SlideShare a Scribd company logo
1 of 196
Download to read offline
Luc_Faucheux_2021
Stochastic Calculus โ€“ ITO โ€“ III
From SDEs to PDEs and back โ€“ part trois
Non-Linear SDE
1
Luc_Faucheux_2021
In this section
ยจ We finally look at the non-linear SDE
ยจ We will introduce the [๐›ผ] calculus in order to not get confused with the factor (1/2)
ยจ We will look also at the backward and forward formalism for PDE
ยจ Maybe if we have time some examples (OU canonical process)
ยจ SDE: Stochastic Differential Equations
ยจ SIE: Stochastic Integral Equations
ยจ PDE: Partial Differential Equations
ยจ PDF: Probability Distribution Function
2
Luc_Faucheux_2021
In this deck - II
ยจ More importantly we will discuss at length (and maybe ad nauseam for some of you) the so
called โ€œITO-STRATANOVITCH controversyโ€
ยจ That was something that for some reason was quite popular in the 90s
ยจ Like a lot of things that were popular in the 90s, there was maybe not a good rationale for it,
and that kind of faded away
ยจ So we will use it more as an example on how to use, and make sure that we understand
stochastic calculus
3
Luc_Faucheux_2021
In this deck - III
ยจ If you Google โ€œITO-STRATANOVITCH controversyโ€, you can see that it is still creating a lot of
confusion out there, especially in the fields of Physics for non-homogeneous diffusion
coefficients (in the field of finance that would be for a volatility that is a function of the
stochastic underlier like equity for example)
ยจ So that would be for modeling an SDE that would look something like that:
ยจ dX(t)= a(X(t),t).dt+b(X(t),t).dW(t)
ยจ We will show, that unlike all the previous examples in the deck II, writing something like the
above is actually not well defined.
ยจ That is where the confusion comes from
ยจ Interestingly enough it also comes from the fact that people started using digital computers
to simulate diffusion process, and in a nutshell (running the risk of oversimplifying):
ยจ DIGITAL COMPUTERS LIVE IN AN ITO WORLD
ยจ ANALOG COMPUTERS LIVE IN A STRATANOVITCH WORLD
4
Luc_Faucheux_2021
In this deck - IV
ยจ To quote Manella:
ยจ โ€ The Itรด versus Stratonovich controversy, about the "correct" calculus to use for integration
of Langevin equations, was settled to general satisfaction some 30 years ago. Recently,
however, it has started to re-emerge, following the advent of new experimental techniques.โ€
ยจ That was written in 2012 or so.
ยจ But a lot of people are still confused by it.
ยจ I get also confused all the time, and then I have a panic attack, then I take a couple of deep
breathes, and redo a bunch of pages of derivations to make sure that I still understand it.
ยจ The older I get the more frequent that happens.
5
Luc_Faucheux_2021
In this deck - V
6
Luc_Faucheux_2021
dX= a(X(t),t).dt+b(X(t),t).dW
โ€œCent deux fois sur le mรฉtier remettez votre ouvrageโ€
Nicolas Boileau
7
Luc_Faucheux_2021
Finally
ยจ We are almost there, so just before we tackle that one, because the factor (1/2) are
confusing (is that coming from the middle point in STRATO, or is is coming from the second
term in a Taylor expansion?), we generalize a little the formalism
8
Luc_Faucheux_2021
Introducing the [๐›ผ]calculus
9
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus
ยจ Here is where we take a pause to avoid being confused with factors 2
ยจ Ito when defining the Ito integral of a function took the left-side
ยจ Strato took the middle point
ยจ Hence the existence of a factor (1/2) in the conversion between Ito and Strato
ยจ HOWEVER, now that we are dealing with PDE, there is another factor (1/2) that crops up a
lot (from the integration by part of the second moment ๐‘ฅ!)
ยจ Those 2 factors are NOT related (although in way they are, Strato chose middle point
specifically so that the rules of usual calculus would be formally conserved, so essentially he
solved for which point to use to offset the (1/2) term in Ito lemma)
ยจ So they will be related, but in the mean-time we could get confused between the two
ยจ Some authors (Gleeson, Arovas) have pointed out that it is somewhat easier to carry
derivations โ€in the [๐›ผ] calculusโ€, and then setting the value of ๐›ผ, as opposed to duplicating
everything in Ito then Strato and carrying (1/2) factors
10
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus - II
ยจ The ITO integral is defined as:
ยจ โˆซ
"#"$
"#"%
๐‘“ ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘‹(๐‘ก) = lim
&โ†’(
{โˆ‘)#*
)#&
๐‘“(๐‘‹(๐‘ก))). [๐‘‹(๐‘ก)+*) โˆ’ ๐‘‹(๐‘ก))]}
ยจ The Stratonovitch integral is defined as:
ยจ โˆซ
"#"$
"#"%
๐‘“ ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘‹(๐‘ก) = lim
&โ†’(
{โˆ‘)#*
)#&
๐‘“ [๐‘‹(๐‘ก) + ๐‘‹(๐‘ก)+*)]/2). [๐‘‹(๐‘ก)+*) โˆ’ ๐‘‹(๐‘ก))]}
ยจ We can define the [๐›ผ] integral as:
ยจ โˆซ
"#"$
"#"%
๐‘“ ๐‘‹ ๐‘ก . ([๐›ผ]). ๐‘‘๐‘‹(๐‘ก) =
lim
&โ†’(
{โˆ‘)#*
)#&
๐‘“(๐‘‹(๐‘ก)) + ๐›ผ. [๐‘‹(๐‘ก)+*) โˆ’ ๐‘‹(๐‘ก))]). [๐‘‹(๐‘ก)+*) โˆ’ ๐‘‹(๐‘ก))]}
ยจ ITO will be the case ๐›ผ = 0
ยจ STRATO will be the case ๐›ผ = 1/2
11
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus - III
ยจ We had the relation on the integrals:
ยจ โˆซ
"#"$
"#"%
๐‘“ ๐‘Š ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก) = โˆซ
"#"$
"#"%
๐‘“ ๐‘Š ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) +
*
!
โˆซ
"#"$
"#"%
๐‘“โ€ฒ ๐‘Š ๐‘ก . ๐‘‘๐‘ก
ยจ This becomes:
ยจ โˆซ
"#"$
"#"%
๐‘“ ๐‘Š ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š(๐‘ก) = โˆซ
"#"$
"#"%
๐‘“ ๐‘Š ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) + ๐›ผ. โˆซ
"#"$
"#"%
๐‘“โ€ฒ ๐‘Š ๐‘ก . ๐‘‘๐‘ก
ยจ Or:
ยจ โˆซ
"#"$
"#"%
๐‘“ ๐‘Š ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š(๐‘ก) = โˆซ
"#"$
"#"%
๐‘“ ๐‘Š ๐‘ก . ([๐›ผ] = 0). ๐‘‘๐‘Š(๐‘ก) + ๐›ผ. โˆซ
"#"$
"#"%
๐‘“โ€ฒ ๐‘Š ๐‘ก . ๐‘‘๐‘ก
12
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus - IV
ยจ For a more complicated stochastic process
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘Š
ยจ We have:
ยจ โˆซ
"#"$
"#"%
๐‘“ ๐‘‹ ๐‘ก . โˆ˜ . ๐‘‘๐‘Š ๐‘ก = โˆซ
"#"$
"#"%
๐‘“ ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) + โˆซ
"#"$
"#"% *
!
. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
๐‘“ ๐‘‹(๐‘ก . ๐‘‘๐‘ก
ยจ This now becomes:
ยจ โˆซ
"#"$
"#"%
๐‘“ ๐‘‹ ๐‘ก . [๐›ผ] . ๐‘‘๐‘Š ๐‘ก = โˆซ
"#"$
"#"%
๐‘“ ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) + โˆซ
"#"$
"#"%
๐›ผ. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
๐‘“ ๐‘‹(๐‘ก . ๐‘‘๐‘ก
13
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus - V
ยจ For the SDE we had the following mapping between ITO and STRATO
ยจ The ITO SDE:
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š
ยจ Has the same solution (is the same) as the STRATO SDE in STRATO calculus:
ยจ ๐‘‘๐‘‹ ๐‘ก = [๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก โˆ’
*
!
. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
๐‘ ๐‘ก, ๐‘‹ ๐‘ก ]. ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š
ยจ The STRATO SDE
ยจ ๐‘‘๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š
ยจ Has the same solution (is the same) as the ITO SDE in ITO calculus
ยจ ๐‘‘๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก +
*
!
. A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š
14
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus - VI
ยจ This now becomes:
ยจ The ITO SDE:
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š
ยจ Has the same solution (is the same) as the [๐›ผ] SDE in [๐›ผ] calculus:
ยจ ๐‘‘๐‘‹ ๐‘ก = [๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก โˆ’ ๐›ผ. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
๐‘ ๐‘ก, ๐‘‹ ๐‘ก ]. ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š
ยจ The [๐›ผ] SDE
ยจ ๐‘‘๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š
ยจ Has the same solution (is the same) as the ITO SDE in ITO calculus
ยจ ๐‘‘๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + ๐›ผ. A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š
15
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus - VII
ยจ The ITO lemma (chain rule) reads:
ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ
"#"$
"#"% ,.
,-
. ([). ๐‘‘๐‘‹(๐‘ก) +
*
!
โˆซ
"#"$
"#"% ,!/
,0! (๐‘‹ ๐‘ก ). ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
!
๐‘‘๐‘ก
ยจ In the โ€limitโ€ of small time increments, this can be written formally as the Ito lemma:
ยจ ๐›ฟ๐‘“ =
,.
,-
. ๐›ฟ๐‘‹ +
*
!
.
,!/
,0! . ๐‘!๐›ฟ๐‘ก
ยจ The STRATO lemma reads:
ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ
"#"$
"#"% ,.
,-
. (โˆ˜). ๐‘‘๐‘‹(๐‘ก)
ยจ In the โ€limitโ€ of small time increments, this can be written formally as the Strato lemma:
ยจ ๐›ฟ๐‘“ =
,.
,-
. โˆ˜ . ๐›ฟ๐‘‹
16
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus - VIII
ยจ In the [๐›ผ] calculus the [๐›ผ] lemma (chain rule) now reads :
ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ
"#"$
"#"% ,.
,-
. ([๐›ผ]). ๐‘‘๐‘‹(๐‘ก) +
*
!
โˆ’ ๐›ผ . โˆซ
"#"$
"#"% ,!/
,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
!
. ๐‘‘๐‘ก
ยจ In the โ€limitโ€ of small time increments, this can be written formally as the [๐›ผ] lemma:
ยจ ๐›ฟ๐‘“ =
,.
,-
. ๐›ฟ๐‘‹ +
*
!
โˆ’ ๐›ผ .
,!/
,0! . ๐‘!. ๐›ฟ๐‘ก
ยจ NOTE: you can convince yourselves by redoing the derivation we had on pages 55-60
ยจ This actually highlights why STRATO took the middle point ๐›ผ = 1/2 , as this is the point
that cancels out the (1/2) coming from the Taylor expansion of ๐‘“ ๐‘‹ ๐‘ก% from the left point.
ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = lim
&โ†’(
โˆ‘)#*
)#&
{๐‘“(๐‘‹(๐‘ก))) โˆ’ ๐‘“(๐‘‹(๐‘ก)1*))}
ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = lim
&โ†’(
โˆ‘)#*
)#&
{
,.
,-
. ([). ๐›ฟ๐‘‹ +
*
!
.
,!/
,0! . ([). (๐›ฟ๐‘‹)!}
17
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus - IX
ยจ This is kind of nice, because we only have to do one derivation, and we do not get confused
where the factor (1/2) comes from in STRATO, and whether or not it is the same factor (1/2)
that comes from the integration by part of the second moment.
ยจ Factors (1/2) are very confusing, so whenever possible it is better to keep them as variable
18
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus - X
ยจ As an illustrated example of [๐›ผ] calculus, we calculate
ยจ ๐ด[๐›ผ] = ๐”ผ{โˆซ
"#"$
"#"%
๐‘Š(๐‘ก). ([๐›ผ]). ๐‘‘๐‘Š(๐‘ก)} and with ๐‘“ ๐‘Š ๐‘ก = ๐‘Š(๐‘ก)
ยจ ๐ด[๐›ผ] = ๐”ผ{ lim
&โ†’(
โˆ‘)#*
)#&
๐‘“(๐‘Š(๐‘ก)) + ๐›ผ. [๐‘Š(๐‘ก)+*) โˆ’ ๐‘Š(๐‘ก))]). [๐‘Š(๐‘ก)+*) โˆ’ ๐‘Š(๐‘ก))]}
ยจ ๐ด[๐›ผ] = ๐”ผ{ lim
&โ†’(
โˆ‘)#*
)#&
(๐‘Š(๐‘ก)) + ๐›ผ. [๐‘Š(๐‘ก)+*) โˆ’ ๐‘Š(๐‘ก))]). [๐‘Š(๐‘ก)+*) โˆ’ ๐‘Š(๐‘ก))]}
ยจ ๐”ผ{๐‘Š ๐‘ก . ๐‘Š(๐‘ก2)} = min(๐‘ก, ๐‘ก2)
ยจ ๐”ผ ๐‘Š ๐‘ก . ๐‘Š ๐‘ก2 โˆ’ ๐‘Š ๐‘ก22 = min ๐‘ก, ๐‘ก2 โˆ’ min ๐‘ก, ๐‘ก22 = 0 when ๐‘ก < ๐‘ก2 < ๐‘กโ€ฒโ€ฒ
ยจ ๐”ผ ๐‘Š ๐‘ก โˆ’ ๐‘Š ๐‘ก2 !
= ๐‘ก + ๐‘ก2 โˆ’ 2. min ๐‘ก, ๐‘ก2 = |๐‘ก โˆ’ ๐‘กโ€ฒ|
ยจ ๐ด[๐›ผ] = ๐”ผ{ lim
&โ†’(
โˆ‘)#*
)#&
๐›ผ. [๐‘ก)+* โˆ’ ๐‘ก)]} = ๐›ผ. [๐‘ก% โˆ’ ๐‘ก$]
ยจ Remember if you could integrate X? Worth revisiting within the [๐›ผ] calculus
19
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus - XI
ยจ We have the following conventions:
ยจ ๐›ผ = 0 ITO
ยจ โˆซ
"#"$
"#"%
๐‘“ ๐‘‹ ๐‘ก . [๐›ผ] . ๐‘‘๐‘Š ๐‘ก = โˆซ
"#"$
"#"%
๐‘“ ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) + โˆซ
"#"$
"#"%
๐›ผ. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
๐‘“ ๐‘‹(๐‘ก . ๐‘‘๐‘ก
ยจ โˆซ
"#"$
"#"%
๐‘“ ๐‘‹ ๐‘ก . [0] . ๐‘‘๐‘Š ๐‘ก = โˆซ
"#"$
"#"%
๐‘“ ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก)
ยจ ITO integral is a MARTINGALE
ยจ ITO integral exhibits the ISOMETRY property
ยจ ITO LEMMA
ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ
"#"$
"#"% ,.
,-
. ([๐›ผ]). ๐‘‘๐‘‹(๐‘ก) +
*
!
โˆ’ ๐›ผ . โˆซ
"#"$
"#"% ,!/
,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
!
. ๐‘‘๐‘ก
ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ
"#"$
"#"% ,.
,-
. ([). ๐‘‘๐‘‹(๐‘ก) +
*
!
. โˆซ
"#"$
"#"% ,!/
,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
!
. ๐‘‘๐‘ก
20
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus - XII
ยจ ๐›ผ = 1/2 STRATONOVITCH
ยจ โˆซ
"#"$
"#"%
๐‘“ ๐‘‹ ๐‘ก . [๐›ผ] . ๐‘‘๐‘Š ๐‘ก = โˆซ
"#"$
"#"%
๐‘“ ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) + โˆซ
"#"$
"#"%
๐›ผ. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
๐‘“ ๐‘‹(๐‘ก . ๐‘‘๐‘ก
ยจ โˆซ
"#"$
"#"%
๐‘“ ๐‘‹ ๐‘ก . [1/2] . ๐‘‘๐‘Š ๐‘ก = โˆซ
"#"$
"#"%
๐‘“ ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก)
ยจ STRATO integral is NOT a MARTINGALE
ยจ STRATO integral does NOT exhibits the ISOMETRY property
ยจ STRATO LEMMA, usual regular Newtonian Leibniz calculus chain rule is FORMALLY respected
ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ
"#"$
"#"% ,.
,-
. ([๐›ผ]). ๐‘‘๐‘‹(๐‘ก) +
*
!
โˆ’ ๐›ผ . โˆซ
"#"$
"#"% ,!/
,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
!
. ๐‘‘๐‘ก
ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ
"#"$
"#"% ,.
,-
. (โˆ˜). ๐‘‘๐‘‹(๐‘ก)
21
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus - XIII
ยจ ๐›ผ = 1 KLIMONTOVITCH (reverse ITO, or OTI)
ยจ โˆซ
"#"$
"#"%
๐‘“ ๐‘‹ ๐‘ก . [๐›ผ] . ๐‘‘๐‘Š ๐‘ก = โˆซ
"#"$
"#"%
๐‘“ ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) + โˆซ
"#"$
"#"%
๐›ผ. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
๐‘“ ๐‘‹(๐‘ก . ๐‘‘๐‘ก
ยจ โˆซ
"#"$
"#"%
๐‘“ ๐‘‹ ๐‘ก . [1] . ๐‘‘๐‘Š ๐‘ก = โˆซ
"#"$
"#"%
๐‘“ ๐‘‹ ๐‘ก . (]). ๐‘‘๐‘Š(๐‘ก)
ยจ KLIMONTOVITCH integral is NOT a MARTINGALE
ยจ KLIMONTOVITCH integral does NOT exhibits the ISOMETRY property
ยจ KLIMONTOVITCH LEMMA, does NOT formally looks like the regular Leibniz rule
ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ
"#"$
"#"% ,.
,-
. ([๐›ผ]). ๐‘‘๐‘‹(๐‘ก) +
*
!
โˆ’ ๐›ผ . โˆซ
"#"$
"#"% ,!/
,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
!
. ๐‘‘๐‘ก
ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ
"#"$
"#"% ,.
,-
. ] . ๐‘‘๐‘‹ ๐‘ก โˆ’
*
!
. โˆซ
"#"$
"#"% ,!/
,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
!
. ๐‘‘๐‘ก
22
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus โ€“ XIII a
ยจ No idea what the advantages of the KLIMONTOVITCH calculus since, the integral is not a
martingale like ITO, the usual rules of calculus are not respected like STRATO
ยจ But it is in the literature out there so thought I would include it
ยจ Might be useful when you do backward numerical simulations on a digital computer, but not
sure, will look into this further
23
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus - XIV
ยจ ๐ด[๐›ผ] = ๐”ผ{โˆซ
"#"$
"#"%
๐‘Š(๐‘ก). ([๐›ผ]). ๐‘‘๐‘Š(๐‘ก)} and with ๐‘“ ๐‘Š ๐‘ก = ๐‘Š(๐‘ก)
ยจ ๐ด[๐›ผ] = ๐”ผ{ lim
&โ†’(
โˆ‘)#*
)#&
๐›ผ. [๐‘ก)+* โˆ’ ๐‘ก)]} = ๐›ผ. [๐‘ก% โˆ’ ๐‘ก$]
ยจ ITO ๐›ผ = 0
ยจ ๐”ผ{โˆซ
"#"$
"#"%
๐‘Š(๐‘ก). ([). ๐‘‘๐‘Š(๐‘ก)} = 0
ยจ No surprise there as the ITO integral is a MARTINGALE
ยจ STRATO ๐›ผ = 1/2
ยจ ๐”ผ{โˆซ
"#"$
"#"%
๐‘Š(๐‘ก). (โˆ˜). ๐‘‘๐‘Š(๐‘ก)} = (1/2). [๐‘ก% โˆ’ ๐‘ก$]
ยจ KLIMONTOVITCH ๐›ผ = 1
ยจ ๐”ผ{โˆซ
"#"$
"#"%
๐‘Š(๐‘ก). (]). ๐‘‘๐‘Š(๐‘ก)} = [๐‘ก% โˆ’ ๐‘ก$]
24
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus - XV
ยจ Letโ€™s look at the integration before taking the expectation:
ยจ โˆซ
"#"$
"#"%
๐‘Š(๐‘ก). (โˆ˜). ๐‘‘๐‘Š(๐‘ก) =
*
!
(๐‘Š ๐‘ก%
! โˆ’ ๐‘Š ๐‘ก$
!)
ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ
"#"$
"#"% ,.
,-
. ([). ๐‘‘๐‘‹(๐‘ก) +
*
!
. โˆซ
"#"$
"#"% ,!.
,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
!
. ๐‘‘๐‘ก
ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ
"#"$
"#"% ,.
,-
. โˆ˜ . ๐‘‘๐‘‹ ๐‘ก
ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ
"#"$
"#"% ,.
,-
. ] . ๐‘‘๐‘‹ ๐‘ก โˆ’
*
!
. โˆซ
"#"$
"#"% ,!.
,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
!
. ๐‘‘๐‘ก
ยจ โˆซ
"#"$
"#"% ,.
,-
. โˆ˜ . ๐‘‘๐‘‹ ๐‘ก = โˆซ
"#"$
"#"% ,.
,-
. ([). ๐‘‘๐‘‹(๐‘ก) +
*
!
. โˆซ
"#"$
"#"% ,!.
,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
!
. ๐‘‘๐‘ก
ยจ โˆซ
"#"$
"#"% ,.
,-
. ] . ๐‘‘๐‘‹ ๐‘ก = โˆซ
"#"$
"#"% ,.
,-
. ([). ๐‘‘๐‘‹(๐‘ก) + โˆซ
"#"$
"#"% ,!.
,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
!
. ๐‘‘๐‘ก
25
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus - XVI
ยจ โˆซ
"#"$
"#"% ,.
,-
. โˆ˜ . ๐‘‘๐‘‹ ๐‘ก = โˆซ
"#"$
"#"% ,.
,-
. ([). ๐‘‘๐‘‹(๐‘ก) +
*
!
. โˆซ
"#"$
"#"% ,!.
,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
!
. ๐‘‘๐‘ก
ยจ โˆซ
"#"$
"#"% ,.
,-
. ] . ๐‘‘๐‘‹ ๐‘ก = โˆซ
"#"$
"#"% ,.
,-
. ([). ๐‘‘๐‘‹(๐‘ก) + โˆซ
"#"$
"#"% ,!.
,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
!
. ๐‘‘๐‘ก
ยจ โˆซ
"#"$
"#"% ,.
,3
. โˆ˜ . ๐‘‘๐‘Š ๐‘ก = โˆซ
"#"$
"#"% ,.
,3
. ([). ๐‘‘๐‘Š(๐‘ก) +
*
!
. โˆซ
"#"$
"#"% ,!.
,3! ๐‘ค ๐‘ก . ๐‘ ๐‘ก, ๐‘Š ๐‘ก
!
. ๐‘‘๐‘ก
ยจ โˆซ
"#"$
"#"% ,.
,3
. ] . ๐‘‘๐‘Š ๐‘ก = โˆซ
"#"$
"#"% ,.
,3
. ([). ๐‘‘๐‘Š(๐‘ก) + โˆซ
"#"$
"#"% ,!.
,3! ๐‘ค ๐‘ก . ๐‘ ๐‘ก, ๐‘Š ๐‘ก
!
. ๐‘‘๐‘ก
ยจ โˆซ
"#"$
"#"%
๐‘Š(๐‘ก). (โˆ˜). ๐‘‘๐‘Š(๐‘ก) =
*
!
(๐‘Š ๐‘ก%
! โˆ’ ๐‘Š ๐‘ก$
!)
ยจ With ๐‘“ ๐‘Š ๐‘ก = ๐‘Š(๐‘ก) we have
,.
,4
๐‘Š ๐‘ก = ๐‘Š(๐‘ก)
ยจ
,.
,3
๐‘ค = ๐‘ค
ยจ
,!.
,3! = 1
26
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus - XVII
ยจ โˆซ
"#"$
"#"%
๐‘Š(๐‘ก). (โˆ˜). ๐‘‘๐‘Š(๐‘ก) =
*
!
(๐‘Š ๐‘ก%
! โˆ’ ๐‘Š ๐‘ก$
!)
ยจ โˆซ
"#"$
"#"% ,.
,-
. โˆ˜ . ๐‘‘๐‘‹ ๐‘ก = โˆซ
"#"$
"#"% ,.
,-
. ([). ๐‘‘๐‘‹(๐‘ก) +
*
!
. โˆซ
"#"$
"#"% ,!.
,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
!
. ๐‘‘๐‘ก
ยจ โˆซ
"#"$
"#"% ,.
,3
. โˆ˜ . ๐‘‘๐‘Š ๐‘ก = โˆซ
"#"$
"#"% ,.
,3
. ([). ๐‘‘๐‘Š(๐‘ก) +
*
!
. โˆซ
"#"$
"#"% ,!.
,3! ๐‘ค ๐‘ก . ๐‘ ๐‘ก, ๐‘Š ๐‘ก
!
. ๐‘‘๐‘ก
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘‘๐‘Š(๐‘ก)
ยจ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก = 0
ยจ ๐‘ ๐‘ก, ๐‘‹ ๐‘ก = 0
ยจ
,.
,3
๐‘ค = ๐‘ค and
,!.
,3! = 1
27
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus - XVIII
ยจ โˆซ
"#"$
"#"% ,.
,3
. โˆ˜ . ๐‘‘๐‘Š ๐‘ก = โˆซ
"#"$
"#"% ,.
,3
. ([). ๐‘‘๐‘Š(๐‘ก) +
*
!
. โˆซ
"#"$
"#"% ,!.
,3! ๐‘ค ๐‘ก . ๐‘ ๐‘ก, ๐‘Š ๐‘ก
!
. ๐‘‘๐‘ก
ยจ โˆซ
"#"$
"#"% ,.
,3
. โˆ˜ . ๐‘‘๐‘Š ๐‘ก = โˆซ
"#"$
"#"% ,.
,3
. ([). ๐‘‘๐‘Š(๐‘ก) +
*
!
. โˆซ
"#"$
"#"% ,!.
,3! ๐‘ค ๐‘ก . ๐‘‘๐‘ก
ยจ โˆซ
"#"$
"#"% ,.
,3
. ([). ๐‘‘๐‘Š(๐‘ก) = โˆซ
"#"$
"#"% ,.
,3
. โˆ˜ . ๐‘‘๐‘Š ๐‘ก โˆ’
*
!
. โˆซ
"#"$
"#"% ,!.
,3! ๐‘ค ๐‘ก . ๐‘‘๐‘ก
ยจ
,.
,3
๐‘ค = ๐‘ค and
,!.
,3! = 1
ยจ โˆซ
"#"$
"#"%
๐‘Š(๐‘ก). (โˆ˜). ๐‘‘๐‘Š(๐‘ก) =
*
!
(๐‘Š ๐‘ก%
! โˆ’ ๐‘Š ๐‘ก$
!)
ยจ โˆซ
"#"$
"#"%
๐‘Š(๐‘ก). ([). ๐‘‘๐‘Š(๐‘ก) = โˆซ
"#"$
"#"%
๐‘Š(๐‘ก). โˆ˜ . ๐‘‘๐‘Š ๐‘ก โˆ’
*
!
. โˆซ
"#"$
"#"%
1. ๐‘‘๐‘ก
ยจ โˆซ
"#"$
"#"%
๐‘Š(๐‘ก). ([). ๐‘‘๐‘Š(๐‘ก) =
*
!
(๐‘Š ๐‘ก%
! โˆ’ ๐‘Š ๐‘ก$
!) โˆ’
*
!
. (๐‘ก% โˆ’ ๐‘ก$)
28
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus - XIX
ยจ โˆซ
"#"$
"#"%
๐‘Š(๐‘ก). (โˆ˜). ๐‘‘๐‘Š(๐‘ก) =
*
!
(๐‘Š ๐‘ก%
! โˆ’ ๐‘Š ๐‘ก$
!)
ยจ โˆซ
"#"$
"#"%
๐‘Š(๐‘ก). ([). ๐‘‘๐‘Š(๐‘ก) =
*
!
(๐‘Š ๐‘ก%
! โˆ’ ๐‘Š ๐‘ก$
!) โˆ’
*
!
. (๐‘ก% โˆ’ ๐‘ก$)
ยจ โˆซ
"#"$
"#"% ,.
,3
. ] . ๐‘‘๐‘Š ๐‘ก = โˆซ
"#"$
"#"% ,.
,3
. ([). ๐‘‘๐‘Š(๐‘ก) + โˆซ
"#"$
"#"% ,!.
,3! ๐‘ค ๐‘ก . ๐‘ ๐‘ก, ๐‘Š ๐‘ก
!
. ๐‘‘๐‘ก
ยจ โˆซ
"#"$
"#"% ,.
,3
. ] . ๐‘‘๐‘Š ๐‘ก = โˆซ
"#"$
"#"% ,.
,3
. (โˆ˜). ๐‘‘๐‘Š(๐‘ก) +
*
!
. โˆซ
"#"$
"#"% ,!.
,3! ๐‘ค ๐‘ก . ๐‘ ๐‘ก, ๐‘Š ๐‘ก
!
. ๐‘‘๐‘ก
ยจ โˆซ
"#"$
"#"%
๐‘Š(๐‘ก). ] . ๐‘‘๐‘Š ๐‘ก =
*
!
๐‘Š ๐‘ก%
! โˆ’ ๐‘Š ๐‘ก$
! +
*
!
. (๐‘ก% โˆ’ ๐‘ก$)
29
Luc_Faucheux_2021
Introducing the [๐›ผ] calculus - XX
ยจ Making it a little simpler and easier on the eye with:
ยจ ๐‘ก$ = 0 and ๐‘ก% = ๐‘ก
ยจ ๐‘Š ๐‘ก$ = 0 and ๐‘Š ๐‘ก% = ๐‘Š ๐‘ก = ๐‘Š
ยจ ITO: โˆซ
5
"
๐‘Š. ([). ๐‘‘๐‘Š =
4 " !
!
โˆ’
"
!
๐”ผ โˆซ
5
"
๐‘Š. ([). ๐‘‘๐‘Š = 0
ยจ STRATONOVITCH: โˆซ
5
"
๐‘Š. (โˆ˜). ๐‘‘๐‘Š =
4 " !
!
๐”ผ โˆซ
5
"
๐‘Š. ([). ๐‘‘๐‘Š =
"
!
ยจ KLIMONTOVITCH: โˆซ
5
"
๐‘Š. (]). ๐‘‘๐‘Š =
4 " !
!
+
"
!
๐”ผ โˆซ
5
"
๐‘Š. ([). ๐‘‘๐‘Š = ๐‘ก
ยจ Remember of course that ๐”ผ ๐‘Š = 0 and ๐”ผ ๐‘Š! = ๐‘ก
30
Luc_Faucheux_2021
A very quick summary
31
Luc_Faucheux_2021
A very quick summary
ยจ In the deck II on stochastic calculus, we reviewed the following equations in increasing order
of complexity:
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž. ๐‘‘๐‘ก
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘. ๐‘‘๐‘Š
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž. ๐‘‘๐‘ก + ๐‘. ๐‘‘๐‘Š
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก . ๐‘‘๐‘ก + ๐‘. ๐‘‘๐‘Š
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘ ๐‘ก . ๐‘‘๐‘Š
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก . ๐‘‘๐‘Š
ยจ In all cases we were justified to write โ€œ.โ€, and this was the โ€usualโ€ product that we are used
to. There was no confusion. This is because the drift and diffusion coefficients were NOT
functions of the stochastic variables
32
Luc_Faucheux_2021
A very quick summary - II
ยจ HOWEVER, if there is one thing that you should remember from this deck, is that you
CANNOT write something like:
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘Š(๐‘ก)
ยจ The โ€œ.โ€ is meaningless if you want in this formulation
ยจ As soon as the diffusion coefficient becomes inhomogeneous (depends on the position if
you thinking about a physical diffusion process), or the volatility depends on the level of the
underlier in Finance, you need to pick a convention (Ito, Strato, Klimontovitch, or anything
else) to explain what you mean by โ€œ.โ€ in the term ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘Š(๐‘ก)
ยจ Stochastic calculus is a very strange world as we saw in deck I, and just because you are
writing something that looks familiar does not mean that you are allowed to do it and use it.
ยจ You should just go the extra mile and really write:
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ). ๐‘‘๐‘Š
ยจ Just to know that you cannot use โ€œ.โ€ in the diffusive term
33
Luc_Faucheux_2021
A very quick summary - III
ยจ Note that the drift term is ok
ยจ You do not need to specify:
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ). ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ). ๐‘‘๐‘Š(๐‘ก)
ยจ This has to do with the definition of the integral
ยจ Remember you cannot differentiate in stochastic calculus (or at least if you do you should be
very careful), always better to integrate and write SIE instead of SDE
ยจ For all the previous examples, and even for a non-homogeneous drift term, the point you
pick inside the bin when you define the integral as the limit of a sum, does not matter, tey all
converge to the same usual Riemann integral.
ยจ HOWEVER for non-homogeneous diffusion, the point you pick does matter
34
Luc_Faucheux_2021
A very quick summary - IV
ยจ You pick the left point -> ITO
ยจ which in the [๐›ผ] calculus means ๐›ผ = 0
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ). ๐‘‘๐‘Š(๐‘ก)
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ = 0). ๐‘‘๐‘Š(๐‘ก)
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก)
35
Luc_Faucheux_2021
A very quick summary - V
ยจ You pick the middle point -> STRATANOVITCH
ยจ which in the [๐›ผ] calculus means ๐›ผ = 1/2
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ). ๐‘‘๐‘Š(๐‘ก)
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ = 1/2). ๐‘‘๐‘Š(๐‘ก)
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก)
36
Luc_Faucheux_2021
A very quick summary - VI
ยจ You pick the right point -> KLIMONTOVITCH
ยจ which in the [๐›ผ] calculus means ๐›ผ = 1
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ). ๐‘‘๐‘Š(๐‘ก)
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ = 1). ๐‘‘๐‘Š(๐‘ก)
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (]). ๐‘‘๐‘Š(๐‘ก)
37
Luc_Faucheux_2021
A very quick summary - VII
ยจ ALWAYS go back to the SIE:
ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ
"#"$
"#"% ,.
,-
. ([๐›ผ]). ๐‘‘๐‘‹(๐‘ก) +
*
!
โˆ’ ๐›ผ . โˆซ
"#"$
"#"% ,!.
,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
!
. ๐‘‘๐‘ก
ยจ ONLY in the STRATNOVITCH case of ๐›ผ = 1/2 does the above reduces to:
ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ
"#"$
"#"% ,.
,-
. ([๐›ผ]). ๐‘‘๐‘‹(๐‘ก)
ยจ Which FORMALLY looks like the usual rules of calculus
ยจ But think of this more as a coincidence than something that you can use willy nilly with
utmost confidence.
ยจ A little like Malliavin derivation operator has been designed to formally follow the usual
chain rule, it is more of a coincidence than a given, so always be super careful
38
Luc_Faucheux_2021
A very quick summary - VIII
ยจ Also another quick note
ยจ For sake of simplicity a lot of books or also in this deck we write:
,.
,-
ยจ What we mean really is:
ยจ ๐‘“ is a regular function of say variable ๐‘ฅ
ยจ Partial derivatives
,.
,0
and
,!.
,0! are well defined quantities
ยจ We can evaluate those quantities for ๐‘ฅ = ๐‘‹(๐‘ก)
ยจ We would then write something like
,.
,0
|0#-(") and
,!.
,0! |0#-(")
ยจ But sometimes because we are lazy or want to keep the notations simple we just write:
ยจ
,.
,-
and
,!.
,-!
39
Luc_Faucheux_2021
A very quick summary - IX
ยจ
,.
,-
and
,!.
,-! are really not well defined quantity
ยจ Remember that we said right at the beginning that one of the defining charaterictic of a
stochastic process is that it was NOT differentiable
ยจ So really what a lot of textbooks (like Hull for example) mean (or in this deck also), is that we
are trying to not make the equations too cumbersome, so really when we write:
ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ
"#"$
"#"% ,.
,0
. ([๐›ผ]). ๐‘‘๐‘‹(๐‘ก) +
*
!
โˆ’ ๐›ผ . โˆซ
"#"$
"#"% ,!.
,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
!
. ๐‘‘๐‘ก
ยจ or
ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ
"#"$
"#"% ,.
,-
. ([๐›ผ]). ๐‘‘๐‘‹(๐‘ก) +
*
!
โˆ’ ๐›ผ . โˆซ
"#"$
"#"% ,!.
,-! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
!
. ๐‘‘๐‘ก
40
Luc_Faucheux_2021
A very quick summary - X
ยจ We really should be writing:
ยจ ๐‘“ ๐‘‹ ๐‘ก" โˆ’ ๐‘“ ๐‘‹ ๐‘ก# = โˆซ
$%$#
$%$" &'())
&)
|)%+($). ([๐›ผ]). ๐‘‘๐‘‹(๐‘ก) +
,
-
โˆ’ ๐›ผ . โˆซ
$%$#
$%$" &!'())
&)! |)%+($). ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
-
. ๐‘‘๐‘ก
ยจ That is getting a little cumbersome
ยจ Baxter is one of the few textbooks that actually bothers to be fully rigorous in his notations.
ยจ So many apologies in those deck if depending on how much space I have on the line, but sometimes
I will use one of those notations or another, but what I mean really is, going from the most exact and
rigorous to the simplest in notation but least rigorous
ยจ
,.(0)
,0
|0#-(") =
,.(0#-("))
,0
=
,.(-("))
,0
=
,.(-("))
,-
=
,.
,-
ยจ Same for the other higher order of derivatives
41
Luc_Faucheux_2021
The McKean and Goodman derivation of the
link between SDE and PDE for the PDF, a
first taste of the Feynmann-Kac theorem
42
Luc_Faucheux_2021
Non-linear SDE โ€“ backward equation
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š
ยจ We are in the general case now ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก and ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
ยจ ๐‘‹ ๐‘ก% โˆ’ ๐‘‹ ๐‘ก$ = โˆซ
"#"$
"#"%
๐‘‘๐‘‹ ๐‘ก = โˆซ
"#"$
"#"%
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก) + โˆซ
"#"$
"#"%
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก)
ยจ We follow here the derivation from McKean and Goodman, but might actually also re-derive
it from first principle after just to convince ourselves
ยจ We are in ITO calculus, and will use the fact that the ITO integral of a trading strategy is a
martingale.
ยจ For any function ๐ถ(๐‘‹ ๐‘ก , ๐‘ก) (hint: we are using ๐ถ notation like in a call option, so people who
have derived before the Black-Sholes equation should be in familiar territory).
ยจ The Ito lemma reads:
43
Luc_Faucheux_2021
Non-linear SDE โ€“ backward equation - II
ยจ ๐ถ ๐‘‹ ๐‘ก% , ๐‘ก% โˆ’ ๐ถ ๐‘‹ ๐‘ก$ , ๐‘ก$ = โˆซ
"#"$
"#"% ,8
,-
. ([). ๐‘‘๐‘‹(๐‘ก) +
*
!
โˆซ
"#"$
"#"% ,!8
,-! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
!
. ๐‘‘๐‘ก +
โˆซ
"#"$
"#"% ,8
,"
(๐‘‹ ๐‘ก ). ๐‘‘๐‘ก
ยจ In the โ€limitโ€ of small time increments, this can be written formally as the Ito lemma:
ยจ ๐›ฟ๐ถ =
,8
,-
. ๐›ฟ๐‘‹ +
*
!
.
,!8
,-! . ๐›ฟ๐‘‹! +
,8
,"
. ๐›ฟ๐‘ก
ยจ Note here the factor (1/2) coming from the Taylor expansion
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š
ยจ ๐‘‘๐‘‹ ๐‘ก ! = ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
!
. ๐‘‘๐‘ก
ยจ ๐›ฟ๐ถ =
,8
,-
. ๐›ฟ๐‘‹ +
*
!
.
,!8
,-! . ๐‘!๐›ฟ๐‘ก +
,8
,"
. ๐›ฟ๐‘ก
44
Luc_Faucheux_2021
Non-linear SDE โ€“ backward equation - III
ยจ ๐›ฟ๐ถ =
,8
,-
. ๐›ฟ๐‘‹ +
*
!
.
,!8
,-! . ๐‘!๐›ฟ๐‘ก +
,8
,"
. ๐›ฟ๐‘ก
ยจ ๐›ฟ๐ถ =
,8
,"
+
*
!
.
,!8
,-! . ๐‘! +
,8
,-
. ๐‘Ž . ๐›ฟ๐‘ก +
,8
,-
. ๐‘Ž. ๐›ฟ๐‘Š
ยจ ๐ถ ๐‘‹ ๐‘ก% , ๐‘ก% โˆ’ ๐ถ ๐‘‹ ๐‘ก$ , ๐‘ก$ = โˆซ
"#"$
"#"% ,8
,-
. ๐‘ ๐‘‹ ๐‘ก , ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก)
+ โˆซ
"#"$
"#"% ,8(- " ,")
,"
+
*
!
.
,!8(- " ,")
,-! . ๐‘(๐‘‹ ๐‘ก , ๐‘ก)! +
,8(- " ,")
,-
. ๐‘Ž(๐‘‹ ๐‘ก , ๐‘ก) . ๐‘‘๐‘ก
ยจ It would be nice if the term of first order in time would disappear.
ยจ This is what the Feynman-Kac theorem is about
ยจ IF the function ๐ถ(๐‘‹ ๐‘ก , ๐‘ก) is such that:
ยจ
,8(- " ,")
,"
+
*
!
. ๐‘(๐‘‹ ๐‘ก , ๐‘ก)!.
,!8(- " ,")
,-! + ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก .
,8(- " ,")
,-
= 0
45
Luc_Faucheux_2021
Non-linear SDE โ€“ backward equation - IV
ยจ Then:
ยจ ๐ถ ๐‘‹ ๐‘ก% , ๐‘ก% โˆ’ ๐ถ ๐‘‹ ๐‘ก$ , ๐‘ก$ = โˆซ
"#"$
"#"% ,8
,-
. ๐‘ ๐‘‹ ๐‘ก , ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก)
ยจ Fixing ๐‘ก% = ๐‘‡ maturity of the option (boundary condition)
ยจ And the conditional expectation:
ยจ ๐”ผ{๐ถ(๐‘‹ ๐‘ก , ๐‘ก)|๐‘‹ ๐‘ก$ = ๐‘‹$}
ยจ ๐”ผ ๐ถ ๐‘‹ ๐‘ก% , ๐‘ก% ๐‘‹ ๐‘ก$ = ๐‘‹$ = ๐”ผ ๐ถ ๐‘‹ ๐‘ก$ , ๐‘ก$ ๐‘‹ ๐‘ก$ = ๐‘‹$ +
๐”ผ โˆซ
"#"$
"#"% ,8
,-
. ๐‘ ๐‘‹ ๐‘ก , ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ๐‘‹ ๐‘ก$ = ๐‘‹$
ยจ ๐”ผ ๐ถ ๐‘‹ ๐‘ก$ , ๐‘ก$ ๐‘‹ ๐‘ก$ = ๐‘‹$ = ๐ถ ๐‘‹ ๐‘ก$ , ๐‘ก$
ยจ ๐”ผ โˆซ
"#"$
"#"% ,8
,-
. ๐‘ ๐‘‹ ๐‘ก , ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ๐‘‹ ๐‘ก$ = ๐‘‹$ = 0
ยจ because this is an ITO integral (note that this would NOT be the case in STRATO, so here it
actually pays to work in ITO)
46
Luc_Faucheux_2021
Non-linear SDE โ€“ backward equation - V
ยจ ๐ถ ๐‘‹ ๐‘ก$ , ๐‘ก$ = ๐”ผ ๐ถ ๐‘‹ ๐‘ก% , ๐‘ก% ๐‘‹ ๐‘ก$ = ๐‘‹$
ยจ This is essentially the Feynman-Kac formula
ยจ IF ๐ถ(๐‘ฅ, ๐‘ก) follows the BACKWARD Kolmogorov (FP) equation:
ยจ
,8(0,")
,"
+
*
!
. ๐‘(๐‘ฅ, ๐‘ก)!.
,!8(0,")
,0! + ๐‘Ž ๐‘ฅ, ๐‘ก .
,8(0,")
,0
= 0
ยจ Then ๐ถ(๐‘ฅ, ๐‘ก) can be written as a conditional expectation:
ยจ ๐ถ ๐‘ฅ, ๐‘ก = ๐”ผ ๐ถ ๐‘‹ ๐‘ก% , ๐‘ก% ๐‘‹ ๐‘ก = ๐‘ฅ
ยจ Under the probability measure such that ๐‘‹ ๐‘ก is an ITO process with the ITO SDE:
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š
ยจ ๐‘‘๐‘Š is a Wiener process (Brownian motion) under that probability measure, and the initial
condition is ๐‘‹ ๐‘ก = ๐‘ฅ
47
Luc_Faucheux_2021
Non-linear SDE โ€“ backward equation - VI
ยจ Note that it is called BACKWARD because it concerns the expectation of something that is
fixed in the future at time ๐‘ก% = ๐‘‡ (Boundary condition)
ยจ Just to be in familiar territory, if the boundary condition is a call payoff
ยจ ๐ถ ๐‘‹ ๐‘ก% , ๐‘ก% = ๐ถ ๐‘‹ ๐‘‡ , ๐‘‡ = ๐‘€๐ด๐‘‹(๐‘‹ ๐‘‡ โˆ’ ๐พ, 0)
ยจ ๐ถ ๐‘ฅ, ๐‘ก = ๐”ผ ๐‘€๐ด๐‘‹(๐‘‹ ๐‘‡ โˆ’ ๐พ, 0) ๐‘‹ ๐‘ก = ๐‘ฅ
ยจ
,8(0,")
,"
+
*
!
. ๐‘(๐‘ฅ, ๐‘ก)!.
,!8(0,")
,0! + ๐‘Ž ๐‘ฅ, ๐‘ก .
,8(0,")
,0
= 0
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š
ยจ From the deck on numeraire, using the Money Market account:
ยจ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก = ๐‘Ÿ๐‘†
ยจ ๐‘ ๐‘ก, ๐‘‹ ๐‘ก = ๐œŽ๐‘†
ยจ ๐‘‘๐‘† ๐‘ก = ๐‘Ÿ๐‘†. ๐‘‘๐‘ก + ๐œŽ๐‘†. ([). ๐‘‘๐‘Š
48
Luc_Faucheux_2021
Non-linear SDE โ€“ backward equation - VII
ยจ
,8(0,")
,"
+
*
!
. ๐‘(๐‘ฅ, ๐‘ก)!.
,!8(0,")
,0! + ๐‘Ž ๐‘ฅ, ๐‘ก .
,8(0,")
,0
= 0
ยจ
,8(:,")
,"
+
*
!
. (๐œŽ๐‘†)!.
,!8(:,")
,:! + ๐‘Ÿ๐‘†.
,8(:,")
,:
= 0
ยจ This is exactly the Black-Sholes equation for ๐ถ ๐‘†, ๐‘ก = ๐ถ5(๐‘†, ๐‘ก)/๐‘’1;(<1")
ยจ Backward equations are for conditional expectations (we know the final value, we want to
evaluate the value of a derivative or a claim now)
ยจ Forward equations like the one we have been working on for the PDF tend to be for
probability distributions (we know the starting point, usually a Dirac peak, and we look at
the evolution in time)
ยจ Note that in the BACKWARD equation, the terms ๐‘(๐‘ฅ, ๐‘ก) and ๐‘Ž(๐‘ฅ, ๐‘ก) are OUTSIDE the partial
derivatives (because that came from using Ito lemma)
ยจ Note that in the FORWARD equation, the terms will be INSIDE
ยจ Note that the PDF will follow BOTH the backward and forward (PDF is expectation of a Delta
peak payoff, following Dupire, we will do it again)
49
Luc_Faucheux_2021
Non-linear SDE โ€“ forward equation -
ยจ OK, so we still do not know what equation does the PDF follow.
ยจ We almost there
ยจ ๐ถ ๐‘‹ ๐‘ก% , ๐‘ก% โˆ’ ๐ถ ๐‘‹ ๐‘ก$ , ๐‘ก$ = โˆซ
"#"$
"#"% ,8
,-
. ๐‘ ๐‘‹ ๐‘ก , ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก)
+ โˆซ
"#"$
"#"% ,8(- " ,")
,"
+
*
!
.
,!8(- " ,")
,-! . ๐‘(๐‘‹ ๐‘ก , ๐‘ก)! +
,8(- " ,")
,-
. ๐‘Ž(๐‘‹ ๐‘ก , ๐‘ก) . ๐‘‘๐‘ก
ยจ Letโ€™s not assume that the term in time vanishes
ยจ Letโ€™s actually assume that ๐ถ ๐‘‹ ๐‘ก , ๐‘ก can be ANY function, and in particular we can choose
ยจ ๐ถ ๐‘‹ ๐‘ก = โˆž , ๐‘ก = โˆž = 0
ยจ ๐ถ ๐‘‹ ๐‘ก = ๐‘ก$ , ๐‘ก = ๐‘ก$ = 0
ยจ 0 = โˆซ
"#".
"#( ,8
,-
. ๐‘ ๐‘‹ ๐‘ก , ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก)
+ โˆซ
"#".
"#( ,8(- " ,")
,"
+
*
!
.
,!8(- " ,")
,-! . ๐‘(๐‘‹ ๐‘ก , ๐‘ก)! +
,8(- " ,")
,-
. ๐‘Ž(๐‘‹ ๐‘ก , ๐‘ก) . ๐‘‘๐‘ก
50
Luc_Faucheux_2021
Non-linear SDE โ€“ forward equation - II
ยจ Let us look again at the conditional expectation, but this time we will explicitly write it as an
integral over the possible outcomes
ยจ ๐”ผ ๐ต ๐‘‹ ๐‘ก , ๐‘ก ๐‘‹ ๐‘ก$ = ๐‘‹$ = โˆซ
0#1(
0#+(
๐ต ๐‘ฅ, ๐‘ก . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$) . ๐‘‘๐‘ฅ
ยจ We are obviously after what kind of equation could ๐‘ƒ(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$) verify
ยจ Now again, because we are in ITO calculus:
ยจ ๐”ผ โˆซ
"#".
"#< ,8
,-
. ๐‘ ๐‘‹ ๐‘ก , ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ๐‘‹ ๐‘ก$ = ๐‘‹$ = 0
ยจ In particular:
ยจ ๐”ผ โˆซ
"#".
"#+( ,8
,-
. ๐‘ ๐‘‹ ๐‘ก , ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ๐‘‹ ๐‘ก$ = ๐‘‹$ = 0
51
Luc_Faucheux_2021
Non-linear SDE โ€“ forward equation - III
ยจ 0 =
๐”ผ โˆซ
"#".
"#( ,8(- " ,")
,"
+
*
!
.
,!8(- " ,")
,-! . ๐‘(๐‘‹ ๐‘ก , ๐‘ก)! +
,8(- " ,")
,-
. ๐‘Ž(๐‘‹ ๐‘ก , ๐‘ก) . ๐‘‘๐‘ก ๐‘‹ ๐‘ก$ = ๐‘‹$
ยจ 0 = ๐”ผ โˆซ
"#".
"#( ,8
,"
+
*
!
. ๐‘(๐‘‹ ๐‘ก , ๐‘ก)!.
,!8
,-! + ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก .
,8
,-
. ๐‘‘๐‘ก ๐‘‹ ๐‘ก$ = ๐‘‹$
ยจ 0 = โˆซ
"#".
"#(
โˆซ
0#1(
0#+( ,8
,"
+
*
!
. ๐‘(๐‘ฅ, ๐‘ก)!.
,!8
,0! + ๐‘Ž ๐‘ฅ, ๐‘ก .
,8
,0
. ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$) . ๐‘‘๐‘ฅ. ๐‘‘๐‘ก
ยจ Remember that this is true for ANY function ๐ถ(๐‘‹ ๐‘ก , ๐‘ก) that we can choose arbitrarily so
that the boundary values, as well as the derivatives, vanishes
ยจ We now integrate by part the above integral (we are now in the world of regular calculus, so
the usual rules of calculus apply, none of this ITO / STRATO issue)
ยจ 0 = โˆซ
"#".
"#(
โˆซ
0#1(
0#+(
โˆ’๐ถ
,=
,"
+ ๐ถ.
*
!
.
,!
,0! [๐‘(๐‘ฅ, ๐‘ก)!. ๐‘] โˆ’ ๐ถ.
,
,0
[๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘] . ๐‘‘๐‘ฅ. ๐‘‘๐‘ก
52
Luc_Faucheux_2021
Non-linear SDE โ€“ forward equation - IV
ยจ 0 = โˆซ
"#".
"#(
โˆซ
0#1(
0#+(
โˆ’๐ถ
,=
,"
+ ๐ถ.
*
!
.
,!
,0! [๐‘(๐‘ฅ, ๐‘ก)!. ๐‘] โˆ’ ๐ถ.
,
,0
[๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘] . ๐‘‘๐‘ฅ. ๐‘‘๐‘ก
ยจ This is true for any function ๐ถ(๐‘ฅ, ๐‘ก)
ยจ And so:
ยจ 0 = โˆ’
,=
,"
+
*
!
.
,!
,0! [๐‘(๐‘ฅ, ๐‘ก)!. ๐‘] โˆ’
,
,0
[๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘]
ยจ
,=(0,"|-.,".)
,"
= โˆ’
,
,0
๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ +
*
!
.
,!
,0! [๐‘(๐‘ฅ, ๐‘ก)!. ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)]
ยจ
,=(0,"|-.,".)
,"
= โˆ’
,
,0
๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’
,
,0
[
*
!
. [๐‘(๐‘ฅ, ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)]
ยจ This is the FORWARD Kolmogorov (FP) equation, because we know the initial condition and
we are looking at an advection/diffusion in time
ยจ Note that in the FORWARD equation the terms ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก and ๐‘ ๐‘‹ ๐‘ก , ๐‘ก are INSIDE the
partial derivatives (because that came from integrating by parts the ITO lemma)
53
Luc_Faucheux_2021
Non-linear SDE โ€“ forward equation - V
ยจ
,=(0,"|-.,".)
,"
= โˆ’
,
,0
๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’
,
,0
[
*
!
. [๐‘(๐‘ฅ, ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)]
ยจ ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ = ๐‘ ๐‘ฅ, ๐‘ก subject to the initial condition: ๐‘(๐‘ฅ, ๐‘ก$) = ๐›ฟ ๐‘ฅ โˆ’ ๐‘‹$
ยจ
,=(0,")
,"
= โˆ’
,
,0
[๐‘€* ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก โˆ’
,
,0
[๐‘€! ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ]]
ยจ < โˆ†๐‘‹ > = ๐ธ โˆ†๐‘‹ =< ๐‘ฅ >"+โˆ†" โˆ’< ๐‘ฅ >"= ๐น* ๐‘‹ ๐‘ก , ๐‘ก . โˆ†๐‘ก (drift term)
ยจ < โˆ†๐‘‹!> = ๐ธ โˆ†๐‘‹! =< (๐‘ฅโˆ’< ๐‘ฅ >"+โˆ†")!>"+โˆ†"= ๐น! ๐‘‹ ๐‘ก , ๐‘ก . โˆ†๐‘ก (diffusion term)
ยจ We showed that ๐น* ๐‘‹ ๐‘ก , ๐‘ก = ๐‘€* ๐‘‹ ๐‘ก , ๐‘ก and ๐น! ๐‘‹ ๐‘ก , ๐‘ก = 2. ๐‘€! ๐‘‹ ๐‘ก , ๐‘ก
ยจ We now have also shown that:
ยจ ๐‘€* ๐‘ฅ, ๐‘ก = ๐‘Ž ๐‘ฅ, ๐‘ก
ยจ ๐‘€! ๐‘ฅ, ๐‘ก =
*
!
. ๐‘(๐‘ฅ, ๐‘ก)!
ยจ Note that we do not always have an explicit solution for ๐‘ ๐‘ฅ, ๐‘ก at this point.
54
Luc_Faucheux_2021
Non-linear SDE โ€“ Summary
ยจ As we saw with the Dupire approach, what is interesting is that the PDF follows BOTH a
FORWARD and a BACKWARD equation:
ยจ In short in ITO: ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š
ยจ
,=(0,"|-.,".)
,"
= โˆ’
,
,0
๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’
,
,-
[
*
!
. [๐‘(๐‘ฅ, ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)]
ยจ
,=(0,"|-.,".)
,"
= โˆ’
,
,0
๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ ] +
*
!
.
,!
,-! [. [๐‘(๐‘ฅ, ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)]
ยจ This is the FORWARD equation: given the starting point (๐‘‹$, ๐‘ก$), the conditional probability
๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$) evolves forward in time
ยจ
,=(0,"|-.,".)
,".
= โˆ’๐‘Ž ๐‘‹$, ๐‘ก$
,
,-.
๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’
*
!
. ๐‘(๐‘‹$, ๐‘ก$)! ,!
,-.
! ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)
ยจ This is the BACKWARD equation: given the end point (๐‘ฅ, ๐‘ก), the conditional probability can
also be expressed as an expectation (of the Delta peak), and will thus follow a BACKWARD FP
diffusing backward in time
55
Luc_Faucheux_2021
Non-linear SDE โ€“ Summary II
ยจ ๐‘(๐‘‹%, ๐‘ก%|๐‘‹$, ๐‘ก$) = โˆซ
0#1(
0#+(
๐‘(๐‘ฅ, ๐‘ก%|๐‘‹$, ๐‘ก$). ๐›ฟ(๐‘ฅ โˆ’ ๐‘‹%). ๐‘‘๐‘ฅ
ยจ This is fairly obvious, and recast ๐‘ƒ(๐‘‹%, ๐‘ก%|๐‘‹$, ๐‘ก$) as an expectation of a terminal payoff,
which is the Delta peak ๐›ฟ(๐‘ฅ โˆ’ ๐‘‹%), with the probability measure associated to
๐‘(๐‘‹%, ๐‘ก%|๐‘‹$, ๐‘ก$) .
ยจ Just like the call option was the expectation of the terminal payoff ๐‘€๐ด๐‘‹(๐‘†2 โˆ’ ๐พ, 0)
ยจ So the idea is that we can apply ITO lemma to ๐‘(๐‘‹%, ๐‘ก%|๐‘‹$, ๐‘ก$) (this time on the โ€œstartingโ€
variables (๐‘‹$, ๐‘ก$)
ยจ This s exactly what we just did, but because it can be confusing since in some ways, the
conditional probability is โ€œtwo thingsโ€: it is the conditional probability that we use in order
to calculate the expectation (integral) of ANY derivative function, and hence will follow the
FORWARD equation, it is ALSO the expectation of a claim (mainly in an evident manner the
Delta peak), and so will follow the BACKWARD equation
ยจ Note: only in ITO calculus can we zero out the integral as a martingale, otherwise we would
have to carry those terms in the derivation
56
Luc_Faucheux_2021
Why going through this trouble?
The Ito-Stratanovitch controversy
57
Luc_Faucheux_2021
Why did we go through all this trouble?
ยจ Letโ€™s recap:
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š in ITO calculus
ยจ We have shown that in that case the PDF follows a FORWARD ITO Kolmogorov (FP)
ยจ
,=(0,"|-.,".)
,"
= โˆ’
,
,0
๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’
,
,0
[
*
!
. [๐‘(๐‘ฅ, ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)]
ยจ < โˆ†๐‘‹ > = ๐ธ โˆ†๐‘‹ =< ๐‘ฅ >"+โˆ†" โˆ’< ๐‘ฅ >"= ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก . โˆ†๐‘ก (advection term)
ยจ < โˆ†๐‘‹!> = ๐ธ โˆ†๐‘‹! =< (๐‘ฅโˆ’< ๐‘ฅ >"+โˆ†")!>"+โˆ†"= ๐‘(๐‘‹ ๐‘ก , ๐‘ก)!. โˆ†๐‘ก (diffusion term)
58
Luc_Faucheux_2021
Why did we go through all this trouble? - II
ยจ We ALSO know that going between ITO and [๐›ผ]:
ยจ The ITO SDE:
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š
ยจ Has the same solution (is the same) as the [๐›ผ] SDE in [๐›ผ] calculus:
ยจ ๐‘‘๐‘‹ ๐‘ก = [๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก โˆ’ ๐›ผ. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
๐‘ ๐‘ก, ๐‘‹ ๐‘ก ]. ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š
ยจ The [๐›ผ] SDE
ยจ ๐‘‘๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š
ยจ Has the same solution (is the same) as the ITO SDE in ITO calculus
ยจ ๐‘‘๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + ๐›ผ. A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š
59
Luc_Faucheux_2021
Why did we go through all this trouble? - III
ยจ And so, if we start with an [๐›ผ] SDE: ๐‘‘๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š
ยจ It has the same solution (is the same) as the ITO SDE in ITO calculus
ยจ ๐‘‘๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + ๐›ผ. A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š
ยจ Which will then follow the ITO FORWARD Kolmogorov (FP):
ยจ
,@(0,"|-.,".)
,"
= โˆ’
,
,0
๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ƒ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’
,
,0
[
*
!
. [๐‘(๐‘ฅ, ๐‘ก)! . ๐‘ƒ(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)]
ยจ With:
ยจ ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + ๐›ผ. A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก
ยจ ๐‘ ๐‘‹ ๐‘ก , ๐‘ก = A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก
ยจ ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + ๐›ผ. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
๐‘ ๐‘ก, ๐‘‹ ๐‘ก
60
Luc_Faucheux_2021
Why did we go through all this trouble? โ€“ III - a
ยจ < โˆ†๐‘‹ > = ๐ธ โˆ†๐‘‹ =< ๐‘ฅ >"+โˆ†" โˆ’< ๐‘ฅ >"= ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก . โˆ†๐‘ก (advection term)
ยจ < โˆ†๐‘‹!> = ๐ธ โˆ†๐‘‹! =< (๐‘ฅโˆ’< ๐‘ฅ >"+โˆ†")!>"+โˆ†"= ๐‘(๐‘‹ ๐‘ก , ๐‘ก)!. โˆ†๐‘ก (diffusion term)
ยจ With:
ยจ ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + ๐›ผ. A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก
ยจ ๐‘ ๐‘‹ ๐‘ก , ๐‘ก = A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก
ยจ Remember that we started from : ๐‘‘๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š
ยจ So: < โˆ†๐‘‹ > = ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก . โˆ†๐‘ก = {@
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + ๐›ผ. A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก }. โˆ†๐‘ก
ยจ < โˆ†๐‘‹ > = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . โˆ†๐‘ก + ๐›ผ. A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . โˆ†๐‘ก
ยจ < โˆ†๐‘‹ > =< @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . [๐›ผ] . ๐‘‘๐‘Š >
61
Luc_Faucheux_2021
Why did we go through all this trouble? โ€“ III - b
ยจ < โˆ†๐‘‹ > = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . โˆ†๐‘ก + ๐›ผ. A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . โˆ†๐‘ก
ยจ < โˆ†๐‘‹ > = < @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก > + < A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . [๐›ผ] . ๐‘‘๐‘Š >
ยจ And
ยจ < @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก > = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . โˆ†๐‘ก
ยจ So we have:
ยจ < A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . [๐›ผ] . ๐‘‘๐‘Š > = ๐›ผ. A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . โˆ†๐‘ก
ยจ In particular:
ยจ ITO : < A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . [ . ๐‘‘๐‘Š > = 0
ยจ STRATO: < A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . โˆ˜ . ๐‘‘๐‘Š > = [
*
!
]. A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . โˆ†๐‘ก
62
Luc_Faucheux_2021
Why did we go through all this trouble? โ€“ III - c
ยจ ITO : < A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . [ . ๐‘‘๐‘Š > = 0
ยจ One can also go back to the definition of the ITO integral (because remember it is never a
SDE, it is ALWAYS and SIE) , but essentially the convention [ of taking the value โ€œbefore the
jumpโ€, implies that the ITO integral is a martingale of expected value 0
ยจ STRATO: < A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . โˆ˜ . ๐‘‘๐‘Š > = [
*
!
]. A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . โˆ†๐‘ก
ยจ Again we can explicitly derive this from the integral, but the convention โˆ˜ implies taking
the value โ€œin the middle of the jumpโ€, hence the STRATO integral CANNOT be a martingale
and has a non zero expected value.
ยจ We did this derivation when we looking at the correspondence between the ITO and
STRATO.
ยจ It has been a while so might take a few pages here to redo it.
63
Luc_Faucheux_2021
Why did we go through all this trouble? โ€“ III - d
ยจ So essentially here is the deal:
ยจ A PDE is well defined, it is just a Partial Differential Equation where the
usual/Leibniz/regular/Leibniz calculus applies
ยจ A SDE is NOT well defined, unless you specify which ๐›ผ-calculus you use (essentially which
point within the bin you take when defining the integral as a limit of sum)
ยจ Left point of the bin -> ITO , ๐›ผ = 0
ยจ Middle point of the bin -> STRATO, ๐›ผ = 1/2
ยจ Right point of the bin -> KLIMONT, ๐›ผ = 1
ยจ Letโ€™s summarize again before we delve more into the reason for the controversy
64
Luc_Faucheux_2021
A nice little summary
65
Luc_Faucheux_2021
We need a nice summary to avoid any confusion
ยจ ITO SDE is: ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š
ยจ This implies that the PDF follows the FORWARD ITO Kolmogorov PDE
ยจ
,=(0,"|-.,".)
,"
= โˆ’
,
,0
๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’
,
,0
[
*
!
. [๐‘(๐‘ฅ, ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)]
ยจ
,=
,"
= โˆ’
,
,0
๐‘Ž๐‘ โˆ’
,
,0
%!=
!
= โˆ’
,
,0
[๐‘Ž๐‘] +
*
!
,!
,0! [
%!=
!
]
ยจ
,=
,"
= โˆ’
,
,0
๐‘Ž๐‘ โˆ’ ๐‘
,%
,0
. ๐‘ โˆ’
*
!
. ๐‘! .
,
,0
๐‘
ยจ
,=
,"
= โˆ’
,
,-
๐‘Ž๐‘ โˆ’
,A
,0
. ๐‘ โˆ’ ๐ท.
,
,0
๐‘
ยจ
,=
,"
= โˆ’
,
,0
๐‘Ž๐‘ โˆ’
,
,0
(๐ท๐‘) using the convention ๐ท =
*
!
. ๐‘!
66
Luc_Faucheux_2021
We need a nice summary to avoid any confusion - a
ยจ
,=
,"
= โˆ’
,
,0
๐‘Ž๐‘ โˆ’
,
,0
(๐ท๐‘)
ยจ Physicists like a diffusion equation for particles that looks like:
ยจ
,=
,"
= โˆ’
,
,0
๐‘ ๐‘œ๐‘š๐‘’๐‘กโ„Ž๐‘–๐‘›๐‘”
ยจ Because when integrating over the x-axis, and assuming some reasonable (read 0) value for
the distribution function at infinity, the integral of the density is invariant with time
(conservation of matter), or:
ยจ
,
,"
โˆซ
1(
+(
๐‘ ๐‘ฅ, ๐‘ก . ๐‘‘๐‘ฅ = 0
ยจ That is a good thing.
ยจ Physicists also like something called โ€œsteady-state solutionโ€, meaning at equilibrium (think
very long time),
,=
,"
= 0
67
Luc_Faucheux_2021
We need a nice summary to avoid any confusion - b
ยจ
,=
,"
= 0 leads to:
ยจ โˆ’
,
,0
๐‘Ž๐‘ โˆ’
,
,0
(๐ท๐‘)
ยจ For simplicity of argument, in the case of no external forcing (๐‘Ž = 0)
ยจ
,
,0
โˆ’
,
,0
(๐ท๐‘) = 0
ยจ So the solution of that is:
ยจ
,
,0
๐ท๐‘ = ๐‘๐‘ก๐‘’
ยจ Again for things to no blow up at infinity, that means:
,
,0
๐ท๐‘ = ๐‘๐‘ก๐‘’ = 0
ยจ That means that ๐ท๐‘ = ๐‘๐‘ก๐‘’ with the normalization condition of โˆซ
1(
+(
๐‘ ๐‘ฅ, ๐‘ก . ๐‘‘๐‘ฅ = 1 usually
68
Luc_Faucheux_2021
We need a nice summary to avoid any confusion - c
ยจ ๐ท๐‘ = ๐‘๐‘ก๐‘’ with the normalization condition of โˆซ
1(
+(
๐‘ ๐‘ฅ, ๐‘ก . ๐‘‘๐‘ฅ = 1, which allows to
calculate the value of that constant.
ยจ And that is when physicists lost their wits in the 80s/90s, because the diffusion is a dynamic
effect, and should not impact the equilibrium solution.
ยจ The equilibrium solution is given by the laws of thermodynamics, and the equilibrium
density should not depend on something like ๐ท
ยจ Essentially the argument was that, if you give it enough time, the equilibrium distribution
will be uniform, and should not depend on how fast or slow you diffuse in some regions
ยจ That is a first flavor of the controversy that arose when we started to look in earnest in non-
homogeneous diffusion processes
ยจ Remember that it was not that long ago on the scale of the planet, Bachelier / Langevin /
Einstein were just over a century ago now, Ito/Doelin in the 50s, so all in all fairly new
69
Luc_Faucheux_2021
We need a nice summary to avoid any confusion - II
ยจ [๐›ผ] SDE is: ๐‘‘๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š
ยจ This implies that the PDF follows the FORWARD [๐›ผ] Kolmogorov PDE
ยจ
&/(),$|+",$")
&$
= โˆ’
&
&)
{1
๐‘Ž ๐‘ฅ, ๐‘ก + ๐›ผ. 4
๐‘ ๐‘ฅ, ๐‘ก .
&
&)
4
๐‘ ๐‘ฅ, ๐‘ก }. ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹#, ๐‘ก# โˆ’
&
&)
[
,
-
. [4
๐‘(๐‘ฅ, ๐‘ก)-
. ๐‘(๐‘ฅ, ๐‘ก|๐‘‹#, ๐‘ก#)]
ยจ
,=
,"
= โˆ’
,
,0
@
๐‘Ž๐‘ + ๐›ผ. A
๐‘.
,
,0
A
๐‘. ๐‘ โˆ’
,
,0
[
*
!
. [A
๐‘! . ๐‘]
ยจ
,=
,"
= โˆ’
,
,0
@
๐‘Ž๐‘ + ๐›ผ. A
๐‘.
,
,0
A
๐‘. ๐‘ โˆ’
,
,0
B
%!=
!
= โˆ’
,
,0
@
๐‘Ž๐‘ +
,
,0
๐›ผ. A
๐‘.
,
,-
A
๐‘. ๐‘ +
*
!
,!
,0! [
B
%!=
!
]
ยจ
,=
,"
= โˆ’
,
,0
@
๐‘Ž๐‘ + ๐›ผ. A
๐‘.
,B
%
,0
. ๐‘ โˆ’ A
๐‘.
,B
%
,0
. ๐‘ โˆ’
*
!
. A
๐‘!.
,
,0
๐‘
ยจ
,=
,"
= โˆ’
,
,0
๐‘Ž๐‘ + ๐›ผ.
,C
A
,0
. ๐‘ โˆ’
,C
A
,0
. ๐‘ โˆ’ b
๐ท.
,
,0
๐‘
70
Luc_Faucheux_2021
We need a nice summary to avoid any confusion - III
ยจ [๐›ผ = 1/2] STRATO SDE is: ๐‘‘๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š
ยจ This implies that the PDF follows the FORWARD STRATO Kolmogorov PDE
ยจ
&/(),$|+",$")
&$
= โˆ’
&
&)
{1
๐‘Ž ๐‘ฅ, ๐‘ก +
,
-
. 4
๐‘ ๐‘ฅ, ๐‘ก .
&
&)
4
๐‘ ๐‘ฅ, ๐‘ก }. ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹#, ๐‘ก# โˆ’
&
&)
[
,
-
. [4
๐‘(๐‘ฅ, ๐‘ก)-
. ๐‘(๐‘ฅ, ๐‘ก|๐‘‹#, ๐‘ก#)]
ยจ
,=
,"
= โˆ’
,
,0
@
๐‘Ž๐‘ +
*
!
. A
๐‘.
,B
%
,0
. ๐‘ โˆ’
,
,0
*
!
. [A
๐‘! . ๐‘
ยจ
,=
,"
= โˆ’
,
,0
@
๐‘Ž๐‘ +
*
!
. A
๐‘.
,
,0
A
๐‘. ๐‘ โˆ’
,
,0
B
%!=
!
= โˆ’
,
,0
@
๐‘Ž๐‘ +
,
,0
*
!
. A
๐‘.
,
,-
A
๐‘. ๐‘ +
*
!
,!
,0! [
B
%!=
!
]
ยจ
,=
,"
= โˆ’
,
,0
@
๐‘Ž๐‘ +
*
!
. A
๐‘.
,B
%
,0
. ๐‘ โˆ’ A
๐‘.
,B
%
,0
. ๐‘ โˆ’
*
!
. A
๐‘!.
,
,0
๐‘ = โˆ’
,
,0
@
๐‘Ž๐‘ โˆ’
*
!
A
๐‘
,B
%
,0
๐‘ โˆ’
*
!
A
๐‘! ,
,0
๐‘
ยจ
,=
,"
= โˆ’
,
,0
@
๐‘Ž๐‘ โˆ’
*
!
.
,C
A
,0
. ๐‘ โˆ’ b
๐ท.
,
,0
๐‘ = โˆ’
,
,0
@
๐‘Ž๐‘ +
*
!
.
,C
A
,0
. ๐‘ โˆ’
,
,0
(b
๐ท๐‘)
71
Luc_Faucheux_2021
We need a nice summary to avoid any confusion - IIIa
ยจ Note that we can also write as:
ยจ
,=
,"
= โˆ’
,
,0
@
๐‘Ž๐‘ +
*
!
. A
๐‘.
,B
%
,0
. ๐‘ โˆ’
,
,0
*
!
. [A
๐‘! . ๐‘
ยจ
,=
,"
= โˆ’
,
,0
[@
๐‘Ž๐‘] โˆ’
,
,0
[
*
!
. A
๐‘.
,B
%
,0
. ๐‘ โˆ’
*
!
, B
%!=
,0
]
ยจ And since:
*
!
, B
%!=
,0
=
*
!
,B
%.B
%=
,0
=
*
!
A
๐‘
,B
%=
,0
+
*
!
A
๐‘๐‘
,B
%
,0
ยจ
,=
,"
= โˆ’
,
,0
@
๐‘Ž๐‘ +
*
!
,
,0
[A
๐‘
,(B
%=)
,0
]
ยจ
,=(0,"|-.,".)
,"
= โˆ’
,
,0
@
๐‘Ž ๐‘ฅ, ๐‘ก ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ +
*
!
,
,0
[A
๐‘ ๐‘ฅ, ๐‘ก
,
,0
{A
๐‘ ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ }]
72
Luc_Faucheux_2021
We need a nice summary to avoid any confusion - IIIb
ยจ That is right there in essence the crux of the matter:
ยจ
,@
,"
= โˆ’
,
,0
@
๐‘Ž๐‘ โˆ’
*
!
.
,C
A
,0
. ๐‘ โˆ’ b
๐ท.
,
,0
๐‘ = โˆ’
,
,0
@
๐‘Ž๐‘ +
*
!
.
,C
A
,0
. ๐‘ โˆ’
,
,0
(b
๐ท๐‘)
ยจ Because of course:
ยจ
,
,0
b
๐ท๐‘ = b
๐ท.
,
,0
๐‘ + ๐‘.
,
,0
b
๐ท
ยจ So you can see how that, combined with the {
*
!
.
,C
A
,0
. ๐‘} term, can easily get people confused,
because in a way you are โ€œmixingโ€ the drift term and the diffusion term.
ยจ In some ways you should not do that, what is drift is drift, and what is diffusion is diffusion
ยจ It is clear when you write the SDE/SIE, because the drift term is in ๐‘‘๐‘ก, and the diffusive term
is in ๐‘‘๐‘Š, assuming that you have specified in which calculus you operate
ยจ But when writing the PDE, it looks like you can willy nilly mix the two terms.
73
Luc_Faucheux_2021
We need a nice summary to avoid any confusion - IV
ยจ ITO SDE is: ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š
ยจ This implies that the PDF follows the FORWARD ITO Kolmogorov PDE
ยจ
,=(0,"|-.,".)
,"
= โˆ’
,
,0
๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’
,
,0
[
*
!
. [๐‘(๐‘ฅ, ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)]
ยจ The PDF ALSO follows the BACKWARD ITO Kolmogorov PDE:
ยจ
,=(0,"|-.,".)
,".
= โˆ’๐‘Ž ๐‘‹$, ๐‘ก$
,
,-.
๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’
*
!
. ๐‘(๐‘‹$, ๐‘ก$)! ,!
,-.
! ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)
ยจ
,=
,".
= โˆ’๐‘Ž
,=
,-.
โˆ’
*
!
. ๐‘! ,!=
,-.
!
ยจ
,=
,".
= โˆ’๐‘Ž
,=
,-.
โˆ’ ๐ท
,!=
,-.
!
ยจ Where I have explicitly kept the notation ๐‘ก$ and ๐‘‹$ to indicate the fact that this is a
BACKWARD PDE (expectation of a payoff at maturity)
74
Luc_Faucheux_2021
We need a nice summary to avoid any confusion - V
ยจ [๐›ผ] SDE is: ๐‘‘๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š
ยจ This implies that the PDF follows the FORWARD [๐›ผ] Kolmogorov PDE
ยจ
&/(),$|+",$")
&$
= โˆ’
&
&)
{1
๐‘Ž ๐‘ฅ, ๐‘ก + ๐›ผ. 4
๐‘ ๐‘ฅ, ๐‘ก .
&
&)
4
๐‘ ๐‘ฅ, ๐‘ก }. ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹#, ๐‘ก# โˆ’
&
&)
[
,
-
. [4
๐‘(๐‘ฅ, ๐‘ก)-
. ๐‘(๐‘ฅ, ๐‘ก|๐‘‹#, ๐‘ก#)]
ยจ The PDF ALSO follows the BACKWARD ITO Kolmogorov PDE:
ยจ
!"($,&|(!,&!)
!&!
= โˆ’ G
๐‘Ž ๐‘ก*, ๐‘‹ ๐‘ก* + ๐›ผ. B
๐‘ ๐‘ก*, ๐‘‹ ๐‘ก* .
!
!(!
B
๐‘ ๐‘ก*, ๐‘‹ ๐‘ก*
!
!(!
๐‘ ๐‘ฅ, ๐‘ก ๐‘‹*, ๐‘ก* โˆ’
+
,
. ๐‘(๐‘‹*, ๐‘ก*), !"
!(!
" ๐‘(๐‘ฅ, ๐‘ก|๐‘‹*, ๐‘ก*)
ยจ
,=
,".
= โˆ’ @
๐‘Ž + ๐›ผ. A
๐‘
,
,-.
A
๐‘
,
,-.
๐‘ โˆ’
*
!
. ๐‘! ,!
,-.
! ๐‘
ยจ
,=
,".
= โˆ’ @
๐‘Ž + ๐›ผ.
,C
A
,-.
,=
,-.
โˆ’ b
๐ท
,!=
,-.
!
75
Luc_Faucheux_2021
We need a nice summary to avoid any confusion - VI
ยจ [๐›ผ = 1/2] STRATO SDE is: ๐‘‘๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š
ยจ This implies that the PDF follows the FORWARD STRATO Kolmogorov PDE
ยจ
&/(),$|+",$")
&$
= โˆ’
&
&)
{1
๐‘Ž ๐‘ฅ, ๐‘ก +
,
-
. 4
๐‘ ๐‘ฅ, ๐‘ก .
&
&)
4
๐‘ ๐‘ฅ, ๐‘ก }. ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹#, ๐‘ก# โˆ’
&
&)
[
,
-
. [4
๐‘(๐‘ฅ, ๐‘ก)-
. ๐‘(๐‘ฅ, ๐‘ก|๐‘‹#, ๐‘ก#)]
ยจ The PDF ALSO follows the BACKWARD ITO Kolmogorov PDE:
ยจ
!"($,&|(!,&!)
!&!
= โˆ’ G
๐‘Ž ๐‘ก*, ๐‘‹ ๐‘ก* +
+
,
. B
๐‘ ๐‘ก*, ๐‘‹ ๐‘ก* .
!
!(!
B
๐‘ ๐‘ก*, ๐‘‹ ๐‘ก*
!
!(!
๐‘ ๐‘ฅ, ๐‘ก ๐‘‹*, ๐‘ก* โˆ’
+
,
. ๐‘(๐‘‹*, ๐‘ก*), !"
!(!
" ๐‘(๐‘ฅ, ๐‘ก|๐‘‹*, ๐‘ก*)
ยจ
,=
,".
= โˆ’ @
๐‘Ž +
*
!
. A
๐‘
,
,-.
A
๐‘
,
,-.
๐‘ โˆ’
*
!
. A
๐‘! ,!
,-.
! ๐‘
ยจ
,=
,".
= โˆ’ @
๐‘Ž +
*
!
.
,C
A
,-.
,=
,-.
โˆ’ b
๐ท
,!=
,-.
!
76
Luc_Faucheux_2021
We need a nice summary to avoid any confusion โ€“ VI-a
ยจ Note that we can also write as:
ยจ
,=
,".
= โˆ’ @
๐‘Ž +
*
!
.
,C
A
,-.
,=
,-.
โˆ’ b
๐ท
,!=
,-.
!
ยจ
,=
,".
= โˆ’ @
๐‘Ž +
*
!
. A
๐‘
,
,-.
A
๐‘
,
,-.
๐‘ โˆ’
*
!
. A
๐‘! ,!
,-.
! ๐‘
ยจ And since:
,
,-.
A
๐‘.
,=
,-.
=
,B
%
,-.
.
,=
,-.
+ A
๐‘.
,!=
,-.
!
ยจ
,=
,".
= โˆ’@
๐‘Ž.
,=
,-.
โˆ’
*
!
. A
๐‘
,
,-.
[A
๐‘.
,=
,-.
]
ยจ
,=(0,"|-.,".)
,"
= โˆ’@
๐‘Ž ๐‘ก$, ๐‘‹ ๐‘ก$ .
,=(0,"|-.,".)
,-.
โˆ’
*
!
. A
๐‘ ๐‘ก$, ๐‘‹ ๐‘ก$
,
,-.
A
๐‘ ๐‘ก$, ๐‘‹ ๐‘ก$ .
,=(0,"|-.,".)
,-.
77
Luc_Faucheux_2021
We need a nice summary to avoid any confusion โ€“ VII
ยจ ITO FORWARD
ยจ
,=(0,"|-.,".)
,"
= โˆ’
,
,0
๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’
,
,0
[
*
!
. [๐‘(๐‘ฅ, ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)]
ยจ ITO BACKWARD
ยจ
,=(0,"|-.,".)
,".
= โˆ’๐‘Ž ๐‘‹$, ๐‘ก$
,
,-.
๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’
*
!
. ๐‘(๐‘‹$, ๐‘ก$)! ,!
,-.
! ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)
ยจ STRATO FORWARD
ยจ
,=(0,"|-.,".)
,"
= โˆ’
,
,0
@
๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ +
*
!
,
,0
[A
๐‘ ๐‘ฅ, ๐‘ก
,
,0
{A
๐‘ ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ }]
ยจ STRATO BACKWARD
ยจ
,=(0,"|-.,".)
,"
= โˆ’@
๐‘Ž ๐‘ก$, ๐‘‹ ๐‘ก$ .
,=(0,"|-.,".)
,-.
โˆ’
*
!
. A
๐‘ ๐‘ก$, ๐‘‹ ๐‘ก$
,
,-.
A
๐‘ ๐‘ก$, ๐‘‹ ๐‘ก$ .
,=(0,"|-.,".)
,-.
78
Luc_Faucheux_2021
We need a nice summary to avoid any confusion โ€“ VIII
ยจ ITO FORWARD
ยจ
,=(0,"|-.,".)
,"
= โˆ’
,
,0
๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’
,
,0
[๐ท(๐‘ฅ, ๐‘ก). ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)]
ยจ ITO BACKWARD
ยจ
,=(0,"|-.,".)
,".
= โˆ’๐‘Ž ๐‘‹$, ๐‘ก$
,
,-.
๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’ ๐ท ๐‘‹$, ๐‘ก$
,!
,-.
! ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)
ยจ STRATO FORWARD
ยจ
,=(0,"|-.,".)
,"
= โˆ’
,
,0
@
๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ +
,
,0
[ b
๐ท ๐‘ฅ, ๐‘ก
,
,0
{ b
๐ท ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ }]
ยจ STRATO BACKWARD
ยจ
,=(0,"|-.,".)
,"
= โˆ’@
๐‘Ž ๐‘ก$, ๐‘‹ ๐‘ก$ .
,=(0,"|-.,".)
,-.
โˆ’ b
๐ท ๐‘ก$, ๐‘‹ ๐‘ก$
,
,-.
b
๐ท ๐‘ก$, ๐‘‹ ๐‘ก$ .
,=(0,"|-.,".)
,-.
79
Luc_Faucheux_2021
We need a nice summary to avoid any confusion โ€“ IX
ยจ Note that one of the characteristic of the STRATO equations is that they are much more
symmetrical forward and backward (physicist like that because it does not break the time
symmetry as much as ITO)
ยจ In Ito, there is a clear distinction: the diffusion appears INSIDE the derivative for the
FORWARD equation, and OUTSIDE for the BACKWARD
ยจ In STRATO, it is split more equally
ยจ STRATO FORWARD
ยจ
,=
,"
= โˆ’
,
,0
@
๐‘Ž๐‘ +
,
,0
[ b
๐ท
,
,-
{ b
๐ท. ๐‘}]
ยจ STRATO BACKWARD
ยจ
,=
,"
= โˆ’@
๐‘Ž.
,=
,-.
โˆ’ b
๐ท
,
,-.
b
๐ท.
,=
,-.
80
Luc_Faucheux_2021
We need a nice summary to avoid any confusion โ€“ X
ยจ Note that we in doubt it is also safe to recast the STRATO as an ITO using the rules of
transformation:
ยจ If we start from a STRATO SDE: ๐‘‘๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š
ยจ The corresponding ITO SDE is: ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š
ยจ With:
ยจ ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + [
*
!
]. A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก
ยจ ๐‘ ๐‘‹ ๐‘ก , ๐‘ก = A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก
ยจ ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + [
*
!
]. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
๐‘ ๐‘ก, ๐‘‹ ๐‘ก
81
Luc_Faucheux_2021
We need a nice summary to avoid any confusion โ€“ XI
ยจ And conversely:
ยจ If we start from a ITO SDE: ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š
ยจ The corresponding STRATO SDE is: ๐‘‘๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š
ยจ With:
ยจ @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก = ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก โˆ’ [
*
!
]. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
๐‘ ๐‘ก, ๐‘‹ ๐‘ก
ยจ A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก = ๐‘ ๐‘‹ ๐‘ก , ๐‘ก
ยจ @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก = ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก โˆ’ [
*
!
]. A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก
82
Luc_Faucheux_2021
The crux of the matter
83
Luc_Faucheux_2021
The crux of the matter
ยจ The main reason for the confusion is that people write something that looks like an ODE and
assume that because we are using the same usual notations, a lot of the baggage we
accumulated on regular calculus (Taylor expansion, Leibniz rule, chain rule,..) will also carry
over.
ยจ I think an easy rule to remember is that writing an SDE like:
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘Š(๐‘ก)
ยจ Makes NO sense and is not well defined.
84
Luc_Faucheux_2021
The crux of the matter - I
ยจ A slight improvement from writing something wrong like:
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘Š(๐‘ก)
ยจ Would be to write something like:
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก)
ยจ Or:
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก)
ยจ Or:
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (ฮฑ). ๐‘‘๐‘Š(๐‘ก)
ยจ
85
Luc_Faucheux_2021
The crux of the matter - II
ยจ That way it is obvious that we are not writing the same thing:
ยจ ITO
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก)
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐›ผ = 0 . ๐‘‘๐‘Š(๐‘ก)
ยจ STRATONOVITCH
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก)
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐›ผ = 1/2 . ๐‘‘๐‘Š(๐‘ก)
86
Luc_Faucheux_2021
The crux of the matter - III
ยจ Writing it like this makes it obvious that we are NOT writing the same SDE, we are writing
two very different processes
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก)
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก)
ยจ IN FACT, if we want the two different equations to represent the same process, then
obviously the coefficients cannot be the same. It turns out that we are quite lucky in the
sense that there is a simple relations between the two. We are maybe too lucky because
the fact that this relationship is simple sort of pushed people into believing that the two
representations (Ito and Strato) are very close to each other, if not similar. Maybe if would
have been easier if there had been not simple relationship between the coefficients, or a
very complicated one
87
Luc_Faucheux_2021
The crux of the matter โ€“ III a
ยจ Say it another way
ยจ Stochastic (Brownian) calculus is very different from regular (Newtonian/Leibniz) calculus
ยจ ITO calculus is one possible definition of stochastic calculus
ยจ STRATO calculus is one possible definition of stochastic calculus
ยจ ITO calculus is very different from STRATO calculus
ยจ However there is a relationship between the ITO integral and the STRATO integral
ยจ ALWAYS go back to the integral form when dealing with stochastic calculus
ยจ ALWAYS use an SIE if possible as opposed to an SDE
88
Luc_Faucheux_2021
The crux of the matter - IV
ยจ So it is actually easier in order not to get confused to add the โ€œtildeโ€ on top of the
coefficients when dealing with Strato
ยจ STRATO SDE: ๐‘‘๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก)
ยจ ITO SDE: ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก)
ยจ Just to emphasize the point that we are not writing the same equation and that those are
two different processes
89
Luc_Faucheux_2021
The crux of the matter - V
ยจ If we start from a STRATO SDE: ๐‘‘๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก)
ยจ The corresponding ITO SDE is: ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก)
ยจ With:
ยจ ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + [
*
!
]. A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก
ยจ ๐‘ ๐‘ก, ๐‘‹ ๐‘ก = A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก
ยจ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + [
*
!
]. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
๐‘ ๐‘ก, ๐‘‹ ๐‘ก
ยจ In that case, the two different SDE will describe the same stochastic process
90
Luc_Faucheux_2021
The crux of the matter - VI
ยจ And conversely:
ยจ If we start from a ITO SDE: ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก)
ยจ The corresponding STRATO SDE is: ๐‘‘๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก)
ยจ With:
ยจ @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก โˆ’ [
*
!
]. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
๐‘ ๐‘ก, ๐‘‹ ๐‘ก
ยจ A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก = ๐‘ ๐‘ก, ๐‘‹ ๐‘ก
ยจ @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก โˆ’ [
*
!
]. A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก
ยจ In that case, the two different SDE will describe the same stochastic process
91
Luc_Faucheux_2021
The crux of the matter - VII
ยจ Some note on notation:
ยจ To be even more rigorous, we should not even write something like this:
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก)
ยจ We should say that there is an analytical function ๐‘Ž ๐‘ก, ๐‘ฅ and ๐‘ ๐‘ก, ๐‘ฅ , and that the process is:
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘ฅ = ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘ฅ = ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก)
ยจ ๐‘‘๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘ฅ = ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘ฅ = ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก)
ยจ And so really we should not write:
ยจ @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก โˆ’ [
*
!
]. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
๐‘ ๐‘ก, ๐‘‹ ๐‘ก
ยจ But:
ยจ @
๐‘Ž ๐‘ก, ๐‘ฅ = ๐‘Ž ๐‘ก, ๐‘ฅ โˆ’
*
!
. ๐‘ ๐‘ก, ๐‘ฅ .
,
,0
๐‘ ๐‘ก, ๐‘ฅ
92
Luc_Faucheux_2021
The crux of the matter - VIII
ยจ In fact, whenever the notation doe not get too cumbersome, we should not even write and
SDE but and SIE
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก)
ยจ ๐‘‹ ๐‘ก% โˆ’ ๐‘‹ ๐‘ก$ = โˆซ
"#"$
"#"%
๐‘‘๐‘‹ ๐‘ก = โˆซ
"#"$
"#"%
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก) + โˆซ
"#"$
"#"%
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก)
ยจ Knowing that all terms are well defined with the added :
ยจ โˆซ
"#"$
"#"%
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) = lim
&โ†’(
{โˆ‘)#*
)#&
๐‘(๐‘ก), ๐‘‹(๐‘ก))). [๐‘Š(๐‘ก)+*) โˆ’ ๐‘Š(๐‘ก))]}
ยจ Where the mesh or partition {๐‘ก)} is not completely pathological
ยจ Also for STRATO:
ยจ โˆซ
"#"$
"#"%
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก) = lim
&โ†’(
{โˆ‘)#*
)#&
๐‘“ ๐‘ก),
[-("2)+-("234)]
!
. [๐‘Š(๐‘ก)+*) โˆ’ ๐‘Š(๐‘ก))]}
93
Luc_Faucheux_2021
The crux of the matter - IX
ยจ So in the end, this is the only way to write a process that leaves no room for error:
ยจ When writing something like this, we have no idea what it means
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘Š(๐‘ก)
ยจ Really we should write:
ยจ ๐‘‹ ๐‘ก% โˆ’ ๐‘‹ ๐‘ก$ = โˆซ
"#"$
"#"%
๐‘‘๐‘‹ ๐‘ก = โˆซ
"#"$
"#"%
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก) + โˆซ
"#"$
"#"%
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ). ๐‘‘๐‘Š(๐‘ก)
ยจ โˆซ
"#"$
"#"%
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ). ๐‘‘๐‘Š(๐‘ก) = lim
&โ†’(
{โˆ‘)#*
)#&
๐‘(๐‘ก), ๐‘€R[๐‘ก), ๐‘ก)+*]). [๐‘Š(๐‘ก)+*) โˆ’ ๐‘Š(๐‘ก))]}
ยจ Where the function ๐‘€R[๐‘ก), ๐‘ก)+*] indicates where in the partition [๐‘ก), ๐‘ก)+*] we should take
the value of the stochastic process ๐‘‹ ๐‘ก
94
Luc_Faucheux_2021
The crux of the matter - X
ยจ If ITO, ๐›ผ = 0, and ๐‘€R ๐‘ก), ๐‘ก)+* = ๐‘‹(๐‘ก))
ยจ If STRATO, ๐›ผ = 1/2, and ๐‘€R ๐‘ก), ๐‘ก)+* =
[-("2)+-("234)]
!
ยจ If anything in between, ๐‘€R ๐‘ก), ๐‘ก)+* = ๐‘‹(๐‘ก)) + ๐›ผ. [ ๐‘‹(๐‘ก)+* โˆ’ ๐‘‹(๐‘ก))]
ยจ The nice thing about ITO is that the integral is a martingale, and that the integral also flows
the isometry rule.
ยจ The price to pay is that the usual rules of calculus (chain rule, Leibniz, Taylor expansion,
integration by part,..) are pretty much out of the window
ยจ The nice thing about STRATO is that the usual rules of calculus carry over FORMALLY in the
same manner (careful, they are not the same, it is a formal relation), so that is nice
ยจ However the integral is not a martingale and does not follow the isometry rule
95
Luc_Faucheux_2021
A brief history of the Ito-Stratanovitch
controversy
96
Luc_Faucheux_2021
A brief history
ยจ Because :
ยจ 1) it is a nice way to apply our knowledge of stochastic calculus
ยจ 2) you encounter it in textbooks
ยจ 3) it is still super confusing at times, I know I am still confused
ยจ It pays to explain why that came about.
ยจ But first of all, letโ€™s go one more time over the fact that actually there is NO controversy
ยจ By the way, if you meet anyone who tells you that they understand Ito versus Stratanovitch
perfectly, that person is either a liar or Van Kampen
97
Luc_Faucheux_2021
First of all there is no controversy
98
Luc_Faucheux_2021
There is no controversy
ยจ In order to make things less confusing that they could be, we are going to be a little literal on
the notation
ยจ Essentially,
ยจ PDEs are well defined, and rely on Newtonian/Leibniz/regular calculus
ยจ SDEs and SIEs for non-homogeneous diffusion coefficients (non-linear), or level depenedent
volatilities, are NOT well defined, and you need to choose a value for ๐›ผ
ยจ There is a correspondence between the different [๐›ผ] calculus
99
Luc_Faucheux_2021
There is no controversy - II
ยจ Ito is ๐›ผ = 0
ยจ Strato is ๐›ผ = 1/2
ยจ Because the factor (1/2) also shows up in Taylor expansion and Ito lemma, that is where a
lot of the confusion comes from. BTW, Ito lemma is NOT a Taylor expansion, it formally
looks like one and comes from using a Taylor expansion in calculating the integral as a limit
of sum, but it is NOT a Taylor expansion
ยจ So you cannot say (as sometimes it seems to be implied in textbooks).
ยจ Hey in stochastic calculus, it is the same as regular calculus, just go up one more level in the
Taylor expansion
100
Luc_Faucheux_2021
There is no controversy - III
ยจ An ITO SDE is of the form:
ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก)
ยจ This implies that the PDF follows the FORWARD ITO Kolmogorov PDE
ยจ
,=(0,"|-.,".)
,"
= โˆ’
,
,0
๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’
,
,0
[
*
!
. [๐‘(๐‘ฅ, ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)]
ยจ With simpler notation:
ยจ An ITO SDE is of the form:
ยจ ๐‘‘๐‘‹ = ๐‘Ž. ๐‘‘๐‘ก + ๐‘. ([). ๐‘‘๐‘Š
ยจ This implies that the PDF follows the FORWARD ITO Kolmogorov PDE (Fokker-Planck)
ยจ
,=
,"
= โˆ’
,
,0
๐‘Ž. ๐‘ โˆ’
,
,0
[
*
!
. [๐‘! . ๐‘]
101
Luc_Faucheux_2021
There is no controversy - IV
ยจ An STRATO SDE is of the form:
ยจ ๐‘‘๐‘‹ ๐‘ก = @
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก)
ยจ This implies that the PDF follows the FORWARD STRATO Kolmogorov PDE
ยจ
!"($,&|(!,&!)
!&
= โˆ’
!
!(
{G
๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก +
+
,
. B
๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
!
!(
B
๐‘ ๐‘ก, ๐‘‹ ๐‘ก }. ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹*, ๐‘ก* โˆ’
!
!(
[
+
,
. [B
๐‘(๐‘‹ ๐‘ก , ๐‘ก), . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹*, ๐‘ก*)]
ยจ With simpler notation:
ยจ An STRATO SDE is of the form:
ยจ ๐‘‘๐‘‹ = @
๐‘Ž. ๐‘‘๐‘ก + A
๐‘. (โˆ˜). ๐‘‘๐‘Š
ยจ This implies that the PDF follows the FORWARD STRATO Kolmogorov PDE (Fokker-Planck)
ยจ
,=
,"
= โˆ’
,
,0
{@
๐‘Ž +
*
!
. A
๐‘.
,
,0
A
๐‘}. ๐‘ โˆ’
,
,0
[
*
!
. [A
๐‘! . ๐‘]
102
Luc_Faucheux_2021
There is no controversy - V
ยจ So and ITO SDE is:
ยจ ๐‘‘๐‘‹ = ๐‘Ž. ๐‘‘๐‘ก + ๐‘. ([). ๐‘‘๐‘Š
ยจ And corresponds to a diffusion equation for the PDF:
ยจ
,=
,"
= โˆ’
,
,0
๐‘Ž. ๐‘ โˆ’
,
,0
[
*
!
. ๐‘! . ๐‘]
ยจ A STRATO SDE is:
ยจ ๐‘‘๐‘‹ = @
๐‘Ž. ๐‘‘๐‘ก + A
๐‘. (โˆ˜). ๐‘‘๐‘Š
ยจ And corresponds to a diffusion equation for the PDF:
ยจ
,=
,"
= โˆ’
,
,0
{@
๐‘Ž +
*
!
. A
๐‘.
,
,0
A
๐‘}. ๐‘ โˆ’
,
,0
[
*
!
. A
๐‘! . ๐‘]
103
Luc_Faucheux_2021
There is no controversy - VI
ยจ So if we want the same solution, meaning we want the 2 equations for the PDE to be equal:
ยจ
,=
,"
= โˆ’
,
,0
๐‘Ž. ๐‘ โˆ’
,
,0
[
*
!
. ๐‘! . ๐‘]
ยจ
,=
,"
= โˆ’
,
,0
{@
๐‘Ž +
*
!
. A
๐‘.
,
,0
A
๐‘}. ๐‘ โˆ’
,
,0
[
*
!
. A
๐‘! . ๐‘]
ยจ That means as we have seen before that we need to write:
ยจ A
๐‘ = ๐‘
ยจ ๐‘Ž = @
๐‘Ž +
*
!
. A
๐‘.
,
,0
A
๐‘
ยจ That is the correspondence between the ITO and STRATO coefficients for the SDEs
ยจ Pointing out the obvious one more time, when
,
,0
A
๐‘ =
,
,0
๐‘ = 0, both calculus will return
exactly the same SDE and PDE, the SDE is called linear
104
Luc_Faucheux_2021
Back to a brief history of the Ito-
Stratanovitch controversy
(a physicistโ€™s point of view)
105
Luc_Faucheux_2021
A brief history - II
ยจ So if you are a physicist, you like to write diffusion equations like:
ยจ
,=
,"
= โˆ’
,
,0
[๐ฝ/ + ๐ฝA]
ยจ ๐ฝA is the diffusion current (hence the D notation)
ยจ ๐ฝ/ is the drift current, coming usually from an external force (hence the F notation)
ยจ In the absence of external force, physicist would expect the equilibrium solution to be
uniform (meaning ๐‘ = ๐‘๐‘ก๐‘’), and so it makes sense for them to write the diffusion current
as:
ยจ ๐ฝA = โˆ’๐ท.
,=
,0
ยจ If the diffusion coefficient is a function of the position (and the physical system described is
such that the diffusive process does not impact the long term equilibrium steady state
derived from thermodynamics), physicists will still expect a uniform distribution and will still
want the diffusion coefficient to be outside the derivative
ยจ ๐ฝA = โˆ’๐ท(๐‘ฅ).
,=
,0
106
Luc_Faucheux_2021
A brief history - III
ยจ For sake of simplicity letโ€™s say that there is no external forcing right now, ๐ฝ/ = 0
ยจ So physicist in the presence of non-homogeneous diffusion would like the PDE to look like:
ยจ
,=
,"
= โˆ’
,
,0
๐ฝA = โˆ’
,
,0
โˆ’๐ท(๐‘ฅ).
,=
,0
ยจ Nothing wrong so far.
ยจ The problem arose when we started using digital computers and started running Monte
Carlo simulations
ยจ This is fairly recent so we should not beat ourselves too much on the fact that we tripped on
that one.
ยจ Following the work of Bachelier, Bernouilli and such, we started modeling the stochastic
process as:
ยจ ๐›ฟ๐‘‹ ๐‘ก = ยฑ 2๐ท with the ยฑ being a random number generator (head or tail)
107
Luc_Faucheux_2021
A brief history - IV
ยจ ๐›ฟ๐‘‹ ๐‘ก = ยฑ 2๐ท
ยจ Is essentially what we looked at in the firs deck, as well as the Bachelier deck
108
X
i i+1
i-1
Luc_Faucheux_2021
A brief history - V
ยจ We know for a contant spacing (constant
diffusion coefficient), the distribution is the
usual binomial which converges to the
Gaussian, as illustrated by the Galton
machine.
ยจ By the way Galton was born not even 200
years ago, so again we should go easy on
ourselves
109
Luc_Faucheux_2021
A brief history - VI
ยจ Ok so so far so good.
ยจ The issue is when we started simulating on a digital computer something like that:
ยจ ๐›ฟ๐‘‹ ๐‘ก = ยฑ 2๐ท(๐‘‹ ๐‘ก )
ยจ Because, when we write something like the above, even without knowing it, we are in ITO
calculus and we are writing really:
ยจ ๐‘‘๐‘‹(๐‘ก) = 2๐ท(๐‘‹ ๐‘ก ). ([). ๐‘‘๐‘Š(๐‘ก)
ยจ This will give a PDE:
ยจ
,=
,"
= โˆ’
,
,0
๐‘Ž. ๐‘ โˆ’
,
,0
[
*
!
. ๐‘! . ๐‘]
ยจ With:
ยจ ๐‘Ž = 0
ยจ ๐‘ = 2๐ท(๐‘ฅ)
110
Luc_Faucheux_2021
A brief history - VII
ยจ
,=
,"
= โˆ’
,
,0
โˆ’
,
,0
[
*
!
. 2๐ท(๐‘ฅ)
!
. ๐‘]
ยจ
,=
,"
= โˆ’
,
,0
โˆ’
,
,0
[๐ท(๐‘ฅ). ๐‘]
ยจ The solution of that (subject to non-diverging boundary conditions) is:
ยจ ๐‘ = 1/๐ท(๐‘ฅ)
ยจ And NOT ๐‘ = ๐‘๐‘ก๐‘’, uniform distribution as expected.
ยจ So that was the first sign that something was wrong.
111
Luc_Faucheux_2021
A brief history - VIII
ยจ Letโ€™s recap.
ยจ You want to model:
ยจ
,=
,"
= โˆ’
,
,0
๐ฝA = โˆ’
,
,0
โˆ’๐ท(๐‘ฅ).
,=
,0
ยจ You do a discrete numerical simulation:
ยจ ๐›ฟ๐‘‹ ๐‘ก = ยฑ 2๐ท(๐‘‹ ๐‘ก )
ยจ You get:
ยจ
,=
,"
= โˆ’
,
,0
โˆ’
,
,0
[๐ท(๐‘ฅ). ๐‘]
ยจ Not good, you do not get what you expected
112
Luc_Faucheux_2021
A brief history - IX
ยจ So at that point, if physicists in the 80s were as good on stochastic calculus as they should
be, they should have said.
ยจ Well yeah, a discrete simulation is by definition in the world of ITO calculus
ยจ If I write: ๐›ฟ๐‘‹ ๐‘ก = ยฑ 2๐ท(๐‘‹ ๐‘ก )
ยจ I really am writing: ๐‘‘๐‘‹(๐‘ก) = 2๐ท(๐‘‹ ๐‘ก ). ([). ๐‘‘๐‘Š(๐‘ก)
ยจ Which will give me:
,=
,"
= โˆ’
,
,0
โˆ’
,
,0
[๐ท(๐‘ฅ). ๐‘]
ยจ But I want:
,=
,"
= โˆ’
,
,0
โˆ’๐ท(๐‘ฅ).
,=
,0
113
Luc_Faucheux_2021
A brief history - X
ยจ Since I want:
,=
,"
= โˆ’
,
,0
โˆ’๐ท(๐‘ฅ).
,=
,0
ยจ I can write:
ยจ
,=
,"
= โˆ’
,
,0
โˆ’
,
,0
๐ท ๐‘ฅ . ๐‘ = โˆ’
,
,0
โˆ’
,
,0
๐ท ๐‘ฅ . ๐‘ โˆ’
,
,0
๐‘ . ๐ท(๐‘ฅ)
ยจ
,=
,"
= โˆ’
,
,0
โˆ’
,
,0
๐ท ๐‘ฅ . ๐‘ โˆ’ ๐ท(๐‘ฅ).
,=
,0
ยจ That is of the form:
ยจ
,=
,"
= โˆ’
,
,0
๐‘Ž. ๐‘ โˆ’
,
,0
[
*
!
. ๐‘! . ๐‘]
ยจ Which in ITO will give the SDE
ยจ ๐‘‘๐‘‹ = ๐‘Ž. ๐‘‘๐‘ก + ๐‘. ([). ๐‘‘๐‘Š
ยจ With: ๐‘Ž = โˆ’
,
,0
๐ท ๐‘ฅ and ๐‘ = 2๐ท(๐‘‹ ๐‘ก )
114
Luc_Faucheux_2021
A brief history - XI
ยจ Soooโ€ฆ bear with me here a few more slides..
ยจ If I write:
ยจ ๐‘‘๐‘‹ = {
,
,0
๐ท ๐‘ฅ }. ๐‘‘๐‘ก + 2๐ท(๐‘‹ ๐‘ก ). ([). ๐‘‘๐‘Š
ยจ Numerically I will simulate it as:
ยจ ๐›ฟ๐‘‹ ๐‘ก =
,
,0
๐ท ๐‘ฅ ยฑ 2๐ท(๐‘‹ ๐‘ก )
ยจ That is assuming units of time equal to 1, if not you write
ยจ ๐›ฟ๐‘‹ ๐‘ก =
,
,0
๐ท ๐‘ฅ . ๐›ฟ๐‘ก ยฑ 2๐ท ๐‘‹ ๐‘ก . ๐›ฟ๐‘ก
115
Luc_Faucheux_2021
A brief history - XII
ยจ I simulate: ๐›ฟ๐‘‹ ๐‘ก =
,
,0
๐ท ๐‘ฅ . ๐›ฟ๐‘ก ยฑ 2๐ท ๐‘‹ ๐‘ก . ๐›ฟ๐‘ก
ยจ That is an ITO SDE: ๐‘‘๐‘‹ = {
,
,0
๐ท ๐‘ฅ }. ๐‘‘๐‘ก + 2๐ท(๐‘‹ ๐‘ก ). ([). ๐‘‘๐‘Š
ยจ That has a PDE solution:
,=
,"
= โˆ’
,
,0
๐‘Ž. ๐‘ โˆ’
,
,0
[
*
!
. ๐‘! . ๐‘]
ยจ With: ๐‘Ž =
,
,0
๐ท ๐‘ฅ and ๐‘ = 2๐ท(๐‘‹ ๐‘ก )
ยจ So you get the PDE:
,=
,"
= โˆ’
,
,0
,
,0
๐ท ๐‘ฅ . ๐‘ โˆ’
,
,0
[๐ท. ๐‘]
ยจ And:
ยจ
,=
,"
= โˆ’
,
,0
,
,0
๐ท ๐‘ฅ . ๐‘ โˆ’
,
,0
๐ท. ๐‘ = โˆ’
,
,0
,A
,0
. ๐‘ โˆ’
,A
,0
. ๐‘ โˆ’ ๐ท.
,=
,0
= โˆ’
,
,0
โˆ’๐ท.
,=
,0
ยจ Which is what I want
116
Luc_Faucheux_2021
A brief history - XIII
ยจ SOโ€ฆ
ยจ If I want to simulate:
,=
,"
= โˆ’
,
,0
โˆ’๐ท(๐‘ฅ).
,=
,0
ยจ Which has no drift term, and has the equilibrium long-term steady state uniform distribution
as the desired solution, I need to simulate on a discrete digital computer:
ยจ ๐›ฟ๐‘‹ ๐‘ก =
,
,0
๐ท ๐‘ฅ . ๐›ฟ๐‘ก ยฑ 2๐ท ๐‘‹ ๐‘ก . ๐›ฟ๐‘ก
ยจ Which seems weird, because the first term
,
,0
๐ท ๐‘ฅ . ๐›ฟ๐‘ก looks like the drift term.
ยจ This is when the whole confusion started
117
Luc_Faucheux_2021
A brief history - XIV
ยจ ๐›ฟ๐‘‹ ๐‘ก =
,
,0
๐ท ๐‘ฅ . ๐›ฟ๐‘ก ยฑ 2๐ท ๐‘‹ ๐‘ก . ๐›ฟ๐‘ก
ยจ The term
,
,0
๐ท ๐‘ฅ . ๐›ฟ๐‘ก looks like indeed what you would simulate if you were to simulate
a drift
ยจ It is NOT a physical drift
ยจ Your PDE that you wanted to simulate was
,=
,"
= โˆ’
,
,0
โˆ’๐ท.
,=
,0
and had NO drift
ยจ It is purely a term that you need to add to the simulation to recover your PDE
ยจ This is why it was called a โ€œspurious driftโ€ (Ryter 1980).
ยจ Again, it is NOT a drift, you are not extraction motion out of noise, you need to have it in
there because you are using Ito calculus, and plain and simple:
ยจ ๐‘‘๐‘‹ = ๐‘Ž. ๐‘‘๐‘ก + ๐‘. ([). ๐‘‘๐‘Š will give you
,=
,"
= โˆ’
,
,0
๐‘Ž. ๐‘ โˆ’
,
,0
[
*
!
. ๐‘! . ๐‘]
ยจ Period..end of storyโ€ฆ
118
Luc_Faucheux_2021
A brief history - XV
ยจ But you can see why that was confusing, and why it tripped at the time a lot of people, yours
truly included.
ยจ And that was confusing for 2 main reasons:
ยจ Reason 1: Maxwell demon and Thermal ratchets
ยจ Reason 2: Stratanovitch
ยจ They are somewhat related but distinct.
ยจ Letโ€™s go over Stratanovitch first
119
Luc_Faucheux_2021
A brief history โ€“ XVI
ยจ To be fair, I was also a little harsh and oversimplifying.
ยจ There is also another reason why the question of Ito versus Stratanovitch came about, it has
to do with the autocorrelation of the random forcing (the noise). Is it white noise with 0
memory, and so Ito would be appropriate, or is it coloured noise, with some non zero
autocorrelation function, and in this case Strato would be more appropriate.
ยจ Again it really depends on the physical system, an what PDE you want
ยจ Radioactive particle decay lends itself well to Ito
ยจ Diffusion in a viscous medium lends itself better to Strato from a physics point of view
ยจ So as always things are a little more complicated that I would make them out to be, so you
should take what I say with a grain of salt.
120
Luc_Faucheux_2021
A brief history โ€“ XVII
ยจ In particular, if there is any kind of correlation, then an non-ITO approach is sometimes
justified.
ยจ Remember that we had from the earlier part of the deck:
ยจ < A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . [๐›ผ] . ๐‘‘๐‘Š > = ๐›ผ. A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก
ยจ In particular:
ยจ ITO : < A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . [ . ๐‘‘๐‘Š > = 0
ยจ STRATO: < A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . โˆ˜ . ๐‘‘๐‘Š > = [
*
!
]. A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก .
,
,-
A
๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก
121
Luc_Faucheux_2021
Reason #2 for the confusion:
People started to hear about Stratanovitch
calculus
122
Luc_Faucheux_2021
Reason #2 for the confusion - Stratanovitch
ยจ So first of all ITO calculus is weird because of ITO lemma, and the fact that the usual rules of
calculus (Leibniz rule, chain rule) are not formally respected
ยจ Second ITO is very well adapted to DISCRETE processes in time (finance, but also simulations
on digital computers)
ยจ Physicists do not like to learn new rules of calculus, they would rather stick to the usual one
ยจ More seriously, in physics, it could be argued that VERY FEW processes are truly discrete
(radioactive decay for example being one of them), but most of the processes (like
diffusion), are not discrete but continuous
123
Luc_Faucheux_2021
Reason #2 for the confusion โ€“ Stratanovitch - II
ยจ We know that using STRATO calculus:
ยจ ๐‘‘๐‘‹ = @
๐‘Ž. ๐‘‘๐‘ก + A
๐‘. (โˆ˜). ๐‘‘๐‘Š
ยจ Corresponds to a diffusion equation for the PDF:
ยจ
,=
,"
= โˆ’
,
,0
{@
๐‘Ž +
*
!
. A
๐‘.
,
,0
A
๐‘}. ๐‘ โˆ’
,
,0
[
*
!
. A
๐‘! . ๐‘]
ยจ We can expand the last term into:
ยจ
,
,0
*
!
. A
๐‘! . ๐‘ =
,
,0
*
!
. A
๐‘. A
๐‘. ๐‘ =
*
!
.
,
,0
A
๐‘. A
๐‘. ๐‘ =
*
!
.
,
,0
A
๐‘. A
๐‘. ๐‘
ยจ
,
,0
*
!
. A
๐‘! . ๐‘ =
*
!
. A
๐‘. ๐‘ .
,
,0
A
๐‘ +
*
!
. A
๐‘.
,
,0
A
๐‘. ๐‘
ยจ
,
,0
*
!
. A
๐‘! . ๐‘ =
*
!
. A
๐‘.
,B
%
,0
. ๐‘ +
*
!
. A
๐‘.
,
,0
A
๐‘. ๐‘
124
Luc_Faucheux_2021
Reason #2 for the confusion โ€“ Stratanovitch - III
ยจ
,=
,"
= โˆ’
,
,0
{@
๐‘Ž +
*
!
. A
๐‘.
,
,0
A
๐‘}. ๐‘ โˆ’
,
,0
[
*
!
. A
๐‘! . ๐‘]
ยจ
,=
,"
= โˆ’
,
,0
@
๐‘Ž +
*
!
. A
๐‘.
,
,0
A
๐‘ . ๐‘ โˆ’ {
*
!
. A
๐‘.
,B
%
,0
. ๐‘ +
*
!
. A
๐‘.
,
,0
A
๐‘. ๐‘ }
ยจ
,=
,"
= โˆ’
,
,0
@
๐‘Ž . ๐‘ โˆ’
*
!
. A
๐‘.
,
,0
A
๐‘. ๐‘
ยจ So a STRATO SDE of the form:
ยจ ๐‘‘๐‘‹ = @
๐‘Ž. ๐‘‘๐‘ก + A
๐‘. (โˆ˜). ๐‘‘๐‘Š
ยจ Will give a PDE of the form:
ยจ
,=
,"
= โˆ’
,
,0
@
๐‘Ž . ๐‘ โˆ’
*
!
. A
๐‘.
,
,0
A
๐‘. ๐‘
125
Luc_Faucheux_2021
Reason #2 for the confusion โ€“ Stratanovitch - IV
ยจ Again in order to illustrate letโ€™s assume no drift term in the PDE
ยจ So a STRATO SDE of the form:
ยจ ๐‘‘๐‘‹ = A
๐‘. (โˆ˜). ๐‘‘๐‘Š
ยจ Will give a PDE of the form:
ยจ
,=
,"
= โˆ’
,
,0
โˆ’
*
!
. A
๐‘.
,
,0
A
๐‘. ๐‘
ยจ Again, nothing wrong there, but physicist got confused because they starting saying things
like the next slide:
126
Luc_Faucheux_2021
Reason #2 for the confusion โ€“ Stratanovitch - V
ยจ If I write: ๐‘‘๐‘‹ = A
๐‘. (โˆ˜). ๐‘‘๐‘Š
ยจ On a computer I simulate ๐›ฟ๐‘‹ = ยฑA
๐‘
ยจ And that gives me a PDE with no drift:
,=
,"
= โˆ’
,
,0
โˆ’
*
!
. A
๐‘.
,
,0
A
๐‘. ๐‘
ยจ So I am happy, but now my diffusion current is no longer:
ยจ
,=
,"
= โˆ’
,
,0
๐ฝA = โˆ’
,
,0
โˆ’๐ท(๐‘ฅ).
,=
,0
with A
๐‘ = 2๐ท
ยจ But:
ยจ
,=
,"
= โˆ’
,
,0
โˆ’
*
!
. A
๐‘.
,
,0
A
๐‘. ๐‘
ยจ So my diffusion current is really: ๐ฝA = โˆ’
*
!
. A
๐‘.
,
,0
A
๐‘. ๐‘ = โˆ’ ๐ท.
,
,0
๐ท. ๐‘
127
Luc_Faucheux_2021
Reason #2 for the confusion โ€“ Stratanovitch - VI
ยจ So if I do not want to simulate a drit term (because my PDE has no drift term), then really I
need to write the diffusion current as:
ยจ ๐ฝA = โˆ’ ๐ท.
,
,0
๐ท. ๐‘
ยจ But thenโ€ฆwait a minute, if I plug this back into the PDE, then my equilibrium distribution is
no longer uniform (in fact it will be ๐ท. ๐‘ = ๐‘๐‘ก๐‘’ or ๐‘ = 1/ ๐ท. )
ยจ So when the diffusion coefficient is non-homogeneous it will change my equilibrium
distribution.
ยจ I have โ€œextractedโ€ motion our of random noise
ยจ This is maybe an example of the Maxwell demon
ยจ I have broken the law of thermodynamics, so what is next ? Perpetual motion ?
128
Luc_Faucheux_2021
Reason #2 for the confusion โ€“ Stratanovitch - VII
ยจ The previous 2 slides are obviously wrong but at the time that was a real debate in Physics
ยจ For a constant diffusion the diffusion current is :
ยจ ๐ฝA = โˆ’ ๐ท.
,
,0
๐ท. ๐‘ = โˆ’๐ท.
,
,0
๐‘ = โˆ’
,
,0
๐ท. ๐‘
ยจ Because who cares, ๐ท is a constant
ยจ But now if we have ๐ท(๐‘‹(๐‘ก)) for a Brownian particle, which one is the real diffusion current?
ยจ And then there were even more confusion about the Ryter spurious drift maybe being an
actual drift because
ยจ
,
,0
๐ท. ๐‘ =
,
,0
๐ท . ๐‘ + ๐ท.
,
,0
๐‘
ยจ So maybe ๐ท.
,
,0
๐‘ is the โ€œrealโ€ diffusion current and the term
,
,0
๐ท . ๐‘ is the Maxwell
demon term extracting directed motion out of a purely random diffusion process?
129
Luc_Faucheux_2021
Reason #2 for the confusion โ€“ Stratanovitch - VIII
ยจ The whole controversy could have been squashed had there been one good math guy who
understood Ito and Strato well enough to explain it simply to physicist and say:
ยจ Listen you idiots, all that really matters is the PDE
ยจ If you want to play with a computer and burn some CPUs and start global warming by
writing something like: ๐›ฟ๐‘‹ ๐‘ก = ยฑ 2๐ท(๐‘‹ ๐‘ก )
ยจ That is fine by me, but you are by definition using ITO calculus and writing:
ยจ ๐‘‘๐‘‹(๐‘ก) = 2๐ท(๐‘‹ ๐‘ก ). ([). ๐‘‘๐‘Š(๐‘ก)
ยจ Which will give you:
,=
,"
= โˆ’
,
,0
โˆ’
,
,0
[๐ท(๐‘ฅ). ๐‘]
ยจ Which will not give you a uniform distribution for the steady state regime.
ยจ Period, full stop, end of story.
130
Luc_Faucheux_2021
Reason #2 for the confusion โ€“ Stratanovitch - IX
ยจ And then this math guy could have added:
ยจ If you idiots want to simulate the SDE that corresponds to the PDE:
,=
,"
= โˆ’
,
,0
โˆ’๐ท(๐‘ฅ).
,=
,0
ยจ And if you insist on using discrete numerical simulations, then sorry mate but you have no
choice but to write:
ยจ ๐›ฟ๐‘‹ ๐‘ก =
,
,0
๐ท ๐‘ฅ ยฑ 2๐ท ๐‘‹ ๐‘ก
ยจ So yeah, the first term looks like a drift term, but there is no physical drift in your PDE, it is
just a โ€œspuriousโ€ drift because you now have entered the mysterious world of stochastic
calculus, and more precisely ITO calculus. And please leave at the door all the intuitions you
had built on Newtonian/Leibniz/regular calculus, and buckle up buttercup, because you not
in Kansas anymore Toto.
ยจ Sometimes I wish math guys could not talk to physicists in such condescending manner.
131
Luc_Faucheux_2021
Reason #1 for the confusion:
Maxwell demon, Brownian motors, Thermal
ratchets
132
Luc_Faucheux_2021
Maxwell demon - I
ยจ The other reason why there was some confusion is that at the same time a lot of physicists
were looking at issues like Maxwell demon, Brownian engines, Brownian ratchets, Thermal
ratchetsโ€ฆ
ยจ The Maxwell demon is a possible (but very improbable) entity that drive a system towards a
state that is possible but very improbable
ยจ To quote the adventures of Mr. Tompkins, it is possible (but very unlikely) the ice cube that
you melted in your coke will spontaneously re-form as an ice cube and that the surrounding
liquid will warm back up again.
ยจ That is possible, there is nothing in Physics that says that it is impossible, it is just very
unlikely
ยจ Winning the lottery is also possible, it is just also highly unlikely
133
Luc_Faucheux_2021
Maxwell demon - II
ยจ The Maxwell demon is intriguing because the 2nd Law of Thermodynamics says that usually
things tend to decay, and that a measure of disorder that you can define (entropy) usually
always increase at the end.
ยจ That law makes sense, we all see it in our lives everyday, things always tend to go to sh..,
never (or highly unlikely) the other way around
ยจ What is true for our personal lives is true for the Universe
ยจ Nothing shocking there, since we are part of the universe, so the laws should be somewhat
universal
134
Luc_Faucheux_2021
Maxwell demon - III
ยจ The Maxwell demon putting the moves on Mr. Tompkins fiancรฉe and trying to impress her
with his tennis skills by reconstructing an ice cube without violating the 2nd law of
Thermodynamics
135
Luc_Faucheux_2021
Maxwell demon - IV
ยจ A more common representation of the demon
136
Luc_Faucheux_2021
Brownian Engines
ยจ So it would make sense for physicists to ask if it was possible to build a Maxwell demon.
ยจ Feynman in his lecture has a great chapter on this.
ยจ He even offered a prize to the first person / team who could build one.
137
Luc_Faucheux_2021
Brownian engines - II
ยจ The canonical example of a thermal ratchet from the Feynmanโ€™s lectures in Physics
138
Luc_Faucheux_2021
Brownian engines - III
ยจ There was also another reason why physicists were excited by all this, (none withstanding
the fact that if you could receive a check from Richard Feynman, that would be the coolest
thing ever, and you would have so much street cred), was that we were starting to look in
real-time in-vivo inside the cells and human body, and whether we like it or not (and again
nothing really to do with intelligent design theory), there were some really cool things
happening there.
ยจ Vesicles are transported within the cell, to the place where they should be, in a very highly
โ€œnoisyโ€ (read bombarded by random fluctuations) environment that should make it
impossible. Yet it works.
ยจ Spermatozoides are able to โ€œswimโ€ to where they were supposed to go, again in a very
random and viscous environment.
ยจ Trying to build a typical inertial engine and computing how much energy that required lead
to the conclusion that it was impossible, so clearly there was something akin to a Thermal
ratchet, or Brownian engine at work there. So somehow Nature found a way to extract
directed motion out of random noise, or at least found a way to still be able to build things
in a highly noisy, random and viscous regime without โ€œforcingโ€ its way through the way an
inertial engine (think of a car or a rocket) would do at the scale we are used to
139
Luc_Faucheux_2021
Brownian engines - IV
ยจ I was lucky enough to work at the time in the lab of Albert Libchaber, who was
fascinated with those problems.
ยจ I have never met in my life someone who was so pure in his search of the truth and who
was so keen on lifting the curtains, and who had the intellectual power to cut through
the math, or the obstacles between him and the hidden simple principles at work
ยจ I count myself as being fortunate to have been able to witness him at work, and in his
life.
140
Luc_Faucheux_2021
Brownian engines - V
ยจ I could write a book about how much fun we had in his lab (and maybe I should), but suffice
to say that we were all involved in the subject of Brownian engines, whether on the biology
side or on the more Physics side. If you guys are interested you have to read this one:
141
Luc_Faucheux_2021
Brownian engines - VI
ยจ So anyways long story short, I spent a lot of time then on those issues, had the pleasure of
working with some of the finest minds of the time (too numerous to name all, and am not
even going to try, but of course Albert Libchaber, Mike Shelley, Dave Muraki, Marcello
Magnasco, Erez Braun, Elisha Moses, Drew Belmonte, Deborah Fygenson, Albrecht Ott,
Andreas Tilgner, Gustavo Stolovitsky, Rolf Landauer and is famous shout โ€œcomplex is complex
is complex!โ€, and again way too many I apologize for not including in the short list above)
ยจ But anyways we built the first optical thermal ratchet !
142
Luc_Faucheux_2021
Brownian engines - VII
ยจ It turns out that there is a cool connection to Finance.
ยจ If you replace the position by the accumulated PL, the thermal ratchet becomes what is
known as the Parrondo Paradox.
ยจ If we just had thought about replacing X by $, that could have been known as the Libchaber
paradox
ยจ Another example of the work of an illustrious Frenchman that is easily applicable to finance
but somehow someone else takes it through the goal lineโ€ฆ
ยจ Essentially, just like the thermal ratchet, you can โ€œextractโ€ some direction our of random
fluctuations (no worries about the 2nd law, it is still intact, because you need to dump energy
into the ratchet to make it work)
ยจ So in Finance, you could have 2 losing strategies (2 loser PM in a fund), but if you switch
randomly the allocation of $ between those 2 strategies, you can find a regime where the
resulting strategy will make $ on average
143
Luc_Faucheux_2021
Brownian engines - VIII
ยจ There are actually applications in Finance of the Maxell demon, known as the Parrondo
paradox.
ยจ Two trading strategies (PM at a hedge fund) on average lose money (B and C, blue and green
line)
ยจ However you can alternate between the two strategies to create one (A-red line) that on
average will be profitable, like the thermal ratchet who is extracting work out of thermal
noise, this Parrondo construct extract positive return out of random switches between two
losing strategies
144
Luc_Faucheux_2021
Another draft of retelling the story of the Ito-
Stratanovitch controversy
145
Luc_Faucheux_2021
Another way to tell the story โ€“ I
ยจ This was a first draft of the story, it was a little more complicated than the one I went
through, but thought I would keep it here for sake of completeness.
ยจ No need to read it
ยจ It has nice pictures though
ยจ Please enjoy
146
Luc_Faucheux_2021
Another way to tell the story - II
ยจ All right, this is where all the confusion came about the Ito-Stratanovitch controversy (and
also that physicists tend to glance over stochastic calculus and be quite liberal and not that
mathematically rigorous, I have to admit)
ยจ Suppose that we start with a PDE describing the following process:
ยจ
,=(0,")
,"
= โˆ’
,
,-
๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก . ๐‘(๐‘ฅ, ๐‘ก) โˆ’
,
,-
[
*
!
. [๐‘(๐‘‹ ๐‘ก , ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก)]
ยจ
,
,"
. ๐‘ ๐‘ฅ, ๐‘ก = โˆ’
,
,0
[๐ฝ/+๐ฝA]
ยจ ๐ฝ/ ๐‘ฅ, ๐‘ก = ๐‘Ž(๐‘ก). ๐‘ ๐‘ฅ, ๐‘ก and ๐ฝA ๐‘ฅ, ๐‘ก = โˆ’
,
,-
[
*
!
. [๐‘(๐‘‹ ๐‘ก , ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก)]
ยจ ๐ฝA ๐‘ฅ, ๐‘ก = โˆ’
,
,-
[
*
!
. [๐œŽ(๐‘‹ ๐‘ก , ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก)]
ยจ ๐ฝA ๐‘ฅ, ๐‘ก = โˆ’
,
,-
[๐ท(๐‘ฅ, ๐‘ก). ๐‘(๐‘ฅ, ๐‘ก)] with ๐ท ๐‘ฅ, ๐‘ก =
*
!
. [๐‘(๐‘‹ ๐‘ก , ๐‘ก)! ]
147
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii

More Related Content

What's hot

Lf 2021 rates_vii
Lf 2021 rates_viiLf 2021 rates_vii
Lf 2021 rates_viiluc faucheux
ย 
Lf 2021 rates_v_a
Lf 2021 rates_v_aLf 2021 rates_v_a
Lf 2021 rates_v_aluc faucheux
ย 
Lf 2020 stochastic_calculus_ito-ii
Lf 2020 stochastic_calculus_ito-iiLf 2020 stochastic_calculus_ito-ii
Lf 2020 stochastic_calculus_ito-iiluc faucheux
ย 
Lf 2020 rates_i
Lf 2020 rates_iLf 2020 rates_i
Lf 2020 rates_iluc faucheux
ย 
Lf 2020 rates_iii
Lf 2020 rates_iiiLf 2020 rates_iii
Lf 2020 rates_iiiluc faucheux
ย 
Lf 2021 stochastic_calculus_ito-iii-a
Lf 2021 stochastic_calculus_ito-iii-aLf 2021 stochastic_calculus_ito-iii-a
Lf 2021 stochastic_calculus_ito-iii-aluc faucheux
ย 
Lf 2021 rates_viii
Lf 2021 rates_viiiLf 2021 rates_viii
Lf 2021 rates_viiiluc faucheux
ย 

What's hot (7)

Lf 2021 rates_vii
Lf 2021 rates_viiLf 2021 rates_vii
Lf 2021 rates_vii
ย 
Lf 2021 rates_v_a
Lf 2021 rates_v_aLf 2021 rates_v_a
Lf 2021 rates_v_a
ย 
Lf 2020 stochastic_calculus_ito-ii
Lf 2020 stochastic_calculus_ito-iiLf 2020 stochastic_calculus_ito-ii
Lf 2020 stochastic_calculus_ito-ii
ย 
Lf 2020 rates_i
Lf 2020 rates_iLf 2020 rates_i
Lf 2020 rates_i
ย 
Lf 2020 rates_iii
Lf 2020 rates_iiiLf 2020 rates_iii
Lf 2020 rates_iii
ย 
Lf 2021 stochastic_calculus_ito-iii-a
Lf 2021 stochastic_calculus_ito-iii-aLf 2021 stochastic_calculus_ito-iii-a
Lf 2021 stochastic_calculus_ito-iii-a
ย 
Lf 2021 rates_viii
Lf 2021 rates_viiiLf 2021 rates_viii
Lf 2021 rates_viii
ย 

Similar to Lf 2021 stochastic_calculus_ito-iii

On solving fuzzy delay differential equationusing bezier curves
On solving fuzzy delay differential equationusing bezier curves  On solving fuzzy delay differential equationusing bezier curves
On solving fuzzy delay differential equationusing bezier curves IJECEIAES
ย 
Python for High School Programmers
Python for High School ProgrammersPython for High School Programmers
Python for High School ProgrammersSiva Arunachalam
ย 
2. data types, variables and operators
2. data types, variables and operators2. data types, variables and operators
2. data types, variables and operatorsPhD Research Scholar
ย 
Numerical Solution Of Delay Differential Equations Using The Adomian Decompos...
Numerical Solution Of Delay Differential Equations Using The Adomian Decompos...Numerical Solution Of Delay Differential Equations Using The Adomian Decompos...
Numerical Solution Of Delay Differential Equations Using The Adomian Decompos...theijes
ย 
A Class of Polynomials Associated with Differential Operator and with a Gener...
A Class of Polynomials Associated with Differential Operator and with a Gener...A Class of Polynomials Associated with Differential Operator and with a Gener...
A Class of Polynomials Associated with Differential Operator and with a Gener...iosrjce
ย 
Declare Your Language: Transformation by Strategic Term Rewriting
Declare Your Language: Transformation by Strategic Term RewritingDeclare Your Language: Transformation by Strategic Term Rewriting
Declare Your Language: Transformation by Strategic Term RewritingEelco Visser
ย 
Split and list technique for solving hard problems
Split and list technique for solving hard problemsSplit and list technique for solving hard problems
Split and list technique for solving hard problemsRohit Kumar Singh
ย 
Engineering Equation Solver (Thai)
Engineering Equation Solver (Thai)Engineering Equation Solver (Thai)
Engineering Equation Solver (Thai)Denpong Soodphakdee
ย 
Some approximation properties of modified baskakov stancu operators
Some approximation properties of modified baskakov stancu operatorsSome approximation properties of modified baskakov stancu operators
Some approximation properties of modified baskakov stancu operatorseSAT Journals
ย 
CP2-Chp2-Series.pptx
CP2-Chp2-Series.pptxCP2-Chp2-Series.pptx
CP2-Chp2-Series.pptxNasimSalim2
ย 
Python quickstart for programmers: Python Kung Fu
Python quickstart for programmers: Python Kung FuPython quickstart for programmers: Python Kung Fu
Python quickstart for programmers: Python Kung Fuclimatewarrior
ย 
Basic calculus (ii) recap
Basic calculus (ii) recapBasic calculus (ii) recap
Basic calculus (ii) recapFarzad Javidanrad
ย 
A05330107
A05330107A05330107
A05330107IOSR-JEN
ย 
Solving Partial Integro Differential Equations using Modified Differential Tr...
Solving Partial Integro Differential Equations using Modified Differential Tr...Solving Partial Integro Differential Equations using Modified Differential Tr...
Solving Partial Integro Differential Equations using Modified Differential Tr...Associate Professor in VSB Coimbatore
ย 
CS4200 2019 | Lecture 5 | Transformation by Term Rewriting
CS4200 2019 | Lecture 5 | Transformation by Term RewritingCS4200 2019 | Lecture 5 | Transformation by Term Rewriting
CS4200 2019 | Lecture 5 | Transformation by Term RewritingEelco Visser
ย 
re3 - modern regex syntax with a focus on adoption
re3 - modern regex syntax with a focus on adoptionre3 - modern regex syntax with a focus on adoption
re3 - modern regex syntax with a focus on adoptionAur Saraf
ย 
Introduction to programming - class 3
Introduction to programming - class 3Introduction to programming - class 3
Introduction to programming - class 3Paul Brebner
ย 

Similar to Lf 2021 stochastic_calculus_ito-iii (20)

Periodic Solutions for Non-Linear Systems of Integral Equations
Periodic Solutions for Non-Linear Systems of Integral EquationsPeriodic Solutions for Non-Linear Systems of Integral Equations
Periodic Solutions for Non-Linear Systems of Integral Equations
ย 
On solving fuzzy delay differential equationusing bezier curves
On solving fuzzy delay differential equationusing bezier curves  On solving fuzzy delay differential equationusing bezier curves
On solving fuzzy delay differential equationusing bezier curves
ย 
Python for High School Programmers
Python for High School ProgrammersPython for High School Programmers
Python for High School Programmers
ย 
2. data types, variables and operators
2. data types, variables and operators2. data types, variables and operators
2. data types, variables and operators
ย 
Numerical Solution Of Delay Differential Equations Using The Adomian Decompos...
Numerical Solution Of Delay Differential Equations Using The Adomian Decompos...Numerical Solution Of Delay Differential Equations Using The Adomian Decompos...
Numerical Solution Of Delay Differential Equations Using The Adomian Decompos...
ย 
A Class of Polynomials Associated with Differential Operator and with a Gener...
A Class of Polynomials Associated with Differential Operator and with a Gener...A Class of Polynomials Associated with Differential Operator and with a Gener...
A Class of Polynomials Associated with Differential Operator and with a Gener...
ย 
Declare Your Language: Transformation by Strategic Term Rewriting
Declare Your Language: Transformation by Strategic Term RewritingDeclare Your Language: Transformation by Strategic Term Rewriting
Declare Your Language: Transformation by Strategic Term Rewriting
ย 
Periodic Solutions for Nonlinear Systems of Integro-Differential Equations of...
Periodic Solutions for Nonlinear Systems of Integro-Differential Equations of...Periodic Solutions for Nonlinear Systems of Integro-Differential Equations of...
Periodic Solutions for Nonlinear Systems of Integro-Differential Equations of...
ย 
Split and list technique for solving hard problems
Split and list technique for solving hard problemsSplit and list technique for solving hard problems
Split and list technique for solving hard problems
ย 
Engineering Equation Solver (Thai)
Engineering Equation Solver (Thai)Engineering Equation Solver (Thai)
Engineering Equation Solver (Thai)
ย 
Some approximation properties of modified baskakov stancu operators
Some approximation properties of modified baskakov stancu operatorsSome approximation properties of modified baskakov stancu operators
Some approximation properties of modified baskakov stancu operators
ย 
CP2-Chp2-Series.pptx
CP2-Chp2-Series.pptxCP2-Chp2-Series.pptx
CP2-Chp2-Series.pptx
ย 
AJMS_480_23.pdf
AJMS_480_23.pdfAJMS_480_23.pdf
AJMS_480_23.pdf
ย 
Python quickstart for programmers: Python Kung Fu
Python quickstart for programmers: Python Kung FuPython quickstart for programmers: Python Kung Fu
Python quickstart for programmers: Python Kung Fu
ย 
Basic calculus (ii) recap
Basic calculus (ii) recapBasic calculus (ii) recap
Basic calculus (ii) recap
ย 
A05330107
A05330107A05330107
A05330107
ย 
Solving Partial Integro Differential Equations using Modified Differential Tr...
Solving Partial Integro Differential Equations using Modified Differential Tr...Solving Partial Integro Differential Equations using Modified Differential Tr...
Solving Partial Integro Differential Equations using Modified Differential Tr...
ย 
CS4200 2019 | Lecture 5 | Transformation by Term Rewriting
CS4200 2019 | Lecture 5 | Transformation by Term RewritingCS4200 2019 | Lecture 5 | Transformation by Term Rewriting
CS4200 2019 | Lecture 5 | Transformation by Term Rewriting
ย 
re3 - modern regex syntax with a focus on adoption
re3 - modern regex syntax with a focus on adoptionre3 - modern regex syntax with a focus on adoption
re3 - modern regex syntax with a focus on adoption
ย 
Introduction to programming - class 3
Introduction to programming - class 3Introduction to programming - class 3
Introduction to programming - class 3
ย 

More from luc faucheux

Marcel_Faucheux_Taquin.pdf
Marcel_Faucheux_Taquin.pdfMarcel_Faucheux_Taquin.pdf
Marcel_Faucheux_Taquin.pdfluc faucheux
ย 
Selection_Brownian_Particles.pdf
Selection_Brownian_Particles.pdfSelection_Brownian_Particles.pdf
Selection_Brownian_Particles.pdfluc faucheux
ย 
Binary_Potential.pdf
Binary_Potential.pdfBinary_Potential.pdf
Binary_Potential.pdfluc faucheux
ย 
Periodic_Forcing.pdf
Periodic_Forcing.pdfPeriodic_Forcing.pdf
Periodic_Forcing.pdfluc faucheux
ย 
ConfinedBrownian MotionPhysRevE.49.5158.pdf
ConfinedBrownian MotionPhysRevE.49.5158.pdfConfinedBrownian MotionPhysRevE.49.5158.pdf
ConfinedBrownian MotionPhysRevE.49.5158.pdfluc faucheux
ย 
Optical_Thermal_Ratchet.pdf
Optical_Thermal_Ratchet.pdfOptical_Thermal_Ratchet.pdf
Optical_Thermal_Ratchet.pdfluc faucheux
ย 
Lf 2021 risk_management_101
Lf 2021 risk_management_101Lf 2021 risk_management_101
Lf 2021 risk_management_101luc faucheux
ย 
Lf 2021 risk_management_101
Lf 2021 risk_management_101Lf 2021 risk_management_101
Lf 2021 risk_management_101luc faucheux
ย 
Lf 2021 rates_v_b2
Lf 2021 rates_v_b2Lf 2021 rates_v_b2
Lf 2021 rates_v_b2luc faucheux
ย 
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-iLf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-iluc faucheux
ย 
Lf 2020 structured
Lf 2020 structuredLf 2020 structured
Lf 2020 structuredluc faucheux
ย 
Lf 2020 options
Lf 2020 optionsLf 2020 options
Lf 2020 optionsluc faucheux
ย 
Lf 2020 skew
Lf 2020 skewLf 2020 skew
Lf 2020 skewluc faucheux
ย 
Lf 2020 bachelier
Lf 2020 bachelierLf 2020 bachelier
Lf 2020 bachelierluc faucheux
ย 

More from luc faucheux (14)

Marcel_Faucheux_Taquin.pdf
Marcel_Faucheux_Taquin.pdfMarcel_Faucheux_Taquin.pdf
Marcel_Faucheux_Taquin.pdf
ย 
Selection_Brownian_Particles.pdf
Selection_Brownian_Particles.pdfSelection_Brownian_Particles.pdf
Selection_Brownian_Particles.pdf
ย 
Binary_Potential.pdf
Binary_Potential.pdfBinary_Potential.pdf
Binary_Potential.pdf
ย 
Periodic_Forcing.pdf
Periodic_Forcing.pdfPeriodic_Forcing.pdf
Periodic_Forcing.pdf
ย 
ConfinedBrownian MotionPhysRevE.49.5158.pdf
ConfinedBrownian MotionPhysRevE.49.5158.pdfConfinedBrownian MotionPhysRevE.49.5158.pdf
ConfinedBrownian MotionPhysRevE.49.5158.pdf
ย 
Optical_Thermal_Ratchet.pdf
Optical_Thermal_Ratchet.pdfOptical_Thermal_Ratchet.pdf
Optical_Thermal_Ratchet.pdf
ย 
Lf 2021 risk_management_101
Lf 2021 risk_management_101Lf 2021 risk_management_101
Lf 2021 risk_management_101
ย 
Lf 2021 risk_management_101
Lf 2021 risk_management_101Lf 2021 risk_management_101
Lf 2021 risk_management_101
ย 
Lf 2021 rates_v_b2
Lf 2021 rates_v_b2Lf 2021 rates_v_b2
Lf 2021 rates_v_b2
ย 
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-iLf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
ย 
Lf 2020 structured
Lf 2020 structuredLf 2020 structured
Lf 2020 structured
ย 
Lf 2020 options
Lf 2020 optionsLf 2020 options
Lf 2020 options
ย 
Lf 2020 skew
Lf 2020 skewLf 2020 skew
Lf 2020 skew
ย 
Lf 2020 bachelier
Lf 2020 bachelierLf 2020 bachelier
Lf 2020 bachelier
ย 

Recently uploaded

Famous Kala Jadu, Black magic expert in Faisalabad and Kala ilam specialist i...
Famous Kala Jadu, Black magic expert in Faisalabad and Kala ilam specialist i...Famous Kala Jadu, Black magic expert in Faisalabad and Kala ilam specialist i...
Famous Kala Jadu, Black magic expert in Faisalabad and Kala ilam specialist i...batoole333
ย 
Jual obat aborsi Jogja ( 085657271886 ) Cytote pil telat bulan penggugur kand...
Jual obat aborsi Jogja ( 085657271886 ) Cytote pil telat bulan penggugur kand...Jual obat aborsi Jogja ( 085657271886 ) Cytote pil telat bulan penggugur kand...
Jual obat aborsi Jogja ( 085657271886 ) Cytote pil telat bulan penggugur kand...Klinik kandungan
ย 
The Pfandbrief Roundtable 2024 - Covered Bonds
The Pfandbrief Roundtable 2024 - Covered BondsThe Pfandbrief Roundtable 2024 - Covered Bonds
The Pfandbrief Roundtable 2024 - Covered BondsNeil Day
ย 
Production and Cost of the firm with curves
Production and Cost of the firm with curvesProduction and Cost of the firm with curves
Production and Cost of the firm with curvesArifa Saeed
ย 
Webinar on E-Invoicing for Fintech Belgium
Webinar on E-Invoicing for Fintech BelgiumWebinar on E-Invoicing for Fintech Belgium
Webinar on E-Invoicing for Fintech BelgiumFinTech Belgium
ย 
APPLIED ECONOMICS Sept 9FGHFGHFHGFGHFHGFHGFH
APPLIED ECONOMICS Sept 9FGHFGHFHGFGHFHGFHGFHAPPLIED ECONOMICS Sept 9FGHFGHFHGFGHFHGFHGFH
APPLIED ECONOMICS Sept 9FGHFGHFHGFGHFHGFHGFHgeloencina777
ย 
Certified Kala Jadu, Black magic specialist in Rawalpindi and Bangali Amil ba...
Certified Kala Jadu, Black magic specialist in Rawalpindi and Bangali Amil ba...Certified Kala Jadu, Black magic specialist in Rawalpindi and Bangali Amil ba...
Certified Kala Jadu, Black magic specialist in Rawalpindi and Bangali Amil ba...batoole333
ย 
Significant AI Trends for the Financial Industry in 2024 and How to Utilize Them
Significant AI Trends for the Financial Industry in 2024 and How to Utilize ThemSignificant AI Trends for the Financial Industry in 2024 and How to Utilize Them
Significant AI Trends for the Financial Industry in 2024 and How to Utilize Them360factors
ย 
najoomi asli amil baba kala jadu expert rawalpindi bangladesh uk usa
najoomi asli amil baba kala jadu expert rawalpindi bangladesh uk usanajoomi asli amil baba kala jadu expert rawalpindi bangladesh uk usa
najoomi asli amil baba kala jadu expert rawalpindi bangladesh uk usamazhshah570
ย 
uk-no 1 kala ilam expert specialist in uk and qatar kala ilam expert speciali...
uk-no 1 kala ilam expert specialist in uk and qatar kala ilam expert speciali...uk-no 1 kala ilam expert specialist in uk and qatar kala ilam expert speciali...
uk-no 1 kala ilam expert specialist in uk and qatar kala ilam expert speciali...batoole333
ย 
fundamentals of corporate finance 11th canadian edition test bank.docx
fundamentals of corporate finance 11th canadian edition test bank.docxfundamentals of corporate finance 11th canadian edition test bank.docx
fundamentals of corporate finance 11th canadian edition test bank.docxssuserf63bd7
ย 
TriStar Gold- 05-13-2024 corporate presentation
TriStar Gold- 05-13-2024 corporate presentationTriStar Gold- 05-13-2024 corporate presentation
TriStar Gold- 05-13-2024 corporate presentationAdnet Communications
ย 
Famous Kala Jadu, Kala ilam specialist in USA and Bangali Amil baba in Saudi ...
Famous Kala Jadu, Kala ilam specialist in USA and Bangali Amil baba in Saudi ...Famous Kala Jadu, Kala ilam specialist in USA and Bangali Amil baba in Saudi ...
Famous Kala Jadu, Kala ilam specialist in USA and Bangali Amil baba in Saudi ...mazhshah570
ย 
Black magic specialist in Canada (Kala ilam specialist in UK) Bangali Amil ba...
Black magic specialist in Canada (Kala ilam specialist in UK) Bangali Amil ba...Black magic specialist in Canada (Kala ilam specialist in UK) Bangali Amil ba...
Black magic specialist in Canada (Kala ilam specialist in UK) Bangali Amil ba...batoole333
ย 
No 1 Top Love marriage specialist baba ji amil baba kala ilam powerful vashik...
No 1 Top Love marriage specialist baba ji amil baba kala ilam powerful vashik...No 1 Top Love marriage specialist baba ji amil baba kala ilam powerful vashik...
No 1 Top Love marriage specialist baba ji amil baba kala ilam powerful vashik...batoole333
ย 
amil baba in australia amil baba in canada amil baba in london amil baba in g...
amil baba in australia amil baba in canada amil baba in london amil baba in g...amil baba in australia amil baba in canada amil baba in london amil baba in g...
amil baba in australia amil baba in canada amil baba in london amil baba in g...israjan914
ย 
Test bank for advanced assessment interpreting findings and formulating diffe...
Test bank for advanced assessment interpreting findings and formulating diffe...Test bank for advanced assessment interpreting findings and formulating diffe...
Test bank for advanced assessment interpreting findings and formulating diffe...robinsonayot
ย 
Certified Kala Jadu, Black magic specialist in Rawalpindi and Bangali Amil ba...
Certified Kala Jadu, Black magic specialist in Rawalpindi and Bangali Amil ba...Certified Kala Jadu, Black magic specialist in Rawalpindi and Bangali Amil ba...
Certified Kala Jadu, Black magic specialist in Rawalpindi and Bangali Amil ba...batoole333
ย 
Q1 2024 Conference Call Presentation vF.pdf
Q1 2024 Conference Call Presentation vF.pdfQ1 2024 Conference Call Presentation vF.pdf
Q1 2024 Conference Call Presentation vF.pdfAdnet Communications
ย 

Recently uploaded (20)

Famous Kala Jadu, Black magic expert in Faisalabad and Kala ilam specialist i...
Famous Kala Jadu, Black magic expert in Faisalabad and Kala ilam specialist i...Famous Kala Jadu, Black magic expert in Faisalabad and Kala ilam specialist i...
Famous Kala Jadu, Black magic expert in Faisalabad and Kala ilam specialist i...
ย 
Jual obat aborsi Jogja ( 085657271886 ) Cytote pil telat bulan penggugur kand...
Jual obat aborsi Jogja ( 085657271886 ) Cytote pil telat bulan penggugur kand...Jual obat aborsi Jogja ( 085657271886 ) Cytote pil telat bulan penggugur kand...
Jual obat aborsi Jogja ( 085657271886 ) Cytote pil telat bulan penggugur kand...
ย 
The Pfandbrief Roundtable 2024 - Covered Bonds
The Pfandbrief Roundtable 2024 - Covered BondsThe Pfandbrief Roundtable 2024 - Covered Bonds
The Pfandbrief Roundtable 2024 - Covered Bonds
ย 
Production and Cost of the firm with curves
Production and Cost of the firm with curvesProduction and Cost of the firm with curves
Production and Cost of the firm with curves
ย 
Webinar on E-Invoicing for Fintech Belgium
Webinar on E-Invoicing for Fintech BelgiumWebinar on E-Invoicing for Fintech Belgium
Webinar on E-Invoicing for Fintech Belgium
ย 
APPLIED ECONOMICS Sept 9FGHFGHFHGFGHFHGFHGFH
APPLIED ECONOMICS Sept 9FGHFGHFHGFGHFHGFHGFHAPPLIED ECONOMICS Sept 9FGHFGHFHGFGHFHGFHGFH
APPLIED ECONOMICS Sept 9FGHFGHFHGFGHFHGFHGFH
ย 
DIGITAL COMMERCE SHAPE VIETNAMESE SHOPPING HABIT IN 4.0 INDUSTRY
DIGITAL COMMERCE SHAPE VIETNAMESE SHOPPING HABIT IN 4.0 INDUSTRYDIGITAL COMMERCE SHAPE VIETNAMESE SHOPPING HABIT IN 4.0 INDUSTRY
DIGITAL COMMERCE SHAPE VIETNAMESE SHOPPING HABIT IN 4.0 INDUSTRY
ย 
Certified Kala Jadu, Black magic specialist in Rawalpindi and Bangali Amil ba...
Certified Kala Jadu, Black magic specialist in Rawalpindi and Bangali Amil ba...Certified Kala Jadu, Black magic specialist in Rawalpindi and Bangali Amil ba...
Certified Kala Jadu, Black magic specialist in Rawalpindi and Bangali Amil ba...
ย 
Significant AI Trends for the Financial Industry in 2024 and How to Utilize Them
Significant AI Trends for the Financial Industry in 2024 and How to Utilize ThemSignificant AI Trends for the Financial Industry in 2024 and How to Utilize Them
Significant AI Trends for the Financial Industry in 2024 and How to Utilize Them
ย 
najoomi asli amil baba kala jadu expert rawalpindi bangladesh uk usa
najoomi asli amil baba kala jadu expert rawalpindi bangladesh uk usanajoomi asli amil baba kala jadu expert rawalpindi bangladesh uk usa
najoomi asli amil baba kala jadu expert rawalpindi bangladesh uk usa
ย 
uk-no 1 kala ilam expert specialist in uk and qatar kala ilam expert speciali...
uk-no 1 kala ilam expert specialist in uk and qatar kala ilam expert speciali...uk-no 1 kala ilam expert specialist in uk and qatar kala ilam expert speciali...
uk-no 1 kala ilam expert specialist in uk and qatar kala ilam expert speciali...
ย 
fundamentals of corporate finance 11th canadian edition test bank.docx
fundamentals of corporate finance 11th canadian edition test bank.docxfundamentals of corporate finance 11th canadian edition test bank.docx
fundamentals of corporate finance 11th canadian edition test bank.docx
ย 
TriStar Gold- 05-13-2024 corporate presentation
TriStar Gold- 05-13-2024 corporate presentationTriStar Gold- 05-13-2024 corporate presentation
TriStar Gold- 05-13-2024 corporate presentation
ย 
Famous Kala Jadu, Kala ilam specialist in USA and Bangali Amil baba in Saudi ...
Famous Kala Jadu, Kala ilam specialist in USA and Bangali Amil baba in Saudi ...Famous Kala Jadu, Kala ilam specialist in USA and Bangali Amil baba in Saudi ...
Famous Kala Jadu, Kala ilam specialist in USA and Bangali Amil baba in Saudi ...
ย 
Black magic specialist in Canada (Kala ilam specialist in UK) Bangali Amil ba...
Black magic specialist in Canada (Kala ilam specialist in UK) Bangali Amil ba...Black magic specialist in Canada (Kala ilam specialist in UK) Bangali Amil ba...
Black magic specialist in Canada (Kala ilam specialist in UK) Bangali Amil ba...
ย 
No 1 Top Love marriage specialist baba ji amil baba kala ilam powerful vashik...
No 1 Top Love marriage specialist baba ji amil baba kala ilam powerful vashik...No 1 Top Love marriage specialist baba ji amil baba kala ilam powerful vashik...
No 1 Top Love marriage specialist baba ji amil baba kala ilam powerful vashik...
ย 
amil baba in australia amil baba in canada amil baba in london amil baba in g...
amil baba in australia amil baba in canada amil baba in london amil baba in g...amil baba in australia amil baba in canada amil baba in london amil baba in g...
amil baba in australia amil baba in canada amil baba in london amil baba in g...
ย 
Test bank for advanced assessment interpreting findings and formulating diffe...
Test bank for advanced assessment interpreting findings and formulating diffe...Test bank for advanced assessment interpreting findings and formulating diffe...
Test bank for advanced assessment interpreting findings and formulating diffe...
ย 
Certified Kala Jadu, Black magic specialist in Rawalpindi and Bangali Amil ba...
Certified Kala Jadu, Black magic specialist in Rawalpindi and Bangali Amil ba...Certified Kala Jadu, Black magic specialist in Rawalpindi and Bangali Amil ba...
Certified Kala Jadu, Black magic specialist in Rawalpindi and Bangali Amil ba...
ย 
Q1 2024 Conference Call Presentation vF.pdf
Q1 2024 Conference Call Presentation vF.pdfQ1 2024 Conference Call Presentation vF.pdf
Q1 2024 Conference Call Presentation vF.pdf
ย 

Lf 2021 stochastic_calculus_ito-iii

  • 1. Luc_Faucheux_2021 Stochastic Calculus โ€“ ITO โ€“ III From SDEs to PDEs and back โ€“ part trois Non-Linear SDE 1
  • 2. Luc_Faucheux_2021 In this section ยจ We finally look at the non-linear SDE ยจ We will introduce the [๐›ผ] calculus in order to not get confused with the factor (1/2) ยจ We will look also at the backward and forward formalism for PDE ยจ Maybe if we have time some examples (OU canonical process) ยจ SDE: Stochastic Differential Equations ยจ SIE: Stochastic Integral Equations ยจ PDE: Partial Differential Equations ยจ PDF: Probability Distribution Function 2
  • 3. Luc_Faucheux_2021 In this deck - II ยจ More importantly we will discuss at length (and maybe ad nauseam for some of you) the so called โ€œITO-STRATANOVITCH controversyโ€ ยจ That was something that for some reason was quite popular in the 90s ยจ Like a lot of things that were popular in the 90s, there was maybe not a good rationale for it, and that kind of faded away ยจ So we will use it more as an example on how to use, and make sure that we understand stochastic calculus 3
  • 4. Luc_Faucheux_2021 In this deck - III ยจ If you Google โ€œITO-STRATANOVITCH controversyโ€, you can see that it is still creating a lot of confusion out there, especially in the fields of Physics for non-homogeneous diffusion coefficients (in the field of finance that would be for a volatility that is a function of the stochastic underlier like equity for example) ยจ So that would be for modeling an SDE that would look something like that: ยจ dX(t)= a(X(t),t).dt+b(X(t),t).dW(t) ยจ We will show, that unlike all the previous examples in the deck II, writing something like the above is actually not well defined. ยจ That is where the confusion comes from ยจ Interestingly enough it also comes from the fact that people started using digital computers to simulate diffusion process, and in a nutshell (running the risk of oversimplifying): ยจ DIGITAL COMPUTERS LIVE IN AN ITO WORLD ยจ ANALOG COMPUTERS LIVE IN A STRATANOVITCH WORLD 4
  • 5. Luc_Faucheux_2021 In this deck - IV ยจ To quote Manella: ยจ โ€ The Itรด versus Stratonovich controversy, about the "correct" calculus to use for integration of Langevin equations, was settled to general satisfaction some 30 years ago. Recently, however, it has started to re-emerge, following the advent of new experimental techniques.โ€ ยจ That was written in 2012 or so. ยจ But a lot of people are still confused by it. ยจ I get also confused all the time, and then I have a panic attack, then I take a couple of deep breathes, and redo a bunch of pages of derivations to make sure that I still understand it. ยจ The older I get the more frequent that happens. 5
  • 7. Luc_Faucheux_2021 dX= a(X(t),t).dt+b(X(t),t).dW โ€œCent deux fois sur le mรฉtier remettez votre ouvrageโ€ Nicolas Boileau 7
  • 8. Luc_Faucheux_2021 Finally ยจ We are almost there, so just before we tackle that one, because the factor (1/2) are confusing (is that coming from the middle point in STRATO, or is is coming from the second term in a Taylor expansion?), we generalize a little the formalism 8
  • 10. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus ยจ Here is where we take a pause to avoid being confused with factors 2 ยจ Ito when defining the Ito integral of a function took the left-side ยจ Strato took the middle point ยจ Hence the existence of a factor (1/2) in the conversion between Ito and Strato ยจ HOWEVER, now that we are dealing with PDE, there is another factor (1/2) that crops up a lot (from the integration by part of the second moment ๐‘ฅ!) ยจ Those 2 factors are NOT related (although in way they are, Strato chose middle point specifically so that the rules of usual calculus would be formally conserved, so essentially he solved for which point to use to offset the (1/2) term in Ito lemma) ยจ So they will be related, but in the mean-time we could get confused between the two ยจ Some authors (Gleeson, Arovas) have pointed out that it is somewhat easier to carry derivations โ€in the [๐›ผ] calculusโ€, and then setting the value of ๐›ผ, as opposed to duplicating everything in Ito then Strato and carrying (1/2) factors 10
  • 11. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus - II ยจ The ITO integral is defined as: ยจ โˆซ "#"$ "#"% ๐‘“ ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘‹(๐‘ก) = lim &โ†’( {โˆ‘)#* )#& ๐‘“(๐‘‹(๐‘ก))). [๐‘‹(๐‘ก)+*) โˆ’ ๐‘‹(๐‘ก))]} ยจ The Stratonovitch integral is defined as: ยจ โˆซ "#"$ "#"% ๐‘“ ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘‹(๐‘ก) = lim &โ†’( {โˆ‘)#* )#& ๐‘“ [๐‘‹(๐‘ก) + ๐‘‹(๐‘ก)+*)]/2). [๐‘‹(๐‘ก)+*) โˆ’ ๐‘‹(๐‘ก))]} ยจ We can define the [๐›ผ] integral as: ยจ โˆซ "#"$ "#"% ๐‘“ ๐‘‹ ๐‘ก . ([๐›ผ]). ๐‘‘๐‘‹(๐‘ก) = lim &โ†’( {โˆ‘)#* )#& ๐‘“(๐‘‹(๐‘ก)) + ๐›ผ. [๐‘‹(๐‘ก)+*) โˆ’ ๐‘‹(๐‘ก))]). [๐‘‹(๐‘ก)+*) โˆ’ ๐‘‹(๐‘ก))]} ยจ ITO will be the case ๐›ผ = 0 ยจ STRATO will be the case ๐›ผ = 1/2 11
  • 12. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus - III ยจ We had the relation on the integrals: ยจ โˆซ "#"$ "#"% ๐‘“ ๐‘Š ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก) = โˆซ "#"$ "#"% ๐‘“ ๐‘Š ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) + * ! โˆซ "#"$ "#"% ๐‘“โ€ฒ ๐‘Š ๐‘ก . ๐‘‘๐‘ก ยจ This becomes: ยจ โˆซ "#"$ "#"% ๐‘“ ๐‘Š ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š(๐‘ก) = โˆซ "#"$ "#"% ๐‘“ ๐‘Š ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) + ๐›ผ. โˆซ "#"$ "#"% ๐‘“โ€ฒ ๐‘Š ๐‘ก . ๐‘‘๐‘ก ยจ Or: ยจ โˆซ "#"$ "#"% ๐‘“ ๐‘Š ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š(๐‘ก) = โˆซ "#"$ "#"% ๐‘“ ๐‘Š ๐‘ก . ([๐›ผ] = 0). ๐‘‘๐‘Š(๐‘ก) + ๐›ผ. โˆซ "#"$ "#"% ๐‘“โ€ฒ ๐‘Š ๐‘ก . ๐‘‘๐‘ก 12
  • 13. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus - IV ยจ For a more complicated stochastic process ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘Š ยจ We have: ยจ โˆซ "#"$ "#"% ๐‘“ ๐‘‹ ๐‘ก . โˆ˜ . ๐‘‘๐‘Š ๐‘ก = โˆซ "#"$ "#"% ๐‘“ ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) + โˆซ "#"$ "#"% * ! . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- ๐‘“ ๐‘‹(๐‘ก . ๐‘‘๐‘ก ยจ This now becomes: ยจ โˆซ "#"$ "#"% ๐‘“ ๐‘‹ ๐‘ก . [๐›ผ] . ๐‘‘๐‘Š ๐‘ก = โˆซ "#"$ "#"% ๐‘“ ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) + โˆซ "#"$ "#"% ๐›ผ. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- ๐‘“ ๐‘‹(๐‘ก . ๐‘‘๐‘ก 13
  • 14. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus - V ยจ For the SDE we had the following mapping between ITO and STRATO ยจ The ITO SDE: ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š ยจ Has the same solution (is the same) as the STRATO SDE in STRATO calculus: ยจ ๐‘‘๐‘‹ ๐‘ก = [๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก โˆ’ * ! . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ]. ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š ยจ The STRATO SDE ยจ ๐‘‘๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š ยจ Has the same solution (is the same) as the ITO SDE in ITO calculus ยจ ๐‘‘๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + * ! . A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š 14
  • 15. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus - VI ยจ This now becomes: ยจ The ITO SDE: ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š ยจ Has the same solution (is the same) as the [๐›ผ] SDE in [๐›ผ] calculus: ยจ ๐‘‘๐‘‹ ๐‘ก = [๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก โˆ’ ๐›ผ. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ]. ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š ยจ The [๐›ผ] SDE ยจ ๐‘‘๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š ยจ Has the same solution (is the same) as the ITO SDE in ITO calculus ยจ ๐‘‘๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + ๐›ผ. A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š 15
  • 16. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus - VII ยจ The ITO lemma (chain rule) reads: ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ "#"$ "#"% ,. ,- . ([). ๐‘‘๐‘‹(๐‘ก) + * ! โˆซ "#"$ "#"% ,!/ ,0! (๐‘‹ ๐‘ก ). ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ! ๐‘‘๐‘ก ยจ In the โ€limitโ€ of small time increments, this can be written formally as the Ito lemma: ยจ ๐›ฟ๐‘“ = ,. ,- . ๐›ฟ๐‘‹ + * ! . ,!/ ,0! . ๐‘!๐›ฟ๐‘ก ยจ The STRATO lemma reads: ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ "#"$ "#"% ,. ,- . (โˆ˜). ๐‘‘๐‘‹(๐‘ก) ยจ In the โ€limitโ€ of small time increments, this can be written formally as the Strato lemma: ยจ ๐›ฟ๐‘“ = ,. ,- . โˆ˜ . ๐›ฟ๐‘‹ 16
  • 17. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus - VIII ยจ In the [๐›ผ] calculus the [๐›ผ] lemma (chain rule) now reads : ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ "#"$ "#"% ,. ,- . ([๐›ผ]). ๐‘‘๐‘‹(๐‘ก) + * ! โˆ’ ๐›ผ . โˆซ "#"$ "#"% ,!/ ,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ! . ๐‘‘๐‘ก ยจ In the โ€limitโ€ of small time increments, this can be written formally as the [๐›ผ] lemma: ยจ ๐›ฟ๐‘“ = ,. ,- . ๐›ฟ๐‘‹ + * ! โˆ’ ๐›ผ . ,!/ ,0! . ๐‘!. ๐›ฟ๐‘ก ยจ NOTE: you can convince yourselves by redoing the derivation we had on pages 55-60 ยจ This actually highlights why STRATO took the middle point ๐›ผ = 1/2 , as this is the point that cancels out the (1/2) coming from the Taylor expansion of ๐‘“ ๐‘‹ ๐‘ก% from the left point. ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = lim &โ†’( โˆ‘)#* )#& {๐‘“(๐‘‹(๐‘ก))) โˆ’ ๐‘“(๐‘‹(๐‘ก)1*))} ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = lim &โ†’( โˆ‘)#* )#& { ,. ,- . ([). ๐›ฟ๐‘‹ + * ! . ,!/ ,0! . ([). (๐›ฟ๐‘‹)!} 17
  • 18. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus - IX ยจ This is kind of nice, because we only have to do one derivation, and we do not get confused where the factor (1/2) comes from in STRATO, and whether or not it is the same factor (1/2) that comes from the integration by part of the second moment. ยจ Factors (1/2) are very confusing, so whenever possible it is better to keep them as variable 18
  • 19. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus - X ยจ As an illustrated example of [๐›ผ] calculus, we calculate ยจ ๐ด[๐›ผ] = ๐”ผ{โˆซ "#"$ "#"% ๐‘Š(๐‘ก). ([๐›ผ]). ๐‘‘๐‘Š(๐‘ก)} and with ๐‘“ ๐‘Š ๐‘ก = ๐‘Š(๐‘ก) ยจ ๐ด[๐›ผ] = ๐”ผ{ lim &โ†’( โˆ‘)#* )#& ๐‘“(๐‘Š(๐‘ก)) + ๐›ผ. [๐‘Š(๐‘ก)+*) โˆ’ ๐‘Š(๐‘ก))]). [๐‘Š(๐‘ก)+*) โˆ’ ๐‘Š(๐‘ก))]} ยจ ๐ด[๐›ผ] = ๐”ผ{ lim &โ†’( โˆ‘)#* )#& (๐‘Š(๐‘ก)) + ๐›ผ. [๐‘Š(๐‘ก)+*) โˆ’ ๐‘Š(๐‘ก))]). [๐‘Š(๐‘ก)+*) โˆ’ ๐‘Š(๐‘ก))]} ยจ ๐”ผ{๐‘Š ๐‘ก . ๐‘Š(๐‘ก2)} = min(๐‘ก, ๐‘ก2) ยจ ๐”ผ ๐‘Š ๐‘ก . ๐‘Š ๐‘ก2 โˆ’ ๐‘Š ๐‘ก22 = min ๐‘ก, ๐‘ก2 โˆ’ min ๐‘ก, ๐‘ก22 = 0 when ๐‘ก < ๐‘ก2 < ๐‘กโ€ฒโ€ฒ ยจ ๐”ผ ๐‘Š ๐‘ก โˆ’ ๐‘Š ๐‘ก2 ! = ๐‘ก + ๐‘ก2 โˆ’ 2. min ๐‘ก, ๐‘ก2 = |๐‘ก โˆ’ ๐‘กโ€ฒ| ยจ ๐ด[๐›ผ] = ๐”ผ{ lim &โ†’( โˆ‘)#* )#& ๐›ผ. [๐‘ก)+* โˆ’ ๐‘ก)]} = ๐›ผ. [๐‘ก% โˆ’ ๐‘ก$] ยจ Remember if you could integrate X? Worth revisiting within the [๐›ผ] calculus 19
  • 20. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus - XI ยจ We have the following conventions: ยจ ๐›ผ = 0 ITO ยจ โˆซ "#"$ "#"% ๐‘“ ๐‘‹ ๐‘ก . [๐›ผ] . ๐‘‘๐‘Š ๐‘ก = โˆซ "#"$ "#"% ๐‘“ ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) + โˆซ "#"$ "#"% ๐›ผ. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- ๐‘“ ๐‘‹(๐‘ก . ๐‘‘๐‘ก ยจ โˆซ "#"$ "#"% ๐‘“ ๐‘‹ ๐‘ก . [0] . ๐‘‘๐‘Š ๐‘ก = โˆซ "#"$ "#"% ๐‘“ ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ยจ ITO integral is a MARTINGALE ยจ ITO integral exhibits the ISOMETRY property ยจ ITO LEMMA ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ "#"$ "#"% ,. ,- . ([๐›ผ]). ๐‘‘๐‘‹(๐‘ก) + * ! โˆ’ ๐›ผ . โˆซ "#"$ "#"% ,!/ ,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ! . ๐‘‘๐‘ก ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ "#"$ "#"% ,. ,- . ([). ๐‘‘๐‘‹(๐‘ก) + * ! . โˆซ "#"$ "#"% ,!/ ,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ! . ๐‘‘๐‘ก 20
  • 21. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus - XII ยจ ๐›ผ = 1/2 STRATONOVITCH ยจ โˆซ "#"$ "#"% ๐‘“ ๐‘‹ ๐‘ก . [๐›ผ] . ๐‘‘๐‘Š ๐‘ก = โˆซ "#"$ "#"% ๐‘“ ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) + โˆซ "#"$ "#"% ๐›ผ. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- ๐‘“ ๐‘‹(๐‘ก . ๐‘‘๐‘ก ยจ โˆซ "#"$ "#"% ๐‘“ ๐‘‹ ๐‘ก . [1/2] . ๐‘‘๐‘Š ๐‘ก = โˆซ "#"$ "#"% ๐‘“ ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก) ยจ STRATO integral is NOT a MARTINGALE ยจ STRATO integral does NOT exhibits the ISOMETRY property ยจ STRATO LEMMA, usual regular Newtonian Leibniz calculus chain rule is FORMALLY respected ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ "#"$ "#"% ,. ,- . ([๐›ผ]). ๐‘‘๐‘‹(๐‘ก) + * ! โˆ’ ๐›ผ . โˆซ "#"$ "#"% ,!/ ,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ! . ๐‘‘๐‘ก ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ "#"$ "#"% ,. ,- . (โˆ˜). ๐‘‘๐‘‹(๐‘ก) 21
  • 22. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus - XIII ยจ ๐›ผ = 1 KLIMONTOVITCH (reverse ITO, or OTI) ยจ โˆซ "#"$ "#"% ๐‘“ ๐‘‹ ๐‘ก . [๐›ผ] . ๐‘‘๐‘Š ๐‘ก = โˆซ "#"$ "#"% ๐‘“ ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) + โˆซ "#"$ "#"% ๐›ผ. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- ๐‘“ ๐‘‹(๐‘ก . ๐‘‘๐‘ก ยจ โˆซ "#"$ "#"% ๐‘“ ๐‘‹ ๐‘ก . [1] . ๐‘‘๐‘Š ๐‘ก = โˆซ "#"$ "#"% ๐‘“ ๐‘‹ ๐‘ก . (]). ๐‘‘๐‘Š(๐‘ก) ยจ KLIMONTOVITCH integral is NOT a MARTINGALE ยจ KLIMONTOVITCH integral does NOT exhibits the ISOMETRY property ยจ KLIMONTOVITCH LEMMA, does NOT formally looks like the regular Leibniz rule ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ "#"$ "#"% ,. ,- . ([๐›ผ]). ๐‘‘๐‘‹(๐‘ก) + * ! โˆ’ ๐›ผ . โˆซ "#"$ "#"% ,!/ ,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ! . ๐‘‘๐‘ก ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ "#"$ "#"% ,. ,- . ] . ๐‘‘๐‘‹ ๐‘ก โˆ’ * ! . โˆซ "#"$ "#"% ,!/ ,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ! . ๐‘‘๐‘ก 22
  • 23. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus โ€“ XIII a ยจ No idea what the advantages of the KLIMONTOVITCH calculus since, the integral is not a martingale like ITO, the usual rules of calculus are not respected like STRATO ยจ But it is in the literature out there so thought I would include it ยจ Might be useful when you do backward numerical simulations on a digital computer, but not sure, will look into this further 23
  • 24. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus - XIV ยจ ๐ด[๐›ผ] = ๐”ผ{โˆซ "#"$ "#"% ๐‘Š(๐‘ก). ([๐›ผ]). ๐‘‘๐‘Š(๐‘ก)} and with ๐‘“ ๐‘Š ๐‘ก = ๐‘Š(๐‘ก) ยจ ๐ด[๐›ผ] = ๐”ผ{ lim &โ†’( โˆ‘)#* )#& ๐›ผ. [๐‘ก)+* โˆ’ ๐‘ก)]} = ๐›ผ. [๐‘ก% โˆ’ ๐‘ก$] ยจ ITO ๐›ผ = 0 ยจ ๐”ผ{โˆซ "#"$ "#"% ๐‘Š(๐‘ก). ([). ๐‘‘๐‘Š(๐‘ก)} = 0 ยจ No surprise there as the ITO integral is a MARTINGALE ยจ STRATO ๐›ผ = 1/2 ยจ ๐”ผ{โˆซ "#"$ "#"% ๐‘Š(๐‘ก). (โˆ˜). ๐‘‘๐‘Š(๐‘ก)} = (1/2). [๐‘ก% โˆ’ ๐‘ก$] ยจ KLIMONTOVITCH ๐›ผ = 1 ยจ ๐”ผ{โˆซ "#"$ "#"% ๐‘Š(๐‘ก). (]). ๐‘‘๐‘Š(๐‘ก)} = [๐‘ก% โˆ’ ๐‘ก$] 24
  • 25. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus - XV ยจ Letโ€™s look at the integration before taking the expectation: ยจ โˆซ "#"$ "#"% ๐‘Š(๐‘ก). (โˆ˜). ๐‘‘๐‘Š(๐‘ก) = * ! (๐‘Š ๐‘ก% ! โˆ’ ๐‘Š ๐‘ก$ !) ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ "#"$ "#"% ,. ,- . ([). ๐‘‘๐‘‹(๐‘ก) + * ! . โˆซ "#"$ "#"% ,!. ,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ! . ๐‘‘๐‘ก ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ "#"$ "#"% ,. ,- . โˆ˜ . ๐‘‘๐‘‹ ๐‘ก ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ "#"$ "#"% ,. ,- . ] . ๐‘‘๐‘‹ ๐‘ก โˆ’ * ! . โˆซ "#"$ "#"% ,!. ,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ! . ๐‘‘๐‘ก ยจ โˆซ "#"$ "#"% ,. ,- . โˆ˜ . ๐‘‘๐‘‹ ๐‘ก = โˆซ "#"$ "#"% ,. ,- . ([). ๐‘‘๐‘‹(๐‘ก) + * ! . โˆซ "#"$ "#"% ,!. ,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ! . ๐‘‘๐‘ก ยจ โˆซ "#"$ "#"% ,. ,- . ] . ๐‘‘๐‘‹ ๐‘ก = โˆซ "#"$ "#"% ,. ,- . ([). ๐‘‘๐‘‹(๐‘ก) + โˆซ "#"$ "#"% ,!. ,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ! . ๐‘‘๐‘ก 25
  • 26. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus - XVI ยจ โˆซ "#"$ "#"% ,. ,- . โˆ˜ . ๐‘‘๐‘‹ ๐‘ก = โˆซ "#"$ "#"% ,. ,- . ([). ๐‘‘๐‘‹(๐‘ก) + * ! . โˆซ "#"$ "#"% ,!. ,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ! . ๐‘‘๐‘ก ยจ โˆซ "#"$ "#"% ,. ,- . ] . ๐‘‘๐‘‹ ๐‘ก = โˆซ "#"$ "#"% ,. ,- . ([). ๐‘‘๐‘‹(๐‘ก) + โˆซ "#"$ "#"% ,!. ,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ! . ๐‘‘๐‘ก ยจ โˆซ "#"$ "#"% ,. ,3 . โˆ˜ . ๐‘‘๐‘Š ๐‘ก = โˆซ "#"$ "#"% ,. ,3 . ([). ๐‘‘๐‘Š(๐‘ก) + * ! . โˆซ "#"$ "#"% ,!. ,3! ๐‘ค ๐‘ก . ๐‘ ๐‘ก, ๐‘Š ๐‘ก ! . ๐‘‘๐‘ก ยจ โˆซ "#"$ "#"% ,. ,3 . ] . ๐‘‘๐‘Š ๐‘ก = โˆซ "#"$ "#"% ,. ,3 . ([). ๐‘‘๐‘Š(๐‘ก) + โˆซ "#"$ "#"% ,!. ,3! ๐‘ค ๐‘ก . ๐‘ ๐‘ก, ๐‘Š ๐‘ก ! . ๐‘‘๐‘ก ยจ โˆซ "#"$ "#"% ๐‘Š(๐‘ก). (โˆ˜). ๐‘‘๐‘Š(๐‘ก) = * ! (๐‘Š ๐‘ก% ! โˆ’ ๐‘Š ๐‘ก$ !) ยจ With ๐‘“ ๐‘Š ๐‘ก = ๐‘Š(๐‘ก) we have ,. ,4 ๐‘Š ๐‘ก = ๐‘Š(๐‘ก) ยจ ,. ,3 ๐‘ค = ๐‘ค ยจ ,!. ,3! = 1 26
  • 27. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus - XVII ยจ โˆซ "#"$ "#"% ๐‘Š(๐‘ก). (โˆ˜). ๐‘‘๐‘Š(๐‘ก) = * ! (๐‘Š ๐‘ก% ! โˆ’ ๐‘Š ๐‘ก$ !) ยจ โˆซ "#"$ "#"% ,. ,- . โˆ˜ . ๐‘‘๐‘‹ ๐‘ก = โˆซ "#"$ "#"% ,. ,- . ([). ๐‘‘๐‘‹(๐‘ก) + * ! . โˆซ "#"$ "#"% ,!. ,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ! . ๐‘‘๐‘ก ยจ โˆซ "#"$ "#"% ,. ,3 . โˆ˜ . ๐‘‘๐‘Š ๐‘ก = โˆซ "#"$ "#"% ,. ,3 . ([). ๐‘‘๐‘Š(๐‘ก) + * ! . โˆซ "#"$ "#"% ,!. ,3! ๐‘ค ๐‘ก . ๐‘ ๐‘ก, ๐‘Š ๐‘ก ! . ๐‘‘๐‘ก ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘‘๐‘Š(๐‘ก) ยจ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก = 0 ยจ ๐‘ ๐‘ก, ๐‘‹ ๐‘ก = 0 ยจ ,. ,3 ๐‘ค = ๐‘ค and ,!. ,3! = 1 27
  • 28. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus - XVIII ยจ โˆซ "#"$ "#"% ,. ,3 . โˆ˜ . ๐‘‘๐‘Š ๐‘ก = โˆซ "#"$ "#"% ,. ,3 . ([). ๐‘‘๐‘Š(๐‘ก) + * ! . โˆซ "#"$ "#"% ,!. ,3! ๐‘ค ๐‘ก . ๐‘ ๐‘ก, ๐‘Š ๐‘ก ! . ๐‘‘๐‘ก ยจ โˆซ "#"$ "#"% ,. ,3 . โˆ˜ . ๐‘‘๐‘Š ๐‘ก = โˆซ "#"$ "#"% ,. ,3 . ([). ๐‘‘๐‘Š(๐‘ก) + * ! . โˆซ "#"$ "#"% ,!. ,3! ๐‘ค ๐‘ก . ๐‘‘๐‘ก ยจ โˆซ "#"$ "#"% ,. ,3 . ([). ๐‘‘๐‘Š(๐‘ก) = โˆซ "#"$ "#"% ,. ,3 . โˆ˜ . ๐‘‘๐‘Š ๐‘ก โˆ’ * ! . โˆซ "#"$ "#"% ,!. ,3! ๐‘ค ๐‘ก . ๐‘‘๐‘ก ยจ ,. ,3 ๐‘ค = ๐‘ค and ,!. ,3! = 1 ยจ โˆซ "#"$ "#"% ๐‘Š(๐‘ก). (โˆ˜). ๐‘‘๐‘Š(๐‘ก) = * ! (๐‘Š ๐‘ก% ! โˆ’ ๐‘Š ๐‘ก$ !) ยจ โˆซ "#"$ "#"% ๐‘Š(๐‘ก). ([). ๐‘‘๐‘Š(๐‘ก) = โˆซ "#"$ "#"% ๐‘Š(๐‘ก). โˆ˜ . ๐‘‘๐‘Š ๐‘ก โˆ’ * ! . โˆซ "#"$ "#"% 1. ๐‘‘๐‘ก ยจ โˆซ "#"$ "#"% ๐‘Š(๐‘ก). ([). ๐‘‘๐‘Š(๐‘ก) = * ! (๐‘Š ๐‘ก% ! โˆ’ ๐‘Š ๐‘ก$ !) โˆ’ * ! . (๐‘ก% โˆ’ ๐‘ก$) 28
  • 29. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus - XIX ยจ โˆซ "#"$ "#"% ๐‘Š(๐‘ก). (โˆ˜). ๐‘‘๐‘Š(๐‘ก) = * ! (๐‘Š ๐‘ก% ! โˆ’ ๐‘Š ๐‘ก$ !) ยจ โˆซ "#"$ "#"% ๐‘Š(๐‘ก). ([). ๐‘‘๐‘Š(๐‘ก) = * ! (๐‘Š ๐‘ก% ! โˆ’ ๐‘Š ๐‘ก$ !) โˆ’ * ! . (๐‘ก% โˆ’ ๐‘ก$) ยจ โˆซ "#"$ "#"% ,. ,3 . ] . ๐‘‘๐‘Š ๐‘ก = โˆซ "#"$ "#"% ,. ,3 . ([). ๐‘‘๐‘Š(๐‘ก) + โˆซ "#"$ "#"% ,!. ,3! ๐‘ค ๐‘ก . ๐‘ ๐‘ก, ๐‘Š ๐‘ก ! . ๐‘‘๐‘ก ยจ โˆซ "#"$ "#"% ,. ,3 . ] . ๐‘‘๐‘Š ๐‘ก = โˆซ "#"$ "#"% ,. ,3 . (โˆ˜). ๐‘‘๐‘Š(๐‘ก) + * ! . โˆซ "#"$ "#"% ,!. ,3! ๐‘ค ๐‘ก . ๐‘ ๐‘ก, ๐‘Š ๐‘ก ! . ๐‘‘๐‘ก ยจ โˆซ "#"$ "#"% ๐‘Š(๐‘ก). ] . ๐‘‘๐‘Š ๐‘ก = * ! ๐‘Š ๐‘ก% ! โˆ’ ๐‘Š ๐‘ก$ ! + * ! . (๐‘ก% โˆ’ ๐‘ก$) 29
  • 30. Luc_Faucheux_2021 Introducing the [๐›ผ] calculus - XX ยจ Making it a little simpler and easier on the eye with: ยจ ๐‘ก$ = 0 and ๐‘ก% = ๐‘ก ยจ ๐‘Š ๐‘ก$ = 0 and ๐‘Š ๐‘ก% = ๐‘Š ๐‘ก = ๐‘Š ยจ ITO: โˆซ 5 " ๐‘Š. ([). ๐‘‘๐‘Š = 4 " ! ! โˆ’ " ! ๐”ผ โˆซ 5 " ๐‘Š. ([). ๐‘‘๐‘Š = 0 ยจ STRATONOVITCH: โˆซ 5 " ๐‘Š. (โˆ˜). ๐‘‘๐‘Š = 4 " ! ! ๐”ผ โˆซ 5 " ๐‘Š. ([). ๐‘‘๐‘Š = " ! ยจ KLIMONTOVITCH: โˆซ 5 " ๐‘Š. (]). ๐‘‘๐‘Š = 4 " ! ! + " ! ๐”ผ โˆซ 5 " ๐‘Š. ([). ๐‘‘๐‘Š = ๐‘ก ยจ Remember of course that ๐”ผ ๐‘Š = 0 and ๐”ผ ๐‘Š! = ๐‘ก 30
  • 32. Luc_Faucheux_2021 A very quick summary ยจ In the deck II on stochastic calculus, we reviewed the following equations in increasing order of complexity: ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž. ๐‘‘๐‘ก ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘. ๐‘‘๐‘Š ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž. ๐‘‘๐‘ก + ๐‘. ๐‘‘๐‘Š ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก . ๐‘‘๐‘ก + ๐‘. ๐‘‘๐‘Š ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘ ๐‘ก . ๐‘‘๐‘Š ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก . ๐‘‘๐‘Š ยจ In all cases we were justified to write โ€œ.โ€, and this was the โ€usualโ€ product that we are used to. There was no confusion. This is because the drift and diffusion coefficients were NOT functions of the stochastic variables 32
  • 33. Luc_Faucheux_2021 A very quick summary - II ยจ HOWEVER, if there is one thing that you should remember from this deck, is that you CANNOT write something like: ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘Š(๐‘ก) ยจ The โ€œ.โ€ is meaningless if you want in this formulation ยจ As soon as the diffusion coefficient becomes inhomogeneous (depends on the position if you thinking about a physical diffusion process), or the volatility depends on the level of the underlier in Finance, you need to pick a convention (Ito, Strato, Klimontovitch, or anything else) to explain what you mean by โ€œ.โ€ in the term ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘Š(๐‘ก) ยจ Stochastic calculus is a very strange world as we saw in deck I, and just because you are writing something that looks familiar does not mean that you are allowed to do it and use it. ยจ You should just go the extra mile and really write: ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ). ๐‘‘๐‘Š ยจ Just to know that you cannot use โ€œ.โ€ in the diffusive term 33
  • 34. Luc_Faucheux_2021 A very quick summary - III ยจ Note that the drift term is ok ยจ You do not need to specify: ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ). ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ). ๐‘‘๐‘Š(๐‘ก) ยจ This has to do with the definition of the integral ยจ Remember you cannot differentiate in stochastic calculus (or at least if you do you should be very careful), always better to integrate and write SIE instead of SDE ยจ For all the previous examples, and even for a non-homogeneous drift term, the point you pick inside the bin when you define the integral as the limit of a sum, does not matter, tey all converge to the same usual Riemann integral. ยจ HOWEVER for non-homogeneous diffusion, the point you pick does matter 34
  • 35. Luc_Faucheux_2021 A very quick summary - IV ยจ You pick the left point -> ITO ยจ which in the [๐›ผ] calculus means ๐›ผ = 0 ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ). ๐‘‘๐‘Š(๐‘ก) ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ = 0). ๐‘‘๐‘Š(๐‘ก) ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) 35
  • 36. Luc_Faucheux_2021 A very quick summary - V ยจ You pick the middle point -> STRATANOVITCH ยจ which in the [๐›ผ] calculus means ๐›ผ = 1/2 ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ). ๐‘‘๐‘Š(๐‘ก) ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ = 1/2). ๐‘‘๐‘Š(๐‘ก) ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก) 36
  • 37. Luc_Faucheux_2021 A very quick summary - VI ยจ You pick the right point -> KLIMONTOVITCH ยจ which in the [๐›ผ] calculus means ๐›ผ = 1 ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ). ๐‘‘๐‘Š(๐‘ก) ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ = 1). ๐‘‘๐‘Š(๐‘ก) ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (]). ๐‘‘๐‘Š(๐‘ก) 37
  • 38. Luc_Faucheux_2021 A very quick summary - VII ยจ ALWAYS go back to the SIE: ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ "#"$ "#"% ,. ,- . ([๐›ผ]). ๐‘‘๐‘‹(๐‘ก) + * ! โˆ’ ๐›ผ . โˆซ "#"$ "#"% ,!. ,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ! . ๐‘‘๐‘ก ยจ ONLY in the STRATNOVITCH case of ๐›ผ = 1/2 does the above reduces to: ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ "#"$ "#"% ,. ,- . ([๐›ผ]). ๐‘‘๐‘‹(๐‘ก) ยจ Which FORMALLY looks like the usual rules of calculus ยจ But think of this more as a coincidence than something that you can use willy nilly with utmost confidence. ยจ A little like Malliavin derivation operator has been designed to formally follow the usual chain rule, it is more of a coincidence than a given, so always be super careful 38
  • 39. Luc_Faucheux_2021 A very quick summary - VIII ยจ Also another quick note ยจ For sake of simplicity a lot of books or also in this deck we write: ,. ,- ยจ What we mean really is: ยจ ๐‘“ is a regular function of say variable ๐‘ฅ ยจ Partial derivatives ,. ,0 and ,!. ,0! are well defined quantities ยจ We can evaluate those quantities for ๐‘ฅ = ๐‘‹(๐‘ก) ยจ We would then write something like ,. ,0 |0#-(") and ,!. ,0! |0#-(") ยจ But sometimes because we are lazy or want to keep the notations simple we just write: ยจ ,. ,- and ,!. ,-! 39
  • 40. Luc_Faucheux_2021 A very quick summary - IX ยจ ,. ,- and ,!. ,-! are really not well defined quantity ยจ Remember that we said right at the beginning that one of the defining charaterictic of a stochastic process is that it was NOT differentiable ยจ So really what a lot of textbooks (like Hull for example) mean (or in this deck also), is that we are trying to not make the equations too cumbersome, so really when we write: ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ "#"$ "#"% ,. ,0 . ([๐›ผ]). ๐‘‘๐‘‹(๐‘ก) + * ! โˆ’ ๐›ผ . โˆซ "#"$ "#"% ,!. ,0! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ! . ๐‘‘๐‘ก ยจ or ยจ ๐‘“ ๐‘‹ ๐‘ก% โˆ’ ๐‘“ ๐‘‹ ๐‘ก$ = โˆซ "#"$ "#"% ,. ,- . ([๐›ผ]). ๐‘‘๐‘‹(๐‘ก) + * ! โˆ’ ๐›ผ . โˆซ "#"$ "#"% ,!. ,-! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ! . ๐‘‘๐‘ก 40
  • 41. Luc_Faucheux_2021 A very quick summary - X ยจ We really should be writing: ยจ ๐‘“ ๐‘‹ ๐‘ก" โˆ’ ๐‘“ ๐‘‹ ๐‘ก# = โˆซ $%$# $%$" &'()) &) |)%+($). ([๐›ผ]). ๐‘‘๐‘‹(๐‘ก) + , - โˆ’ ๐›ผ . โˆซ $%$# $%$" &!'()) &)! |)%+($). ๐‘ ๐‘ก, ๐‘‹ ๐‘ก - . ๐‘‘๐‘ก ยจ That is getting a little cumbersome ยจ Baxter is one of the few textbooks that actually bothers to be fully rigorous in his notations. ยจ So many apologies in those deck if depending on how much space I have on the line, but sometimes I will use one of those notations or another, but what I mean really is, going from the most exact and rigorous to the simplest in notation but least rigorous ยจ ,.(0) ,0 |0#-(") = ,.(0#-(")) ,0 = ,.(-(")) ,0 = ,.(-(")) ,- = ,. ,- ยจ Same for the other higher order of derivatives 41
  • 42. Luc_Faucheux_2021 The McKean and Goodman derivation of the link between SDE and PDE for the PDF, a first taste of the Feynmann-Kac theorem 42
  • 43. Luc_Faucheux_2021 Non-linear SDE โ€“ backward equation ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š ยจ We are in the general case now ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก and ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ยจ ๐‘‹ ๐‘ก% โˆ’ ๐‘‹ ๐‘ก$ = โˆซ "#"$ "#"% ๐‘‘๐‘‹ ๐‘ก = โˆซ "#"$ "#"% ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก) + โˆซ "#"$ "#"% ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ยจ We follow here the derivation from McKean and Goodman, but might actually also re-derive it from first principle after just to convince ourselves ยจ We are in ITO calculus, and will use the fact that the ITO integral of a trading strategy is a martingale. ยจ For any function ๐ถ(๐‘‹ ๐‘ก , ๐‘ก) (hint: we are using ๐ถ notation like in a call option, so people who have derived before the Black-Sholes equation should be in familiar territory). ยจ The Ito lemma reads: 43
  • 44. Luc_Faucheux_2021 Non-linear SDE โ€“ backward equation - II ยจ ๐ถ ๐‘‹ ๐‘ก% , ๐‘ก% โˆ’ ๐ถ ๐‘‹ ๐‘ก$ , ๐‘ก$ = โˆซ "#"$ "#"% ,8 ,- . ([). ๐‘‘๐‘‹(๐‘ก) + * ! โˆซ "#"$ "#"% ,!8 ,-! ๐‘‹ ๐‘ก . ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ! . ๐‘‘๐‘ก + โˆซ "#"$ "#"% ,8 ," (๐‘‹ ๐‘ก ). ๐‘‘๐‘ก ยจ In the โ€limitโ€ of small time increments, this can be written formally as the Ito lemma: ยจ ๐›ฟ๐ถ = ,8 ,- . ๐›ฟ๐‘‹ + * ! . ,!8 ,-! . ๐›ฟ๐‘‹! + ,8 ," . ๐›ฟ๐‘ก ยจ Note here the factor (1/2) coming from the Taylor expansion ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š ยจ ๐‘‘๐‘‹ ๐‘ก ! = ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ! . ๐‘‘๐‘ก ยจ ๐›ฟ๐ถ = ,8 ,- . ๐›ฟ๐‘‹ + * ! . ,!8 ,-! . ๐‘!๐›ฟ๐‘ก + ,8 ," . ๐›ฟ๐‘ก 44
  • 45. Luc_Faucheux_2021 Non-linear SDE โ€“ backward equation - III ยจ ๐›ฟ๐ถ = ,8 ,- . ๐›ฟ๐‘‹ + * ! . ,!8 ,-! . ๐‘!๐›ฟ๐‘ก + ,8 ," . ๐›ฟ๐‘ก ยจ ๐›ฟ๐ถ = ,8 ," + * ! . ,!8 ,-! . ๐‘! + ,8 ,- . ๐‘Ž . ๐›ฟ๐‘ก + ,8 ,- . ๐‘Ž. ๐›ฟ๐‘Š ยจ ๐ถ ๐‘‹ ๐‘ก% , ๐‘ก% โˆ’ ๐ถ ๐‘‹ ๐‘ก$ , ๐‘ก$ = โˆซ "#"$ "#"% ,8 ,- . ๐‘ ๐‘‹ ๐‘ก , ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) + โˆซ "#"$ "#"% ,8(- " ,") ," + * ! . ,!8(- " ,") ,-! . ๐‘(๐‘‹ ๐‘ก , ๐‘ก)! + ,8(- " ,") ,- . ๐‘Ž(๐‘‹ ๐‘ก , ๐‘ก) . ๐‘‘๐‘ก ยจ It would be nice if the term of first order in time would disappear. ยจ This is what the Feynman-Kac theorem is about ยจ IF the function ๐ถ(๐‘‹ ๐‘ก , ๐‘ก) is such that: ยจ ,8(- " ,") ," + * ! . ๐‘(๐‘‹ ๐‘ก , ๐‘ก)!. ,!8(- " ,") ,-! + ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก . ,8(- " ,") ,- = 0 45
  • 46. Luc_Faucheux_2021 Non-linear SDE โ€“ backward equation - IV ยจ Then: ยจ ๐ถ ๐‘‹ ๐‘ก% , ๐‘ก% โˆ’ ๐ถ ๐‘‹ ๐‘ก$ , ๐‘ก$ = โˆซ "#"$ "#"% ,8 ,- . ๐‘ ๐‘‹ ๐‘ก , ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ยจ Fixing ๐‘ก% = ๐‘‡ maturity of the option (boundary condition) ยจ And the conditional expectation: ยจ ๐”ผ{๐ถ(๐‘‹ ๐‘ก , ๐‘ก)|๐‘‹ ๐‘ก$ = ๐‘‹$} ยจ ๐”ผ ๐ถ ๐‘‹ ๐‘ก% , ๐‘ก% ๐‘‹ ๐‘ก$ = ๐‘‹$ = ๐”ผ ๐ถ ๐‘‹ ๐‘ก$ , ๐‘ก$ ๐‘‹ ๐‘ก$ = ๐‘‹$ + ๐”ผ โˆซ "#"$ "#"% ,8 ,- . ๐‘ ๐‘‹ ๐‘ก , ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ๐‘‹ ๐‘ก$ = ๐‘‹$ ยจ ๐”ผ ๐ถ ๐‘‹ ๐‘ก$ , ๐‘ก$ ๐‘‹ ๐‘ก$ = ๐‘‹$ = ๐ถ ๐‘‹ ๐‘ก$ , ๐‘ก$ ยจ ๐”ผ โˆซ "#"$ "#"% ,8 ,- . ๐‘ ๐‘‹ ๐‘ก , ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ๐‘‹ ๐‘ก$ = ๐‘‹$ = 0 ยจ because this is an ITO integral (note that this would NOT be the case in STRATO, so here it actually pays to work in ITO) 46
  • 47. Luc_Faucheux_2021 Non-linear SDE โ€“ backward equation - V ยจ ๐ถ ๐‘‹ ๐‘ก$ , ๐‘ก$ = ๐”ผ ๐ถ ๐‘‹ ๐‘ก% , ๐‘ก% ๐‘‹ ๐‘ก$ = ๐‘‹$ ยจ This is essentially the Feynman-Kac formula ยจ IF ๐ถ(๐‘ฅ, ๐‘ก) follows the BACKWARD Kolmogorov (FP) equation: ยจ ,8(0,") ," + * ! . ๐‘(๐‘ฅ, ๐‘ก)!. ,!8(0,") ,0! + ๐‘Ž ๐‘ฅ, ๐‘ก . ,8(0,") ,0 = 0 ยจ Then ๐ถ(๐‘ฅ, ๐‘ก) can be written as a conditional expectation: ยจ ๐ถ ๐‘ฅ, ๐‘ก = ๐”ผ ๐ถ ๐‘‹ ๐‘ก% , ๐‘ก% ๐‘‹ ๐‘ก = ๐‘ฅ ยจ Under the probability measure such that ๐‘‹ ๐‘ก is an ITO process with the ITO SDE: ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š ยจ ๐‘‘๐‘Š is a Wiener process (Brownian motion) under that probability measure, and the initial condition is ๐‘‹ ๐‘ก = ๐‘ฅ 47
  • 48. Luc_Faucheux_2021 Non-linear SDE โ€“ backward equation - VI ยจ Note that it is called BACKWARD because it concerns the expectation of something that is fixed in the future at time ๐‘ก% = ๐‘‡ (Boundary condition) ยจ Just to be in familiar territory, if the boundary condition is a call payoff ยจ ๐ถ ๐‘‹ ๐‘ก% , ๐‘ก% = ๐ถ ๐‘‹ ๐‘‡ , ๐‘‡ = ๐‘€๐ด๐‘‹(๐‘‹ ๐‘‡ โˆ’ ๐พ, 0) ยจ ๐ถ ๐‘ฅ, ๐‘ก = ๐”ผ ๐‘€๐ด๐‘‹(๐‘‹ ๐‘‡ โˆ’ ๐พ, 0) ๐‘‹ ๐‘ก = ๐‘ฅ ยจ ,8(0,") ," + * ! . ๐‘(๐‘ฅ, ๐‘ก)!. ,!8(0,") ,0! + ๐‘Ž ๐‘ฅ, ๐‘ก . ,8(0,") ,0 = 0 ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š ยจ From the deck on numeraire, using the Money Market account: ยจ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก = ๐‘Ÿ๐‘† ยจ ๐‘ ๐‘ก, ๐‘‹ ๐‘ก = ๐œŽ๐‘† ยจ ๐‘‘๐‘† ๐‘ก = ๐‘Ÿ๐‘†. ๐‘‘๐‘ก + ๐œŽ๐‘†. ([). ๐‘‘๐‘Š 48
  • 49. Luc_Faucheux_2021 Non-linear SDE โ€“ backward equation - VII ยจ ,8(0,") ," + * ! . ๐‘(๐‘ฅ, ๐‘ก)!. ,!8(0,") ,0! + ๐‘Ž ๐‘ฅ, ๐‘ก . ,8(0,") ,0 = 0 ยจ ,8(:,") ," + * ! . (๐œŽ๐‘†)!. ,!8(:,") ,:! + ๐‘Ÿ๐‘†. ,8(:,") ,: = 0 ยจ This is exactly the Black-Sholes equation for ๐ถ ๐‘†, ๐‘ก = ๐ถ5(๐‘†, ๐‘ก)/๐‘’1;(<1") ยจ Backward equations are for conditional expectations (we know the final value, we want to evaluate the value of a derivative or a claim now) ยจ Forward equations like the one we have been working on for the PDF tend to be for probability distributions (we know the starting point, usually a Dirac peak, and we look at the evolution in time) ยจ Note that in the BACKWARD equation, the terms ๐‘(๐‘ฅ, ๐‘ก) and ๐‘Ž(๐‘ฅ, ๐‘ก) are OUTSIDE the partial derivatives (because that came from using Ito lemma) ยจ Note that in the FORWARD equation, the terms will be INSIDE ยจ Note that the PDF will follow BOTH the backward and forward (PDF is expectation of a Delta peak payoff, following Dupire, we will do it again) 49
  • 50. Luc_Faucheux_2021 Non-linear SDE โ€“ forward equation - ยจ OK, so we still do not know what equation does the PDF follow. ยจ We almost there ยจ ๐ถ ๐‘‹ ๐‘ก% , ๐‘ก% โˆ’ ๐ถ ๐‘‹ ๐‘ก$ , ๐‘ก$ = โˆซ "#"$ "#"% ,8 ,- . ๐‘ ๐‘‹ ๐‘ก , ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) + โˆซ "#"$ "#"% ,8(- " ,") ," + * ! . ,!8(- " ,") ,-! . ๐‘(๐‘‹ ๐‘ก , ๐‘ก)! + ,8(- " ,") ,- . ๐‘Ž(๐‘‹ ๐‘ก , ๐‘ก) . ๐‘‘๐‘ก ยจ Letโ€™s not assume that the term in time vanishes ยจ Letโ€™s actually assume that ๐ถ ๐‘‹ ๐‘ก , ๐‘ก can be ANY function, and in particular we can choose ยจ ๐ถ ๐‘‹ ๐‘ก = โˆž , ๐‘ก = โˆž = 0 ยจ ๐ถ ๐‘‹ ๐‘ก = ๐‘ก$ , ๐‘ก = ๐‘ก$ = 0 ยจ 0 = โˆซ "#". "#( ,8 ,- . ๐‘ ๐‘‹ ๐‘ก , ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) + โˆซ "#". "#( ,8(- " ,") ," + * ! . ,!8(- " ,") ,-! . ๐‘(๐‘‹ ๐‘ก , ๐‘ก)! + ,8(- " ,") ,- . ๐‘Ž(๐‘‹ ๐‘ก , ๐‘ก) . ๐‘‘๐‘ก 50
  • 51. Luc_Faucheux_2021 Non-linear SDE โ€“ forward equation - II ยจ Let us look again at the conditional expectation, but this time we will explicitly write it as an integral over the possible outcomes ยจ ๐”ผ ๐ต ๐‘‹ ๐‘ก , ๐‘ก ๐‘‹ ๐‘ก$ = ๐‘‹$ = โˆซ 0#1( 0#+( ๐ต ๐‘ฅ, ๐‘ก . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$) . ๐‘‘๐‘ฅ ยจ We are obviously after what kind of equation could ๐‘ƒ(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$) verify ยจ Now again, because we are in ITO calculus: ยจ ๐”ผ โˆซ "#". "#< ,8 ,- . ๐‘ ๐‘‹ ๐‘ก , ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ๐‘‹ ๐‘ก$ = ๐‘‹$ = 0 ยจ In particular: ยจ ๐”ผ โˆซ "#". "#+( ,8 ,- . ๐‘ ๐‘‹ ๐‘ก , ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ๐‘‹ ๐‘ก$ = ๐‘‹$ = 0 51
  • 52. Luc_Faucheux_2021 Non-linear SDE โ€“ forward equation - III ยจ 0 = ๐”ผ โˆซ "#". "#( ,8(- " ,") ," + * ! . ,!8(- " ,") ,-! . ๐‘(๐‘‹ ๐‘ก , ๐‘ก)! + ,8(- " ,") ,- . ๐‘Ž(๐‘‹ ๐‘ก , ๐‘ก) . ๐‘‘๐‘ก ๐‘‹ ๐‘ก$ = ๐‘‹$ ยจ 0 = ๐”ผ โˆซ "#". "#( ,8 ," + * ! . ๐‘(๐‘‹ ๐‘ก , ๐‘ก)!. ,!8 ,-! + ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก . ,8 ,- . ๐‘‘๐‘ก ๐‘‹ ๐‘ก$ = ๐‘‹$ ยจ 0 = โˆซ "#". "#( โˆซ 0#1( 0#+( ,8 ," + * ! . ๐‘(๐‘ฅ, ๐‘ก)!. ,!8 ,0! + ๐‘Ž ๐‘ฅ, ๐‘ก . ,8 ,0 . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$) . ๐‘‘๐‘ฅ. ๐‘‘๐‘ก ยจ Remember that this is true for ANY function ๐ถ(๐‘‹ ๐‘ก , ๐‘ก) that we can choose arbitrarily so that the boundary values, as well as the derivatives, vanishes ยจ We now integrate by part the above integral (we are now in the world of regular calculus, so the usual rules of calculus apply, none of this ITO / STRATO issue) ยจ 0 = โˆซ "#". "#( โˆซ 0#1( 0#+( โˆ’๐ถ ,= ," + ๐ถ. * ! . ,! ,0! [๐‘(๐‘ฅ, ๐‘ก)!. ๐‘] โˆ’ ๐ถ. , ,0 [๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘] . ๐‘‘๐‘ฅ. ๐‘‘๐‘ก 52
  • 53. Luc_Faucheux_2021 Non-linear SDE โ€“ forward equation - IV ยจ 0 = โˆซ "#". "#( โˆซ 0#1( 0#+( โˆ’๐ถ ,= ," + ๐ถ. * ! . ,! ,0! [๐‘(๐‘ฅ, ๐‘ก)!. ๐‘] โˆ’ ๐ถ. , ,0 [๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘] . ๐‘‘๐‘ฅ. ๐‘‘๐‘ก ยจ This is true for any function ๐ถ(๐‘ฅ, ๐‘ก) ยจ And so: ยจ 0 = โˆ’ ,= ," + * ! . ,! ,0! [๐‘(๐‘ฅ, ๐‘ก)!. ๐‘] โˆ’ , ,0 [๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘] ยจ ,=(0,"|-.,".) ," = โˆ’ , ,0 ๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ + * ! . ,! ,0! [๐‘(๐‘ฅ, ๐‘ก)!. ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)] ยจ ,=(0,"|-.,".) ," = โˆ’ , ,0 ๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’ , ,0 [ * ! . [๐‘(๐‘ฅ, ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)] ยจ This is the FORWARD Kolmogorov (FP) equation, because we know the initial condition and we are looking at an advection/diffusion in time ยจ Note that in the FORWARD equation the terms ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก and ๐‘ ๐‘‹ ๐‘ก , ๐‘ก are INSIDE the partial derivatives (because that came from integrating by parts the ITO lemma) 53
  • 54. Luc_Faucheux_2021 Non-linear SDE โ€“ forward equation - V ยจ ,=(0,"|-.,".) ," = โˆ’ , ,0 ๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’ , ,0 [ * ! . [๐‘(๐‘ฅ, ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)] ยจ ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ = ๐‘ ๐‘ฅ, ๐‘ก subject to the initial condition: ๐‘(๐‘ฅ, ๐‘ก$) = ๐›ฟ ๐‘ฅ โˆ’ ๐‘‹$ ยจ ,=(0,") ," = โˆ’ , ,0 [๐‘€* ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก โˆ’ , ,0 [๐‘€! ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ]] ยจ < โˆ†๐‘‹ > = ๐ธ โˆ†๐‘‹ =< ๐‘ฅ >"+โˆ†" โˆ’< ๐‘ฅ >"= ๐น* ๐‘‹ ๐‘ก , ๐‘ก . โˆ†๐‘ก (drift term) ยจ < โˆ†๐‘‹!> = ๐ธ โˆ†๐‘‹! =< (๐‘ฅโˆ’< ๐‘ฅ >"+โˆ†")!>"+โˆ†"= ๐น! ๐‘‹ ๐‘ก , ๐‘ก . โˆ†๐‘ก (diffusion term) ยจ We showed that ๐น* ๐‘‹ ๐‘ก , ๐‘ก = ๐‘€* ๐‘‹ ๐‘ก , ๐‘ก and ๐น! ๐‘‹ ๐‘ก , ๐‘ก = 2. ๐‘€! ๐‘‹ ๐‘ก , ๐‘ก ยจ We now have also shown that: ยจ ๐‘€* ๐‘ฅ, ๐‘ก = ๐‘Ž ๐‘ฅ, ๐‘ก ยจ ๐‘€! ๐‘ฅ, ๐‘ก = * ! . ๐‘(๐‘ฅ, ๐‘ก)! ยจ Note that we do not always have an explicit solution for ๐‘ ๐‘ฅ, ๐‘ก at this point. 54
  • 55. Luc_Faucheux_2021 Non-linear SDE โ€“ Summary ยจ As we saw with the Dupire approach, what is interesting is that the PDF follows BOTH a FORWARD and a BACKWARD equation: ยจ In short in ITO: ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š ยจ ,=(0,"|-.,".) ," = โˆ’ , ,0 ๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’ , ,- [ * ! . [๐‘(๐‘ฅ, ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)] ยจ ,=(0,"|-.,".) ," = โˆ’ , ,0 ๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ ] + * ! . ,! ,-! [. [๐‘(๐‘ฅ, ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)] ยจ This is the FORWARD equation: given the starting point (๐‘‹$, ๐‘ก$), the conditional probability ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$) evolves forward in time ยจ ,=(0,"|-.,".) ,". = โˆ’๐‘Ž ๐‘‹$, ๐‘ก$ , ,-. ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’ * ! . ๐‘(๐‘‹$, ๐‘ก$)! ,! ,-. ! ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$) ยจ This is the BACKWARD equation: given the end point (๐‘ฅ, ๐‘ก), the conditional probability can also be expressed as an expectation (of the Delta peak), and will thus follow a BACKWARD FP diffusing backward in time 55
  • 56. Luc_Faucheux_2021 Non-linear SDE โ€“ Summary II ยจ ๐‘(๐‘‹%, ๐‘ก%|๐‘‹$, ๐‘ก$) = โˆซ 0#1( 0#+( ๐‘(๐‘ฅ, ๐‘ก%|๐‘‹$, ๐‘ก$). ๐›ฟ(๐‘ฅ โˆ’ ๐‘‹%). ๐‘‘๐‘ฅ ยจ This is fairly obvious, and recast ๐‘ƒ(๐‘‹%, ๐‘ก%|๐‘‹$, ๐‘ก$) as an expectation of a terminal payoff, which is the Delta peak ๐›ฟ(๐‘ฅ โˆ’ ๐‘‹%), with the probability measure associated to ๐‘(๐‘‹%, ๐‘ก%|๐‘‹$, ๐‘ก$) . ยจ Just like the call option was the expectation of the terminal payoff ๐‘€๐ด๐‘‹(๐‘†2 โˆ’ ๐พ, 0) ยจ So the idea is that we can apply ITO lemma to ๐‘(๐‘‹%, ๐‘ก%|๐‘‹$, ๐‘ก$) (this time on the โ€œstartingโ€ variables (๐‘‹$, ๐‘ก$) ยจ This s exactly what we just did, but because it can be confusing since in some ways, the conditional probability is โ€œtwo thingsโ€: it is the conditional probability that we use in order to calculate the expectation (integral) of ANY derivative function, and hence will follow the FORWARD equation, it is ALSO the expectation of a claim (mainly in an evident manner the Delta peak), and so will follow the BACKWARD equation ยจ Note: only in ITO calculus can we zero out the integral as a martingale, otherwise we would have to carry those terms in the derivation 56
  • 57. Luc_Faucheux_2021 Why going through this trouble? The Ito-Stratanovitch controversy 57
  • 58. Luc_Faucheux_2021 Why did we go through all this trouble? ยจ Letโ€™s recap: ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š in ITO calculus ยจ We have shown that in that case the PDF follows a FORWARD ITO Kolmogorov (FP) ยจ ,=(0,"|-.,".) ," = โˆ’ , ,0 ๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’ , ,0 [ * ! . [๐‘(๐‘ฅ, ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)] ยจ < โˆ†๐‘‹ > = ๐ธ โˆ†๐‘‹ =< ๐‘ฅ >"+โˆ†" โˆ’< ๐‘ฅ >"= ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก . โˆ†๐‘ก (advection term) ยจ < โˆ†๐‘‹!> = ๐ธ โˆ†๐‘‹! =< (๐‘ฅโˆ’< ๐‘ฅ >"+โˆ†")!>"+โˆ†"= ๐‘(๐‘‹ ๐‘ก , ๐‘ก)!. โˆ†๐‘ก (diffusion term) 58
  • 59. Luc_Faucheux_2021 Why did we go through all this trouble? - II ยจ We ALSO know that going between ITO and [๐›ผ]: ยจ The ITO SDE: ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š ยจ Has the same solution (is the same) as the [๐›ผ] SDE in [๐›ผ] calculus: ยจ ๐‘‘๐‘‹ ๐‘ก = [๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก โˆ’ ๐›ผ. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ]. ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š ยจ The [๐›ผ] SDE ยจ ๐‘‘๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š ยจ Has the same solution (is the same) as the ITO SDE in ITO calculus ยจ ๐‘‘๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + ๐›ผ. A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š 59
  • 60. Luc_Faucheux_2021 Why did we go through all this trouble? - III ยจ And so, if we start with an [๐›ผ] SDE: ๐‘‘๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š ยจ It has the same solution (is the same) as the ITO SDE in ITO calculus ยจ ๐‘‘๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + ๐›ผ. A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š ยจ Which will then follow the ITO FORWARD Kolmogorov (FP): ยจ ,@(0,"|-.,".) ," = โˆ’ , ,0 ๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ƒ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’ , ,0 [ * ! . [๐‘(๐‘ฅ, ๐‘ก)! . ๐‘ƒ(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)] ยจ With: ยจ ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + ๐›ผ. A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ยจ ๐‘ ๐‘‹ ๐‘ก , ๐‘ก = A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ยจ ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + ๐›ผ. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- ๐‘ ๐‘ก, ๐‘‹ ๐‘ก 60
  • 61. Luc_Faucheux_2021 Why did we go through all this trouble? โ€“ III - a ยจ < โˆ†๐‘‹ > = ๐ธ โˆ†๐‘‹ =< ๐‘ฅ >"+โˆ†" โˆ’< ๐‘ฅ >"= ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก . โˆ†๐‘ก (advection term) ยจ < โˆ†๐‘‹!> = ๐ธ โˆ†๐‘‹! =< (๐‘ฅโˆ’< ๐‘ฅ >"+โˆ†")!>"+โˆ†"= ๐‘(๐‘‹ ๐‘ก , ๐‘ก)!. โˆ†๐‘ก (diffusion term) ยจ With: ยจ ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + ๐›ผ. A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ยจ ๐‘ ๐‘‹ ๐‘ก , ๐‘ก = A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ยจ Remember that we started from : ๐‘‘๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š ยจ So: < โˆ†๐‘‹ > = ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก . โˆ†๐‘ก = {@ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + ๐›ผ. A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก }. โˆ†๐‘ก ยจ < โˆ†๐‘‹ > = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . โˆ†๐‘ก + ๐›ผ. A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . โˆ†๐‘ก ยจ < โˆ†๐‘‹ > =< @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . [๐›ผ] . ๐‘‘๐‘Š > 61
  • 62. Luc_Faucheux_2021 Why did we go through all this trouble? โ€“ III - b ยจ < โˆ†๐‘‹ > = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . โˆ†๐‘ก + ๐›ผ. A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . โˆ†๐‘ก ยจ < โˆ†๐‘‹ > = < @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก > + < A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . [๐›ผ] . ๐‘‘๐‘Š > ยจ And ยจ < @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก > = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . โˆ†๐‘ก ยจ So we have: ยจ < A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . [๐›ผ] . ๐‘‘๐‘Š > = ๐›ผ. A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . โˆ†๐‘ก ยจ In particular: ยจ ITO : < A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . [ . ๐‘‘๐‘Š > = 0 ยจ STRATO: < A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . โˆ˜ . ๐‘‘๐‘Š > = [ * ! ]. A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . โˆ†๐‘ก 62
  • 63. Luc_Faucheux_2021 Why did we go through all this trouble? โ€“ III - c ยจ ITO : < A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . [ . ๐‘‘๐‘Š > = 0 ยจ One can also go back to the definition of the ITO integral (because remember it is never a SDE, it is ALWAYS and SIE) , but essentially the convention [ of taking the value โ€œbefore the jumpโ€, implies that the ITO integral is a martingale of expected value 0 ยจ STRATO: < A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . โˆ˜ . ๐‘‘๐‘Š > = [ * ! ]. A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . โˆ†๐‘ก ยจ Again we can explicitly derive this from the integral, but the convention โˆ˜ implies taking the value โ€œin the middle of the jumpโ€, hence the STRATO integral CANNOT be a martingale and has a non zero expected value. ยจ We did this derivation when we looking at the correspondence between the ITO and STRATO. ยจ It has been a while so might take a few pages here to redo it. 63
  • 64. Luc_Faucheux_2021 Why did we go through all this trouble? โ€“ III - d ยจ So essentially here is the deal: ยจ A PDE is well defined, it is just a Partial Differential Equation where the usual/Leibniz/regular/Leibniz calculus applies ยจ A SDE is NOT well defined, unless you specify which ๐›ผ-calculus you use (essentially which point within the bin you take when defining the integral as a limit of sum) ยจ Left point of the bin -> ITO , ๐›ผ = 0 ยจ Middle point of the bin -> STRATO, ๐›ผ = 1/2 ยจ Right point of the bin -> KLIMONT, ๐›ผ = 1 ยจ Letโ€™s summarize again before we delve more into the reason for the controversy 64
  • 66. Luc_Faucheux_2021 We need a nice summary to avoid any confusion ยจ ITO SDE is: ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š ยจ This implies that the PDF follows the FORWARD ITO Kolmogorov PDE ยจ ,=(0,"|-.,".) ," = โˆ’ , ,0 ๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’ , ,0 [ * ! . [๐‘(๐‘ฅ, ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)] ยจ ,= ," = โˆ’ , ,0 ๐‘Ž๐‘ โˆ’ , ,0 %!= ! = โˆ’ , ,0 [๐‘Ž๐‘] + * ! ,! ,0! [ %!= ! ] ยจ ,= ," = โˆ’ , ,0 ๐‘Ž๐‘ โˆ’ ๐‘ ,% ,0 . ๐‘ โˆ’ * ! . ๐‘! . , ,0 ๐‘ ยจ ,= ," = โˆ’ , ,- ๐‘Ž๐‘ โˆ’ ,A ,0 . ๐‘ โˆ’ ๐ท. , ,0 ๐‘ ยจ ,= ," = โˆ’ , ,0 ๐‘Ž๐‘ โˆ’ , ,0 (๐ท๐‘) using the convention ๐ท = * ! . ๐‘! 66
  • 67. Luc_Faucheux_2021 We need a nice summary to avoid any confusion - a ยจ ,= ," = โˆ’ , ,0 ๐‘Ž๐‘ โˆ’ , ,0 (๐ท๐‘) ยจ Physicists like a diffusion equation for particles that looks like: ยจ ,= ," = โˆ’ , ,0 ๐‘ ๐‘œ๐‘š๐‘’๐‘กโ„Ž๐‘–๐‘›๐‘” ยจ Because when integrating over the x-axis, and assuming some reasonable (read 0) value for the distribution function at infinity, the integral of the density is invariant with time (conservation of matter), or: ยจ , ," โˆซ 1( +( ๐‘ ๐‘ฅ, ๐‘ก . ๐‘‘๐‘ฅ = 0 ยจ That is a good thing. ยจ Physicists also like something called โ€œsteady-state solutionโ€, meaning at equilibrium (think very long time), ,= ," = 0 67
  • 68. Luc_Faucheux_2021 We need a nice summary to avoid any confusion - b ยจ ,= ," = 0 leads to: ยจ โˆ’ , ,0 ๐‘Ž๐‘ โˆ’ , ,0 (๐ท๐‘) ยจ For simplicity of argument, in the case of no external forcing (๐‘Ž = 0) ยจ , ,0 โˆ’ , ,0 (๐ท๐‘) = 0 ยจ So the solution of that is: ยจ , ,0 ๐ท๐‘ = ๐‘๐‘ก๐‘’ ยจ Again for things to no blow up at infinity, that means: , ,0 ๐ท๐‘ = ๐‘๐‘ก๐‘’ = 0 ยจ That means that ๐ท๐‘ = ๐‘๐‘ก๐‘’ with the normalization condition of โˆซ 1( +( ๐‘ ๐‘ฅ, ๐‘ก . ๐‘‘๐‘ฅ = 1 usually 68
  • 69. Luc_Faucheux_2021 We need a nice summary to avoid any confusion - c ยจ ๐ท๐‘ = ๐‘๐‘ก๐‘’ with the normalization condition of โˆซ 1( +( ๐‘ ๐‘ฅ, ๐‘ก . ๐‘‘๐‘ฅ = 1, which allows to calculate the value of that constant. ยจ And that is when physicists lost their wits in the 80s/90s, because the diffusion is a dynamic effect, and should not impact the equilibrium solution. ยจ The equilibrium solution is given by the laws of thermodynamics, and the equilibrium density should not depend on something like ๐ท ยจ Essentially the argument was that, if you give it enough time, the equilibrium distribution will be uniform, and should not depend on how fast or slow you diffuse in some regions ยจ That is a first flavor of the controversy that arose when we started to look in earnest in non- homogeneous diffusion processes ยจ Remember that it was not that long ago on the scale of the planet, Bachelier / Langevin / Einstein were just over a century ago now, Ito/Doelin in the 50s, so all in all fairly new 69
  • 70. Luc_Faucheux_2021 We need a nice summary to avoid any confusion - II ยจ [๐›ผ] SDE is: ๐‘‘๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š ยจ This implies that the PDF follows the FORWARD [๐›ผ] Kolmogorov PDE ยจ &/(),$|+",$") &$ = โˆ’ & &) {1 ๐‘Ž ๐‘ฅ, ๐‘ก + ๐›ผ. 4 ๐‘ ๐‘ฅ, ๐‘ก . & &) 4 ๐‘ ๐‘ฅ, ๐‘ก }. ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹#, ๐‘ก# โˆ’ & &) [ , - . [4 ๐‘(๐‘ฅ, ๐‘ก)- . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹#, ๐‘ก#)] ยจ ,= ," = โˆ’ , ,0 @ ๐‘Ž๐‘ + ๐›ผ. A ๐‘. , ,0 A ๐‘. ๐‘ โˆ’ , ,0 [ * ! . [A ๐‘! . ๐‘] ยจ ,= ," = โˆ’ , ,0 @ ๐‘Ž๐‘ + ๐›ผ. A ๐‘. , ,0 A ๐‘. ๐‘ โˆ’ , ,0 B %!= ! = โˆ’ , ,0 @ ๐‘Ž๐‘ + , ,0 ๐›ผ. A ๐‘. , ,- A ๐‘. ๐‘ + * ! ,! ,0! [ B %!= ! ] ยจ ,= ," = โˆ’ , ,0 @ ๐‘Ž๐‘ + ๐›ผ. A ๐‘. ,B % ,0 . ๐‘ โˆ’ A ๐‘. ,B % ,0 . ๐‘ โˆ’ * ! . A ๐‘!. , ,0 ๐‘ ยจ ,= ," = โˆ’ , ,0 ๐‘Ž๐‘ + ๐›ผ. ,C A ,0 . ๐‘ โˆ’ ,C A ,0 . ๐‘ โˆ’ b ๐ท. , ,0 ๐‘ 70
  • 71. Luc_Faucheux_2021 We need a nice summary to avoid any confusion - III ยจ [๐›ผ = 1/2] STRATO SDE is: ๐‘‘๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š ยจ This implies that the PDF follows the FORWARD STRATO Kolmogorov PDE ยจ &/(),$|+",$") &$ = โˆ’ & &) {1 ๐‘Ž ๐‘ฅ, ๐‘ก + , - . 4 ๐‘ ๐‘ฅ, ๐‘ก . & &) 4 ๐‘ ๐‘ฅ, ๐‘ก }. ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹#, ๐‘ก# โˆ’ & &) [ , - . [4 ๐‘(๐‘ฅ, ๐‘ก)- . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹#, ๐‘ก#)] ยจ ,= ," = โˆ’ , ,0 @ ๐‘Ž๐‘ + * ! . A ๐‘. ,B % ,0 . ๐‘ โˆ’ , ,0 * ! . [A ๐‘! . ๐‘ ยจ ,= ," = โˆ’ , ,0 @ ๐‘Ž๐‘ + * ! . A ๐‘. , ,0 A ๐‘. ๐‘ โˆ’ , ,0 B %!= ! = โˆ’ , ,0 @ ๐‘Ž๐‘ + , ,0 * ! . A ๐‘. , ,- A ๐‘. ๐‘ + * ! ,! ,0! [ B %!= ! ] ยจ ,= ," = โˆ’ , ,0 @ ๐‘Ž๐‘ + * ! . A ๐‘. ,B % ,0 . ๐‘ โˆ’ A ๐‘. ,B % ,0 . ๐‘ โˆ’ * ! . A ๐‘!. , ,0 ๐‘ = โˆ’ , ,0 @ ๐‘Ž๐‘ โˆ’ * ! A ๐‘ ,B % ,0 ๐‘ โˆ’ * ! A ๐‘! , ,0 ๐‘ ยจ ,= ," = โˆ’ , ,0 @ ๐‘Ž๐‘ โˆ’ * ! . ,C A ,0 . ๐‘ โˆ’ b ๐ท. , ,0 ๐‘ = โˆ’ , ,0 @ ๐‘Ž๐‘ + * ! . ,C A ,0 . ๐‘ โˆ’ , ,0 (b ๐ท๐‘) 71
  • 72. Luc_Faucheux_2021 We need a nice summary to avoid any confusion - IIIa ยจ Note that we can also write as: ยจ ,= ," = โˆ’ , ,0 @ ๐‘Ž๐‘ + * ! . A ๐‘. ,B % ,0 . ๐‘ โˆ’ , ,0 * ! . [A ๐‘! . ๐‘ ยจ ,= ," = โˆ’ , ,0 [@ ๐‘Ž๐‘] โˆ’ , ,0 [ * ! . A ๐‘. ,B % ,0 . ๐‘ โˆ’ * ! , B %!= ,0 ] ยจ And since: * ! , B %!= ,0 = * ! ,B %.B %= ,0 = * ! A ๐‘ ,B %= ,0 + * ! A ๐‘๐‘ ,B % ,0 ยจ ,= ," = โˆ’ , ,0 @ ๐‘Ž๐‘ + * ! , ,0 [A ๐‘ ,(B %=) ,0 ] ยจ ,=(0,"|-.,".) ," = โˆ’ , ,0 @ ๐‘Ž ๐‘ฅ, ๐‘ก ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ + * ! , ,0 [A ๐‘ ๐‘ฅ, ๐‘ก , ,0 {A ๐‘ ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ }] 72
  • 73. Luc_Faucheux_2021 We need a nice summary to avoid any confusion - IIIb ยจ That is right there in essence the crux of the matter: ยจ ,@ ," = โˆ’ , ,0 @ ๐‘Ž๐‘ โˆ’ * ! . ,C A ,0 . ๐‘ โˆ’ b ๐ท. , ,0 ๐‘ = โˆ’ , ,0 @ ๐‘Ž๐‘ + * ! . ,C A ,0 . ๐‘ โˆ’ , ,0 (b ๐ท๐‘) ยจ Because of course: ยจ , ,0 b ๐ท๐‘ = b ๐ท. , ,0 ๐‘ + ๐‘. , ,0 b ๐ท ยจ So you can see how that, combined with the { * ! . ,C A ,0 . ๐‘} term, can easily get people confused, because in a way you are โ€œmixingโ€ the drift term and the diffusion term. ยจ In some ways you should not do that, what is drift is drift, and what is diffusion is diffusion ยจ It is clear when you write the SDE/SIE, because the drift term is in ๐‘‘๐‘ก, and the diffusive term is in ๐‘‘๐‘Š, assuming that you have specified in which calculus you operate ยจ But when writing the PDE, it looks like you can willy nilly mix the two terms. 73
  • 74. Luc_Faucheux_2021 We need a nice summary to avoid any confusion - IV ยจ ITO SDE is: ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š ยจ This implies that the PDF follows the FORWARD ITO Kolmogorov PDE ยจ ,=(0,"|-.,".) ," = โˆ’ , ,0 ๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’ , ,0 [ * ! . [๐‘(๐‘ฅ, ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)] ยจ The PDF ALSO follows the BACKWARD ITO Kolmogorov PDE: ยจ ,=(0,"|-.,".) ,". = โˆ’๐‘Ž ๐‘‹$, ๐‘ก$ , ,-. ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’ * ! . ๐‘(๐‘‹$, ๐‘ก$)! ,! ,-. ! ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$) ยจ ,= ,". = โˆ’๐‘Ž ,= ,-. โˆ’ * ! . ๐‘! ,!= ,-. ! ยจ ,= ,". = โˆ’๐‘Ž ,= ,-. โˆ’ ๐ท ,!= ,-. ! ยจ Where I have explicitly kept the notation ๐‘ก$ and ๐‘‹$ to indicate the fact that this is a BACKWARD PDE (expectation of a payoff at maturity) 74
  • 75. Luc_Faucheux_2021 We need a nice summary to avoid any confusion - V ยจ [๐›ผ] SDE is: ๐‘‘๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š ยจ This implies that the PDF follows the FORWARD [๐›ผ] Kolmogorov PDE ยจ &/(),$|+",$") &$ = โˆ’ & &) {1 ๐‘Ž ๐‘ฅ, ๐‘ก + ๐›ผ. 4 ๐‘ ๐‘ฅ, ๐‘ก . & &) 4 ๐‘ ๐‘ฅ, ๐‘ก }. ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹#, ๐‘ก# โˆ’ & &) [ , - . [4 ๐‘(๐‘ฅ, ๐‘ก)- . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹#, ๐‘ก#)] ยจ The PDF ALSO follows the BACKWARD ITO Kolmogorov PDE: ยจ !"($,&|(!,&!) !&! = โˆ’ G ๐‘Ž ๐‘ก*, ๐‘‹ ๐‘ก* + ๐›ผ. B ๐‘ ๐‘ก*, ๐‘‹ ๐‘ก* . ! !(! B ๐‘ ๐‘ก*, ๐‘‹ ๐‘ก* ! !(! ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹*, ๐‘ก* โˆ’ + , . ๐‘(๐‘‹*, ๐‘ก*), !" !(! " ๐‘(๐‘ฅ, ๐‘ก|๐‘‹*, ๐‘ก*) ยจ ,= ,". = โˆ’ @ ๐‘Ž + ๐›ผ. A ๐‘ , ,-. A ๐‘ , ,-. ๐‘ โˆ’ * ! . ๐‘! ,! ,-. ! ๐‘ ยจ ,= ,". = โˆ’ @ ๐‘Ž + ๐›ผ. ,C A ,-. ,= ,-. โˆ’ b ๐ท ,!= ,-. ! 75
  • 76. Luc_Faucheux_2021 We need a nice summary to avoid any confusion - VI ยจ [๐›ผ = 1/2] STRATO SDE is: ๐‘‘๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([๐›ผ]). ๐‘‘๐‘Š ยจ This implies that the PDF follows the FORWARD STRATO Kolmogorov PDE ยจ &/(),$|+",$") &$ = โˆ’ & &) {1 ๐‘Ž ๐‘ฅ, ๐‘ก + , - . 4 ๐‘ ๐‘ฅ, ๐‘ก . & &) 4 ๐‘ ๐‘ฅ, ๐‘ก }. ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹#, ๐‘ก# โˆ’ & &) [ , - . [4 ๐‘(๐‘ฅ, ๐‘ก)- . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹#, ๐‘ก#)] ยจ The PDF ALSO follows the BACKWARD ITO Kolmogorov PDE: ยจ !"($,&|(!,&!) !&! = โˆ’ G ๐‘Ž ๐‘ก*, ๐‘‹ ๐‘ก* + + , . B ๐‘ ๐‘ก*, ๐‘‹ ๐‘ก* . ! !(! B ๐‘ ๐‘ก*, ๐‘‹ ๐‘ก* ! !(! ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹*, ๐‘ก* โˆ’ + , . ๐‘(๐‘‹*, ๐‘ก*), !" !(! " ๐‘(๐‘ฅ, ๐‘ก|๐‘‹*, ๐‘ก*) ยจ ,= ,". = โˆ’ @ ๐‘Ž + * ! . A ๐‘ , ,-. A ๐‘ , ,-. ๐‘ โˆ’ * ! . A ๐‘! ,! ,-. ! ๐‘ ยจ ,= ,". = โˆ’ @ ๐‘Ž + * ! . ,C A ,-. ,= ,-. โˆ’ b ๐ท ,!= ,-. ! 76
  • 77. Luc_Faucheux_2021 We need a nice summary to avoid any confusion โ€“ VI-a ยจ Note that we can also write as: ยจ ,= ,". = โˆ’ @ ๐‘Ž + * ! . ,C A ,-. ,= ,-. โˆ’ b ๐ท ,!= ,-. ! ยจ ,= ,". = โˆ’ @ ๐‘Ž + * ! . A ๐‘ , ,-. A ๐‘ , ,-. ๐‘ โˆ’ * ! . A ๐‘! ,! ,-. ! ๐‘ ยจ And since: , ,-. A ๐‘. ,= ,-. = ,B % ,-. . ,= ,-. + A ๐‘. ,!= ,-. ! ยจ ,= ,". = โˆ’@ ๐‘Ž. ,= ,-. โˆ’ * ! . A ๐‘ , ,-. [A ๐‘. ,= ,-. ] ยจ ,=(0,"|-.,".) ," = โˆ’@ ๐‘Ž ๐‘ก$, ๐‘‹ ๐‘ก$ . ,=(0,"|-.,".) ,-. โˆ’ * ! . A ๐‘ ๐‘ก$, ๐‘‹ ๐‘ก$ , ,-. A ๐‘ ๐‘ก$, ๐‘‹ ๐‘ก$ . ,=(0,"|-.,".) ,-. 77
  • 78. Luc_Faucheux_2021 We need a nice summary to avoid any confusion โ€“ VII ยจ ITO FORWARD ยจ ,=(0,"|-.,".) ," = โˆ’ , ,0 ๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’ , ,0 [ * ! . [๐‘(๐‘ฅ, ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)] ยจ ITO BACKWARD ยจ ,=(0,"|-.,".) ,". = โˆ’๐‘Ž ๐‘‹$, ๐‘ก$ , ,-. ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’ * ! . ๐‘(๐‘‹$, ๐‘ก$)! ,! ,-. ! ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$) ยจ STRATO FORWARD ยจ ,=(0,"|-.,".) ," = โˆ’ , ,0 @ ๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ + * ! , ,0 [A ๐‘ ๐‘ฅ, ๐‘ก , ,0 {A ๐‘ ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ }] ยจ STRATO BACKWARD ยจ ,=(0,"|-.,".) ," = โˆ’@ ๐‘Ž ๐‘ก$, ๐‘‹ ๐‘ก$ . ,=(0,"|-.,".) ,-. โˆ’ * ! . A ๐‘ ๐‘ก$, ๐‘‹ ๐‘ก$ , ,-. A ๐‘ ๐‘ก$, ๐‘‹ ๐‘ก$ . ,=(0,"|-.,".) ,-. 78
  • 79. Luc_Faucheux_2021 We need a nice summary to avoid any confusion โ€“ VIII ยจ ITO FORWARD ยจ ,=(0,"|-.,".) ," = โˆ’ , ,0 ๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’ , ,0 [๐ท(๐‘ฅ, ๐‘ก). ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)] ยจ ITO BACKWARD ยจ ,=(0,"|-.,".) ,". = โˆ’๐‘Ž ๐‘‹$, ๐‘ก$ , ,-. ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’ ๐ท ๐‘‹$, ๐‘ก$ ,! ,-. ! ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$) ยจ STRATO FORWARD ยจ ,=(0,"|-.,".) ," = โˆ’ , ,0 @ ๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ + , ,0 [ b ๐ท ๐‘ฅ, ๐‘ก , ,0 { b ๐ท ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ }] ยจ STRATO BACKWARD ยจ ,=(0,"|-.,".) ," = โˆ’@ ๐‘Ž ๐‘ก$, ๐‘‹ ๐‘ก$ . ,=(0,"|-.,".) ,-. โˆ’ b ๐ท ๐‘ก$, ๐‘‹ ๐‘ก$ , ,-. b ๐ท ๐‘ก$, ๐‘‹ ๐‘ก$ . ,=(0,"|-.,".) ,-. 79
  • 80. Luc_Faucheux_2021 We need a nice summary to avoid any confusion โ€“ IX ยจ Note that one of the characteristic of the STRATO equations is that they are much more symmetrical forward and backward (physicist like that because it does not break the time symmetry as much as ITO) ยจ In Ito, there is a clear distinction: the diffusion appears INSIDE the derivative for the FORWARD equation, and OUTSIDE for the BACKWARD ยจ In STRATO, it is split more equally ยจ STRATO FORWARD ยจ ,= ," = โˆ’ , ,0 @ ๐‘Ž๐‘ + , ,0 [ b ๐ท , ,- { b ๐ท. ๐‘}] ยจ STRATO BACKWARD ยจ ,= ," = โˆ’@ ๐‘Ž. ,= ,-. โˆ’ b ๐ท , ,-. b ๐ท. ,= ,-. 80
  • 81. Luc_Faucheux_2021 We need a nice summary to avoid any confusion โ€“ X ยจ Note that we in doubt it is also safe to recast the STRATO as an ITO using the rules of transformation: ยจ If we start from a STRATO SDE: ๐‘‘๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š ยจ The corresponding ITO SDE is: ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š ยจ With: ยจ ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + [ * ! ]. A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ยจ ๐‘ ๐‘‹ ๐‘ก , ๐‘ก = A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ยจ ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + [ * ! ]. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- ๐‘ ๐‘ก, ๐‘‹ ๐‘ก 81
  • 82. Luc_Faucheux_2021 We need a nice summary to avoid any confusion โ€“ XI ยจ And conversely: ยจ If we start from a ITO SDE: ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š ยจ The corresponding STRATO SDE is: ๐‘‘๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š ยจ With: ยจ @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก = ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก โˆ’ [ * ! ]. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ยจ A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก = ๐‘ ๐‘‹ ๐‘ก , ๐‘ก ยจ @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก = ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก โˆ’ [ * ! ]. A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก 82
  • 84. Luc_Faucheux_2021 The crux of the matter ยจ The main reason for the confusion is that people write something that looks like an ODE and assume that because we are using the same usual notations, a lot of the baggage we accumulated on regular calculus (Taylor expansion, Leibniz rule, chain rule,..) will also carry over. ยจ I think an easy rule to remember is that writing an SDE like: ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘Š(๐‘ก) ยจ Makes NO sense and is not well defined. 84
  • 85. Luc_Faucheux_2021 The crux of the matter - I ยจ A slight improvement from writing something wrong like: ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘Š(๐‘ก) ยจ Would be to write something like: ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ยจ Or: ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก) ยจ Or: ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (ฮฑ). ๐‘‘๐‘Š(๐‘ก) ยจ 85
  • 86. Luc_Faucheux_2021 The crux of the matter - II ยจ That way it is obvious that we are not writing the same thing: ยจ ITO ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐›ผ = 0 . ๐‘‘๐‘Š(๐‘ก) ยจ STRATONOVITCH ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก) ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐›ผ = 1/2 . ๐‘‘๐‘Š(๐‘ก) 86
  • 87. Luc_Faucheux_2021 The crux of the matter - III ยจ Writing it like this makes it obvious that we are NOT writing the same SDE, we are writing two very different processes ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก) ยจ IN FACT, if we want the two different equations to represent the same process, then obviously the coefficients cannot be the same. It turns out that we are quite lucky in the sense that there is a simple relations between the two. We are maybe too lucky because the fact that this relationship is simple sort of pushed people into believing that the two representations (Ito and Strato) are very close to each other, if not similar. Maybe if would have been easier if there had been not simple relationship between the coefficients, or a very complicated one 87
  • 88. Luc_Faucheux_2021 The crux of the matter โ€“ III a ยจ Say it another way ยจ Stochastic (Brownian) calculus is very different from regular (Newtonian/Leibniz) calculus ยจ ITO calculus is one possible definition of stochastic calculus ยจ STRATO calculus is one possible definition of stochastic calculus ยจ ITO calculus is very different from STRATO calculus ยจ However there is a relationship between the ITO integral and the STRATO integral ยจ ALWAYS go back to the integral form when dealing with stochastic calculus ยจ ALWAYS use an SIE if possible as opposed to an SDE 88
  • 89. Luc_Faucheux_2021 The crux of the matter - IV ยจ So it is actually easier in order not to get confused to add the โ€œtildeโ€ on top of the coefficients when dealing with Strato ยจ STRATO SDE: ๐‘‘๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก) ยจ ITO SDE: ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ยจ Just to emphasize the point that we are not writing the same equation and that those are two different processes 89
  • 90. Luc_Faucheux_2021 The crux of the matter - V ยจ If we start from a STRATO SDE: ๐‘‘๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก) ยจ The corresponding ITO SDE is: ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ยจ With: ยจ ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + [ * ! ]. A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ยจ ๐‘ ๐‘ก, ๐‘‹ ๐‘ก = A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ยจ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + [ * ! ]. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ยจ In that case, the two different SDE will describe the same stochastic process 90
  • 91. Luc_Faucheux_2021 The crux of the matter - VI ยจ And conversely: ยจ If we start from a ITO SDE: ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ยจ The corresponding STRATO SDE is: ๐‘‘๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก) ยจ With: ยจ @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก โˆ’ [ * ! ]. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ยจ A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก = ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ยจ @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก โˆ’ [ * ! ]. A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ยจ In that case, the two different SDE will describe the same stochastic process 91
  • 92. Luc_Faucheux_2021 The crux of the matter - VII ยจ Some note on notation: ยจ To be even more rigorous, we should not even write something like this: ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ยจ We should say that there is an analytical function ๐‘Ž ๐‘ก, ๐‘ฅ and ๐‘ ๐‘ก, ๐‘ฅ , and that the process is: ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘ฅ = ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘ฅ = ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ยจ ๐‘‘๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘ฅ = ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘ฅ = ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก) ยจ And so really we should not write: ยจ @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก โˆ’ [ * ! ]. ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- ๐‘ ๐‘ก, ๐‘‹ ๐‘ก ยจ But: ยจ @ ๐‘Ž ๐‘ก, ๐‘ฅ = ๐‘Ž ๐‘ก, ๐‘ฅ โˆ’ * ! . ๐‘ ๐‘ก, ๐‘ฅ . , ,0 ๐‘ ๐‘ก, ๐‘ฅ 92
  • 93. Luc_Faucheux_2021 The crux of the matter - VIII ยจ In fact, whenever the notation doe not get too cumbersome, we should not even write and SDE but and SIE ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ยจ ๐‘‹ ๐‘ก% โˆ’ ๐‘‹ ๐‘ก$ = โˆซ "#"$ "#"% ๐‘‘๐‘‹ ๐‘ก = โˆซ "#"$ "#"% ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก) + โˆซ "#"$ "#"% ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ยจ Knowing that all terms are well defined with the added : ยจ โˆซ "#"$ "#"% ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) = lim &โ†’( {โˆ‘)#* )#& ๐‘(๐‘ก), ๐‘‹(๐‘ก))). [๐‘Š(๐‘ก)+*) โˆ’ ๐‘Š(๐‘ก))]} ยจ Where the mesh or partition {๐‘ก)} is not completely pathological ยจ Also for STRATO: ยจ โˆซ "#"$ "#"% ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก) = lim &โ†’( {โˆ‘)#* )#& ๐‘“ ๐‘ก), [-("2)+-("234)] ! . [๐‘Š(๐‘ก)+*) โˆ’ ๐‘Š(๐‘ก))]} 93
  • 94. Luc_Faucheux_2021 The crux of the matter - IX ยจ So in the end, this is the only way to write a process that leaves no room for error: ยจ When writing something like this, we have no idea what it means ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘Š(๐‘ก) ยจ Really we should write: ยจ ๐‘‹ ๐‘ก% โˆ’ ๐‘‹ ๐‘ก$ = โˆซ "#"$ "#"% ๐‘‘๐‘‹ ๐‘ก = โˆซ "#"$ "#"% ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก) + โˆซ "#"$ "#"% ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ). ๐‘‘๐‘Š(๐‘ก) ยจ โˆซ "#"$ "#"% ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (๐›ผ). ๐‘‘๐‘Š(๐‘ก) = lim &โ†’( {โˆ‘)#* )#& ๐‘(๐‘ก), ๐‘€R[๐‘ก), ๐‘ก)+*]). [๐‘Š(๐‘ก)+*) โˆ’ ๐‘Š(๐‘ก))]} ยจ Where the function ๐‘€R[๐‘ก), ๐‘ก)+*] indicates where in the partition [๐‘ก), ๐‘ก)+*] we should take the value of the stochastic process ๐‘‹ ๐‘ก 94
  • 95. Luc_Faucheux_2021 The crux of the matter - X ยจ If ITO, ๐›ผ = 0, and ๐‘€R ๐‘ก), ๐‘ก)+* = ๐‘‹(๐‘ก)) ยจ If STRATO, ๐›ผ = 1/2, and ๐‘€R ๐‘ก), ๐‘ก)+* = [-("2)+-("234)] ! ยจ If anything in between, ๐‘€R ๐‘ก), ๐‘ก)+* = ๐‘‹(๐‘ก)) + ๐›ผ. [ ๐‘‹(๐‘ก)+* โˆ’ ๐‘‹(๐‘ก))] ยจ The nice thing about ITO is that the integral is a martingale, and that the integral also flows the isometry rule. ยจ The price to pay is that the usual rules of calculus (chain rule, Leibniz, Taylor expansion, integration by part,..) are pretty much out of the window ยจ The nice thing about STRATO is that the usual rules of calculus carry over FORMALLY in the same manner (careful, they are not the same, it is a formal relation), so that is nice ยจ However the integral is not a martingale and does not follow the isometry rule 95
  • 96. Luc_Faucheux_2021 A brief history of the Ito-Stratanovitch controversy 96
  • 97. Luc_Faucheux_2021 A brief history ยจ Because : ยจ 1) it is a nice way to apply our knowledge of stochastic calculus ยจ 2) you encounter it in textbooks ยจ 3) it is still super confusing at times, I know I am still confused ยจ It pays to explain why that came about. ยจ But first of all, letโ€™s go one more time over the fact that actually there is NO controversy ยจ By the way, if you meet anyone who tells you that they understand Ito versus Stratanovitch perfectly, that person is either a liar or Van Kampen 97
  • 98. Luc_Faucheux_2021 First of all there is no controversy 98
  • 99. Luc_Faucheux_2021 There is no controversy ยจ In order to make things less confusing that they could be, we are going to be a little literal on the notation ยจ Essentially, ยจ PDEs are well defined, and rely on Newtonian/Leibniz/regular calculus ยจ SDEs and SIEs for non-homogeneous diffusion coefficients (non-linear), or level depenedent volatilities, are NOT well defined, and you need to choose a value for ๐›ผ ยจ There is a correspondence between the different [๐›ผ] calculus 99
  • 100. Luc_Faucheux_2021 There is no controversy - II ยจ Ito is ๐›ผ = 0 ยจ Strato is ๐›ผ = 1/2 ยจ Because the factor (1/2) also shows up in Taylor expansion and Ito lemma, that is where a lot of the confusion comes from. BTW, Ito lemma is NOT a Taylor expansion, it formally looks like one and comes from using a Taylor expansion in calculating the integral as a limit of sum, but it is NOT a Taylor expansion ยจ So you cannot say (as sometimes it seems to be implied in textbooks). ยจ Hey in stochastic calculus, it is the same as regular calculus, just go up one more level in the Taylor expansion 100
  • 101. Luc_Faucheux_2021 There is no controversy - III ยจ An ITO SDE is of the form: ยจ ๐‘‘๐‘‹ ๐‘ก = ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ([). ๐‘‘๐‘Š(๐‘ก) ยจ This implies that the PDF follows the FORWARD ITO Kolmogorov PDE ยจ ,=(0,"|-.,".) ," = โˆ’ , ,0 ๐‘Ž ๐‘ฅ, ๐‘ก . ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹$, ๐‘ก$ โˆ’ , ,0 [ * ! . [๐‘(๐‘ฅ, ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹$, ๐‘ก$)] ยจ With simpler notation: ยจ An ITO SDE is of the form: ยจ ๐‘‘๐‘‹ = ๐‘Ž. ๐‘‘๐‘ก + ๐‘. ([). ๐‘‘๐‘Š ยจ This implies that the PDF follows the FORWARD ITO Kolmogorov PDE (Fokker-Planck) ยจ ,= ," = โˆ’ , ,0 ๐‘Ž. ๐‘ โˆ’ , ,0 [ * ! . [๐‘! . ๐‘] 101
  • 102. Luc_Faucheux_2021 There is no controversy - IV ยจ An STRATO SDE is of the form: ยจ ๐‘‘๐‘‹ ๐‘ก = @ ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก + A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . (โˆ˜). ๐‘‘๐‘Š(๐‘ก) ยจ This implies that the PDF follows the FORWARD STRATO Kolmogorov PDE ยจ !"($,&|(!,&!) !& = โˆ’ ! !( {G ๐‘Ž ๐‘ก, ๐‘‹ ๐‘ก + + , . B ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ! !( B ๐‘ ๐‘ก, ๐‘‹ ๐‘ก }. ๐‘ ๐‘ฅ, ๐‘ก ๐‘‹*, ๐‘ก* โˆ’ ! !( [ + , . [B ๐‘(๐‘‹ ๐‘ก , ๐‘ก), . ๐‘(๐‘ฅ, ๐‘ก|๐‘‹*, ๐‘ก*)] ยจ With simpler notation: ยจ An STRATO SDE is of the form: ยจ ๐‘‘๐‘‹ = @ ๐‘Ž. ๐‘‘๐‘ก + A ๐‘. (โˆ˜). ๐‘‘๐‘Š ยจ This implies that the PDF follows the FORWARD STRATO Kolmogorov PDE (Fokker-Planck) ยจ ,= ," = โˆ’ , ,0 {@ ๐‘Ž + * ! . A ๐‘. , ,0 A ๐‘}. ๐‘ โˆ’ , ,0 [ * ! . [A ๐‘! . ๐‘] 102
  • 103. Luc_Faucheux_2021 There is no controversy - V ยจ So and ITO SDE is: ยจ ๐‘‘๐‘‹ = ๐‘Ž. ๐‘‘๐‘ก + ๐‘. ([). ๐‘‘๐‘Š ยจ And corresponds to a diffusion equation for the PDF: ยจ ,= ," = โˆ’ , ,0 ๐‘Ž. ๐‘ โˆ’ , ,0 [ * ! . ๐‘! . ๐‘] ยจ A STRATO SDE is: ยจ ๐‘‘๐‘‹ = @ ๐‘Ž. ๐‘‘๐‘ก + A ๐‘. (โˆ˜). ๐‘‘๐‘Š ยจ And corresponds to a diffusion equation for the PDF: ยจ ,= ," = โˆ’ , ,0 {@ ๐‘Ž + * ! . A ๐‘. , ,0 A ๐‘}. ๐‘ โˆ’ , ,0 [ * ! . A ๐‘! . ๐‘] 103
  • 104. Luc_Faucheux_2021 There is no controversy - VI ยจ So if we want the same solution, meaning we want the 2 equations for the PDE to be equal: ยจ ,= ," = โˆ’ , ,0 ๐‘Ž. ๐‘ โˆ’ , ,0 [ * ! . ๐‘! . ๐‘] ยจ ,= ," = โˆ’ , ,0 {@ ๐‘Ž + * ! . A ๐‘. , ,0 A ๐‘}. ๐‘ โˆ’ , ,0 [ * ! . A ๐‘! . ๐‘] ยจ That means as we have seen before that we need to write: ยจ A ๐‘ = ๐‘ ยจ ๐‘Ž = @ ๐‘Ž + * ! . A ๐‘. , ,0 A ๐‘ ยจ That is the correspondence between the ITO and STRATO coefficients for the SDEs ยจ Pointing out the obvious one more time, when , ,0 A ๐‘ = , ,0 ๐‘ = 0, both calculus will return exactly the same SDE and PDE, the SDE is called linear 104
  • 105. Luc_Faucheux_2021 Back to a brief history of the Ito- Stratanovitch controversy (a physicistโ€™s point of view) 105
  • 106. Luc_Faucheux_2021 A brief history - II ยจ So if you are a physicist, you like to write diffusion equations like: ยจ ,= ," = โˆ’ , ,0 [๐ฝ/ + ๐ฝA] ยจ ๐ฝA is the diffusion current (hence the D notation) ยจ ๐ฝ/ is the drift current, coming usually from an external force (hence the F notation) ยจ In the absence of external force, physicist would expect the equilibrium solution to be uniform (meaning ๐‘ = ๐‘๐‘ก๐‘’), and so it makes sense for them to write the diffusion current as: ยจ ๐ฝA = โˆ’๐ท. ,= ,0 ยจ If the diffusion coefficient is a function of the position (and the physical system described is such that the diffusive process does not impact the long term equilibrium steady state derived from thermodynamics), physicists will still expect a uniform distribution and will still want the diffusion coefficient to be outside the derivative ยจ ๐ฝA = โˆ’๐ท(๐‘ฅ). ,= ,0 106
  • 107. Luc_Faucheux_2021 A brief history - III ยจ For sake of simplicity letโ€™s say that there is no external forcing right now, ๐ฝ/ = 0 ยจ So physicist in the presence of non-homogeneous diffusion would like the PDE to look like: ยจ ,= ," = โˆ’ , ,0 ๐ฝA = โˆ’ , ,0 โˆ’๐ท(๐‘ฅ). ,= ,0 ยจ Nothing wrong so far. ยจ The problem arose when we started using digital computers and started running Monte Carlo simulations ยจ This is fairly recent so we should not beat ourselves too much on the fact that we tripped on that one. ยจ Following the work of Bachelier, Bernouilli and such, we started modeling the stochastic process as: ยจ ๐›ฟ๐‘‹ ๐‘ก = ยฑ 2๐ท with the ยฑ being a random number generator (head or tail) 107
  • 108. Luc_Faucheux_2021 A brief history - IV ยจ ๐›ฟ๐‘‹ ๐‘ก = ยฑ 2๐ท ยจ Is essentially what we looked at in the firs deck, as well as the Bachelier deck 108 X i i+1 i-1
  • 109. Luc_Faucheux_2021 A brief history - V ยจ We know for a contant spacing (constant diffusion coefficient), the distribution is the usual binomial which converges to the Gaussian, as illustrated by the Galton machine. ยจ By the way Galton was born not even 200 years ago, so again we should go easy on ourselves 109
  • 110. Luc_Faucheux_2021 A brief history - VI ยจ Ok so so far so good. ยจ The issue is when we started simulating on a digital computer something like that: ยจ ๐›ฟ๐‘‹ ๐‘ก = ยฑ 2๐ท(๐‘‹ ๐‘ก ) ยจ Because, when we write something like the above, even without knowing it, we are in ITO calculus and we are writing really: ยจ ๐‘‘๐‘‹(๐‘ก) = 2๐ท(๐‘‹ ๐‘ก ). ([). ๐‘‘๐‘Š(๐‘ก) ยจ This will give a PDE: ยจ ,= ," = โˆ’ , ,0 ๐‘Ž. ๐‘ โˆ’ , ,0 [ * ! . ๐‘! . ๐‘] ยจ With: ยจ ๐‘Ž = 0 ยจ ๐‘ = 2๐ท(๐‘ฅ) 110
  • 111. Luc_Faucheux_2021 A brief history - VII ยจ ,= ," = โˆ’ , ,0 โˆ’ , ,0 [ * ! . 2๐ท(๐‘ฅ) ! . ๐‘] ยจ ,= ," = โˆ’ , ,0 โˆ’ , ,0 [๐ท(๐‘ฅ). ๐‘] ยจ The solution of that (subject to non-diverging boundary conditions) is: ยจ ๐‘ = 1/๐ท(๐‘ฅ) ยจ And NOT ๐‘ = ๐‘๐‘ก๐‘’, uniform distribution as expected. ยจ So that was the first sign that something was wrong. 111
  • 112. Luc_Faucheux_2021 A brief history - VIII ยจ Letโ€™s recap. ยจ You want to model: ยจ ,= ," = โˆ’ , ,0 ๐ฝA = โˆ’ , ,0 โˆ’๐ท(๐‘ฅ). ,= ,0 ยจ You do a discrete numerical simulation: ยจ ๐›ฟ๐‘‹ ๐‘ก = ยฑ 2๐ท(๐‘‹ ๐‘ก ) ยจ You get: ยจ ,= ," = โˆ’ , ,0 โˆ’ , ,0 [๐ท(๐‘ฅ). ๐‘] ยจ Not good, you do not get what you expected 112
  • 113. Luc_Faucheux_2021 A brief history - IX ยจ So at that point, if physicists in the 80s were as good on stochastic calculus as they should be, they should have said. ยจ Well yeah, a discrete simulation is by definition in the world of ITO calculus ยจ If I write: ๐›ฟ๐‘‹ ๐‘ก = ยฑ 2๐ท(๐‘‹ ๐‘ก ) ยจ I really am writing: ๐‘‘๐‘‹(๐‘ก) = 2๐ท(๐‘‹ ๐‘ก ). ([). ๐‘‘๐‘Š(๐‘ก) ยจ Which will give me: ,= ," = โˆ’ , ,0 โˆ’ , ,0 [๐ท(๐‘ฅ). ๐‘] ยจ But I want: ,= ," = โˆ’ , ,0 โˆ’๐ท(๐‘ฅ). ,= ,0 113
  • 114. Luc_Faucheux_2021 A brief history - X ยจ Since I want: ,= ," = โˆ’ , ,0 โˆ’๐ท(๐‘ฅ). ,= ,0 ยจ I can write: ยจ ,= ," = โˆ’ , ,0 โˆ’ , ,0 ๐ท ๐‘ฅ . ๐‘ = โˆ’ , ,0 โˆ’ , ,0 ๐ท ๐‘ฅ . ๐‘ โˆ’ , ,0 ๐‘ . ๐ท(๐‘ฅ) ยจ ,= ," = โˆ’ , ,0 โˆ’ , ,0 ๐ท ๐‘ฅ . ๐‘ โˆ’ ๐ท(๐‘ฅ). ,= ,0 ยจ That is of the form: ยจ ,= ," = โˆ’ , ,0 ๐‘Ž. ๐‘ โˆ’ , ,0 [ * ! . ๐‘! . ๐‘] ยจ Which in ITO will give the SDE ยจ ๐‘‘๐‘‹ = ๐‘Ž. ๐‘‘๐‘ก + ๐‘. ([). ๐‘‘๐‘Š ยจ With: ๐‘Ž = โˆ’ , ,0 ๐ท ๐‘ฅ and ๐‘ = 2๐ท(๐‘‹ ๐‘ก ) 114
  • 115. Luc_Faucheux_2021 A brief history - XI ยจ Soooโ€ฆ bear with me here a few more slides.. ยจ If I write: ยจ ๐‘‘๐‘‹ = { , ,0 ๐ท ๐‘ฅ }. ๐‘‘๐‘ก + 2๐ท(๐‘‹ ๐‘ก ). ([). ๐‘‘๐‘Š ยจ Numerically I will simulate it as: ยจ ๐›ฟ๐‘‹ ๐‘ก = , ,0 ๐ท ๐‘ฅ ยฑ 2๐ท(๐‘‹ ๐‘ก ) ยจ That is assuming units of time equal to 1, if not you write ยจ ๐›ฟ๐‘‹ ๐‘ก = , ,0 ๐ท ๐‘ฅ . ๐›ฟ๐‘ก ยฑ 2๐ท ๐‘‹ ๐‘ก . ๐›ฟ๐‘ก 115
  • 116. Luc_Faucheux_2021 A brief history - XII ยจ I simulate: ๐›ฟ๐‘‹ ๐‘ก = , ,0 ๐ท ๐‘ฅ . ๐›ฟ๐‘ก ยฑ 2๐ท ๐‘‹ ๐‘ก . ๐›ฟ๐‘ก ยจ That is an ITO SDE: ๐‘‘๐‘‹ = { , ,0 ๐ท ๐‘ฅ }. ๐‘‘๐‘ก + 2๐ท(๐‘‹ ๐‘ก ). ([). ๐‘‘๐‘Š ยจ That has a PDE solution: ,= ," = โˆ’ , ,0 ๐‘Ž. ๐‘ โˆ’ , ,0 [ * ! . ๐‘! . ๐‘] ยจ With: ๐‘Ž = , ,0 ๐ท ๐‘ฅ and ๐‘ = 2๐ท(๐‘‹ ๐‘ก ) ยจ So you get the PDE: ,= ," = โˆ’ , ,0 , ,0 ๐ท ๐‘ฅ . ๐‘ โˆ’ , ,0 [๐ท. ๐‘] ยจ And: ยจ ,= ," = โˆ’ , ,0 , ,0 ๐ท ๐‘ฅ . ๐‘ โˆ’ , ,0 ๐ท. ๐‘ = โˆ’ , ,0 ,A ,0 . ๐‘ โˆ’ ,A ,0 . ๐‘ โˆ’ ๐ท. ,= ,0 = โˆ’ , ,0 โˆ’๐ท. ,= ,0 ยจ Which is what I want 116
  • 117. Luc_Faucheux_2021 A brief history - XIII ยจ SOโ€ฆ ยจ If I want to simulate: ,= ," = โˆ’ , ,0 โˆ’๐ท(๐‘ฅ). ,= ,0 ยจ Which has no drift term, and has the equilibrium long-term steady state uniform distribution as the desired solution, I need to simulate on a discrete digital computer: ยจ ๐›ฟ๐‘‹ ๐‘ก = , ,0 ๐ท ๐‘ฅ . ๐›ฟ๐‘ก ยฑ 2๐ท ๐‘‹ ๐‘ก . ๐›ฟ๐‘ก ยจ Which seems weird, because the first term , ,0 ๐ท ๐‘ฅ . ๐›ฟ๐‘ก looks like the drift term. ยจ This is when the whole confusion started 117
  • 118. Luc_Faucheux_2021 A brief history - XIV ยจ ๐›ฟ๐‘‹ ๐‘ก = , ,0 ๐ท ๐‘ฅ . ๐›ฟ๐‘ก ยฑ 2๐ท ๐‘‹ ๐‘ก . ๐›ฟ๐‘ก ยจ The term , ,0 ๐ท ๐‘ฅ . ๐›ฟ๐‘ก looks like indeed what you would simulate if you were to simulate a drift ยจ It is NOT a physical drift ยจ Your PDE that you wanted to simulate was ,= ," = โˆ’ , ,0 โˆ’๐ท. ,= ,0 and had NO drift ยจ It is purely a term that you need to add to the simulation to recover your PDE ยจ This is why it was called a โ€œspurious driftโ€ (Ryter 1980). ยจ Again, it is NOT a drift, you are not extraction motion out of noise, you need to have it in there because you are using Ito calculus, and plain and simple: ยจ ๐‘‘๐‘‹ = ๐‘Ž. ๐‘‘๐‘ก + ๐‘. ([). ๐‘‘๐‘Š will give you ,= ," = โˆ’ , ,0 ๐‘Ž. ๐‘ โˆ’ , ,0 [ * ! . ๐‘! . ๐‘] ยจ Period..end of storyโ€ฆ 118
  • 119. Luc_Faucheux_2021 A brief history - XV ยจ But you can see why that was confusing, and why it tripped at the time a lot of people, yours truly included. ยจ And that was confusing for 2 main reasons: ยจ Reason 1: Maxwell demon and Thermal ratchets ยจ Reason 2: Stratanovitch ยจ They are somewhat related but distinct. ยจ Letโ€™s go over Stratanovitch first 119
  • 120. Luc_Faucheux_2021 A brief history โ€“ XVI ยจ To be fair, I was also a little harsh and oversimplifying. ยจ There is also another reason why the question of Ito versus Stratanovitch came about, it has to do with the autocorrelation of the random forcing (the noise). Is it white noise with 0 memory, and so Ito would be appropriate, or is it coloured noise, with some non zero autocorrelation function, and in this case Strato would be more appropriate. ยจ Again it really depends on the physical system, an what PDE you want ยจ Radioactive particle decay lends itself well to Ito ยจ Diffusion in a viscous medium lends itself better to Strato from a physics point of view ยจ So as always things are a little more complicated that I would make them out to be, so you should take what I say with a grain of salt. 120
  • 121. Luc_Faucheux_2021 A brief history โ€“ XVII ยจ In particular, if there is any kind of correlation, then an non-ITO approach is sometimes justified. ยจ Remember that we had from the earlier part of the deck: ยจ < A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . [๐›ผ] . ๐‘‘๐‘Š > = ๐›ผ. A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก ยจ In particular: ยจ ITO : < A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . [ . ๐‘‘๐‘Š > = 0 ยจ STRATO: < A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . โˆ˜ . ๐‘‘๐‘Š > = [ * ! ]. A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . , ,- A ๐‘ ๐‘ก, ๐‘‹ ๐‘ก . ๐‘‘๐‘ก 121
  • 122. Luc_Faucheux_2021 Reason #2 for the confusion: People started to hear about Stratanovitch calculus 122
  • 123. Luc_Faucheux_2021 Reason #2 for the confusion - Stratanovitch ยจ So first of all ITO calculus is weird because of ITO lemma, and the fact that the usual rules of calculus (Leibniz rule, chain rule) are not formally respected ยจ Second ITO is very well adapted to DISCRETE processes in time (finance, but also simulations on digital computers) ยจ Physicists do not like to learn new rules of calculus, they would rather stick to the usual one ยจ More seriously, in physics, it could be argued that VERY FEW processes are truly discrete (radioactive decay for example being one of them), but most of the processes (like diffusion), are not discrete but continuous 123
  • 124. Luc_Faucheux_2021 Reason #2 for the confusion โ€“ Stratanovitch - II ยจ We know that using STRATO calculus: ยจ ๐‘‘๐‘‹ = @ ๐‘Ž. ๐‘‘๐‘ก + A ๐‘. (โˆ˜). ๐‘‘๐‘Š ยจ Corresponds to a diffusion equation for the PDF: ยจ ,= ," = โˆ’ , ,0 {@ ๐‘Ž + * ! . A ๐‘. , ,0 A ๐‘}. ๐‘ โˆ’ , ,0 [ * ! . A ๐‘! . ๐‘] ยจ We can expand the last term into: ยจ , ,0 * ! . A ๐‘! . ๐‘ = , ,0 * ! . A ๐‘. A ๐‘. ๐‘ = * ! . , ,0 A ๐‘. A ๐‘. ๐‘ = * ! . , ,0 A ๐‘. A ๐‘. ๐‘ ยจ , ,0 * ! . A ๐‘! . ๐‘ = * ! . A ๐‘. ๐‘ . , ,0 A ๐‘ + * ! . A ๐‘. , ,0 A ๐‘. ๐‘ ยจ , ,0 * ! . A ๐‘! . ๐‘ = * ! . A ๐‘. ,B % ,0 . ๐‘ + * ! . A ๐‘. , ,0 A ๐‘. ๐‘ 124
  • 125. Luc_Faucheux_2021 Reason #2 for the confusion โ€“ Stratanovitch - III ยจ ,= ," = โˆ’ , ,0 {@ ๐‘Ž + * ! . A ๐‘. , ,0 A ๐‘}. ๐‘ โˆ’ , ,0 [ * ! . A ๐‘! . ๐‘] ยจ ,= ," = โˆ’ , ,0 @ ๐‘Ž + * ! . A ๐‘. , ,0 A ๐‘ . ๐‘ โˆ’ { * ! . A ๐‘. ,B % ,0 . ๐‘ + * ! . A ๐‘. , ,0 A ๐‘. ๐‘ } ยจ ,= ," = โˆ’ , ,0 @ ๐‘Ž . ๐‘ โˆ’ * ! . A ๐‘. , ,0 A ๐‘. ๐‘ ยจ So a STRATO SDE of the form: ยจ ๐‘‘๐‘‹ = @ ๐‘Ž. ๐‘‘๐‘ก + A ๐‘. (โˆ˜). ๐‘‘๐‘Š ยจ Will give a PDE of the form: ยจ ,= ," = โˆ’ , ,0 @ ๐‘Ž . ๐‘ โˆ’ * ! . A ๐‘. , ,0 A ๐‘. ๐‘ 125
  • 126. Luc_Faucheux_2021 Reason #2 for the confusion โ€“ Stratanovitch - IV ยจ Again in order to illustrate letโ€™s assume no drift term in the PDE ยจ So a STRATO SDE of the form: ยจ ๐‘‘๐‘‹ = A ๐‘. (โˆ˜). ๐‘‘๐‘Š ยจ Will give a PDE of the form: ยจ ,= ," = โˆ’ , ,0 โˆ’ * ! . A ๐‘. , ,0 A ๐‘. ๐‘ ยจ Again, nothing wrong there, but physicist got confused because they starting saying things like the next slide: 126
  • 127. Luc_Faucheux_2021 Reason #2 for the confusion โ€“ Stratanovitch - V ยจ If I write: ๐‘‘๐‘‹ = A ๐‘. (โˆ˜). ๐‘‘๐‘Š ยจ On a computer I simulate ๐›ฟ๐‘‹ = ยฑA ๐‘ ยจ And that gives me a PDE with no drift: ,= ," = โˆ’ , ,0 โˆ’ * ! . A ๐‘. , ,0 A ๐‘. ๐‘ ยจ So I am happy, but now my diffusion current is no longer: ยจ ,= ," = โˆ’ , ,0 ๐ฝA = โˆ’ , ,0 โˆ’๐ท(๐‘ฅ). ,= ,0 with A ๐‘ = 2๐ท ยจ But: ยจ ,= ," = โˆ’ , ,0 โˆ’ * ! . A ๐‘. , ,0 A ๐‘. ๐‘ ยจ So my diffusion current is really: ๐ฝA = โˆ’ * ! . A ๐‘. , ,0 A ๐‘. ๐‘ = โˆ’ ๐ท. , ,0 ๐ท. ๐‘ 127
  • 128. Luc_Faucheux_2021 Reason #2 for the confusion โ€“ Stratanovitch - VI ยจ So if I do not want to simulate a drit term (because my PDE has no drift term), then really I need to write the diffusion current as: ยจ ๐ฝA = โˆ’ ๐ท. , ,0 ๐ท. ๐‘ ยจ But thenโ€ฆwait a minute, if I plug this back into the PDE, then my equilibrium distribution is no longer uniform (in fact it will be ๐ท. ๐‘ = ๐‘๐‘ก๐‘’ or ๐‘ = 1/ ๐ท. ) ยจ So when the diffusion coefficient is non-homogeneous it will change my equilibrium distribution. ยจ I have โ€œextractedโ€ motion our of random noise ยจ This is maybe an example of the Maxwell demon ยจ I have broken the law of thermodynamics, so what is next ? Perpetual motion ? 128
  • 129. Luc_Faucheux_2021 Reason #2 for the confusion โ€“ Stratanovitch - VII ยจ The previous 2 slides are obviously wrong but at the time that was a real debate in Physics ยจ For a constant diffusion the diffusion current is : ยจ ๐ฝA = โˆ’ ๐ท. , ,0 ๐ท. ๐‘ = โˆ’๐ท. , ,0 ๐‘ = โˆ’ , ,0 ๐ท. ๐‘ ยจ Because who cares, ๐ท is a constant ยจ But now if we have ๐ท(๐‘‹(๐‘ก)) for a Brownian particle, which one is the real diffusion current? ยจ And then there were even more confusion about the Ryter spurious drift maybe being an actual drift because ยจ , ,0 ๐ท. ๐‘ = , ,0 ๐ท . ๐‘ + ๐ท. , ,0 ๐‘ ยจ So maybe ๐ท. , ,0 ๐‘ is the โ€œrealโ€ diffusion current and the term , ,0 ๐ท . ๐‘ is the Maxwell demon term extracting directed motion out of a purely random diffusion process? 129
  • 130. Luc_Faucheux_2021 Reason #2 for the confusion โ€“ Stratanovitch - VIII ยจ The whole controversy could have been squashed had there been one good math guy who understood Ito and Strato well enough to explain it simply to physicist and say: ยจ Listen you idiots, all that really matters is the PDE ยจ If you want to play with a computer and burn some CPUs and start global warming by writing something like: ๐›ฟ๐‘‹ ๐‘ก = ยฑ 2๐ท(๐‘‹ ๐‘ก ) ยจ That is fine by me, but you are by definition using ITO calculus and writing: ยจ ๐‘‘๐‘‹(๐‘ก) = 2๐ท(๐‘‹ ๐‘ก ). ([). ๐‘‘๐‘Š(๐‘ก) ยจ Which will give you: ,= ," = โˆ’ , ,0 โˆ’ , ,0 [๐ท(๐‘ฅ). ๐‘] ยจ Which will not give you a uniform distribution for the steady state regime. ยจ Period, full stop, end of story. 130
  • 131. Luc_Faucheux_2021 Reason #2 for the confusion โ€“ Stratanovitch - IX ยจ And then this math guy could have added: ยจ If you idiots want to simulate the SDE that corresponds to the PDE: ,= ," = โˆ’ , ,0 โˆ’๐ท(๐‘ฅ). ,= ,0 ยจ And if you insist on using discrete numerical simulations, then sorry mate but you have no choice but to write: ยจ ๐›ฟ๐‘‹ ๐‘ก = , ,0 ๐ท ๐‘ฅ ยฑ 2๐ท ๐‘‹ ๐‘ก ยจ So yeah, the first term looks like a drift term, but there is no physical drift in your PDE, it is just a โ€œspuriousโ€ drift because you now have entered the mysterious world of stochastic calculus, and more precisely ITO calculus. And please leave at the door all the intuitions you had built on Newtonian/Leibniz/regular calculus, and buckle up buttercup, because you not in Kansas anymore Toto. ยจ Sometimes I wish math guys could not talk to physicists in such condescending manner. 131
  • 132. Luc_Faucheux_2021 Reason #1 for the confusion: Maxwell demon, Brownian motors, Thermal ratchets 132
  • 133. Luc_Faucheux_2021 Maxwell demon - I ยจ The other reason why there was some confusion is that at the same time a lot of physicists were looking at issues like Maxwell demon, Brownian engines, Brownian ratchets, Thermal ratchetsโ€ฆ ยจ The Maxwell demon is a possible (but very improbable) entity that drive a system towards a state that is possible but very improbable ยจ To quote the adventures of Mr. Tompkins, it is possible (but very unlikely) the ice cube that you melted in your coke will spontaneously re-form as an ice cube and that the surrounding liquid will warm back up again. ยจ That is possible, there is nothing in Physics that says that it is impossible, it is just very unlikely ยจ Winning the lottery is also possible, it is just also highly unlikely 133
  • 134. Luc_Faucheux_2021 Maxwell demon - II ยจ The Maxwell demon is intriguing because the 2nd Law of Thermodynamics says that usually things tend to decay, and that a measure of disorder that you can define (entropy) usually always increase at the end. ยจ That law makes sense, we all see it in our lives everyday, things always tend to go to sh.., never (or highly unlikely) the other way around ยจ What is true for our personal lives is true for the Universe ยจ Nothing shocking there, since we are part of the universe, so the laws should be somewhat universal 134
  • 135. Luc_Faucheux_2021 Maxwell demon - III ยจ The Maxwell demon putting the moves on Mr. Tompkins fiancรฉe and trying to impress her with his tennis skills by reconstructing an ice cube without violating the 2nd law of Thermodynamics 135
  • 136. Luc_Faucheux_2021 Maxwell demon - IV ยจ A more common representation of the demon 136
  • 137. Luc_Faucheux_2021 Brownian Engines ยจ So it would make sense for physicists to ask if it was possible to build a Maxwell demon. ยจ Feynman in his lecture has a great chapter on this. ยจ He even offered a prize to the first person / team who could build one. 137
  • 138. Luc_Faucheux_2021 Brownian engines - II ยจ The canonical example of a thermal ratchet from the Feynmanโ€™s lectures in Physics 138
  • 139. Luc_Faucheux_2021 Brownian engines - III ยจ There was also another reason why physicists were excited by all this, (none withstanding the fact that if you could receive a check from Richard Feynman, that would be the coolest thing ever, and you would have so much street cred), was that we were starting to look in real-time in-vivo inside the cells and human body, and whether we like it or not (and again nothing really to do with intelligent design theory), there were some really cool things happening there. ยจ Vesicles are transported within the cell, to the place where they should be, in a very highly โ€œnoisyโ€ (read bombarded by random fluctuations) environment that should make it impossible. Yet it works. ยจ Spermatozoides are able to โ€œswimโ€ to where they were supposed to go, again in a very random and viscous environment. ยจ Trying to build a typical inertial engine and computing how much energy that required lead to the conclusion that it was impossible, so clearly there was something akin to a Thermal ratchet, or Brownian engine at work there. So somehow Nature found a way to extract directed motion out of random noise, or at least found a way to still be able to build things in a highly noisy, random and viscous regime without โ€œforcingโ€ its way through the way an inertial engine (think of a car or a rocket) would do at the scale we are used to 139
  • 140. Luc_Faucheux_2021 Brownian engines - IV ยจ I was lucky enough to work at the time in the lab of Albert Libchaber, who was fascinated with those problems. ยจ I have never met in my life someone who was so pure in his search of the truth and who was so keen on lifting the curtains, and who had the intellectual power to cut through the math, or the obstacles between him and the hidden simple principles at work ยจ I count myself as being fortunate to have been able to witness him at work, and in his life. 140
  • 141. Luc_Faucheux_2021 Brownian engines - V ยจ I could write a book about how much fun we had in his lab (and maybe I should), but suffice to say that we were all involved in the subject of Brownian engines, whether on the biology side or on the more Physics side. If you guys are interested you have to read this one: 141
  • 142. Luc_Faucheux_2021 Brownian engines - VI ยจ So anyways long story short, I spent a lot of time then on those issues, had the pleasure of working with some of the finest minds of the time (too numerous to name all, and am not even going to try, but of course Albert Libchaber, Mike Shelley, Dave Muraki, Marcello Magnasco, Erez Braun, Elisha Moses, Drew Belmonte, Deborah Fygenson, Albrecht Ott, Andreas Tilgner, Gustavo Stolovitsky, Rolf Landauer and is famous shout โ€œcomplex is complex is complex!โ€, and again way too many I apologize for not including in the short list above) ยจ But anyways we built the first optical thermal ratchet ! 142
  • 143. Luc_Faucheux_2021 Brownian engines - VII ยจ It turns out that there is a cool connection to Finance. ยจ If you replace the position by the accumulated PL, the thermal ratchet becomes what is known as the Parrondo Paradox. ยจ If we just had thought about replacing X by $, that could have been known as the Libchaber paradox ยจ Another example of the work of an illustrious Frenchman that is easily applicable to finance but somehow someone else takes it through the goal lineโ€ฆ ยจ Essentially, just like the thermal ratchet, you can โ€œextractโ€ some direction our of random fluctuations (no worries about the 2nd law, it is still intact, because you need to dump energy into the ratchet to make it work) ยจ So in Finance, you could have 2 losing strategies (2 loser PM in a fund), but if you switch randomly the allocation of $ between those 2 strategies, you can find a regime where the resulting strategy will make $ on average 143
  • 144. Luc_Faucheux_2021 Brownian engines - VIII ยจ There are actually applications in Finance of the Maxell demon, known as the Parrondo paradox. ยจ Two trading strategies (PM at a hedge fund) on average lose money (B and C, blue and green line) ยจ However you can alternate between the two strategies to create one (A-red line) that on average will be profitable, like the thermal ratchet who is extracting work out of thermal noise, this Parrondo construct extract positive return out of random switches between two losing strategies 144
  • 145. Luc_Faucheux_2021 Another draft of retelling the story of the Ito- Stratanovitch controversy 145
  • 146. Luc_Faucheux_2021 Another way to tell the story โ€“ I ยจ This was a first draft of the story, it was a little more complicated than the one I went through, but thought I would keep it here for sake of completeness. ยจ No need to read it ยจ It has nice pictures though ยจ Please enjoy 146
  • 147. Luc_Faucheux_2021 Another way to tell the story - II ยจ All right, this is where all the confusion came about the Ito-Stratanovitch controversy (and also that physicists tend to glance over stochastic calculus and be quite liberal and not that mathematically rigorous, I have to admit) ยจ Suppose that we start with a PDE describing the following process: ยจ ,=(0,") ," = โˆ’ , ,- ๐‘Ž ๐‘‹ ๐‘ก , ๐‘ก . ๐‘(๐‘ฅ, ๐‘ก) โˆ’ , ,- [ * ! . [๐‘(๐‘‹ ๐‘ก , ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก)] ยจ , ," . ๐‘ ๐‘ฅ, ๐‘ก = โˆ’ , ,0 [๐ฝ/+๐ฝA] ยจ ๐ฝ/ ๐‘ฅ, ๐‘ก = ๐‘Ž(๐‘ก). ๐‘ ๐‘ฅ, ๐‘ก and ๐ฝA ๐‘ฅ, ๐‘ก = โˆ’ , ,- [ * ! . [๐‘(๐‘‹ ๐‘ก , ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก)] ยจ ๐ฝA ๐‘ฅ, ๐‘ก = โˆ’ , ,- [ * ! . [๐œŽ(๐‘‹ ๐‘ก , ๐‘ก)! . ๐‘(๐‘ฅ, ๐‘ก)] ยจ ๐ฝA ๐‘ฅ, ๐‘ก = โˆ’ , ,- [๐ท(๐‘ฅ, ๐‘ก). ๐‘(๐‘ฅ, ๐‘ก)] with ๐ท ๐‘ฅ, ๐‘ก = * ! . [๐‘(๐‘‹ ๐‘ก , ๐‘ก)! ] 147