2. Luc_Faucheux_2021
That deck
2
š Could have been named âEverything that you ever wanted to know about Ho-Lee but were
too afraid to askâ
š Using Ho Lee as a working example to introduce a lot of concepts, mostly within the HJM
framework, but also illustrating some of the properties of the affine models
š Reached 280 slides and 82M on part V-a, so I had to split it into two sections
š Apologies for that.
š This is the second part
š It also answers the question: âhow many slides does it take Luc to beat up a dead horse?â
8. Luc_Faucheux_2021
Summary, using the SDE formalism - II
š Remember that SDEs are ferocious beasts not to be dealt with lightly, so always better to
either use the SIE formalism or the exact solution.
8
11. Luc_Faucheux_2021
A useful relationship - II
š The Langevin equation is quite commonly used when modeling interest rates.
š Since interest rates are the âspeedâ or âvelocityâ of the Money Market Numeraire, it is quite
natural to have thought about using the Langevin equation which represents the âspeedâ of
a Brownian particle.
š As a result, a number of quantities in Finance are related to the exponential of the integral
over time of the short-term rate (instantaneous spot rate)
š For example (Fabio Mercurio p. 3), the stochastic discount factor ð·(ð¡, ð) between two time
instants ð¡ and ð is the amount at time ð¡ that is âequivalentâ to one unit of currency payable
at time ð, and is equal to
š ð· ð¡, ð =
2(&)
2(4)
= exp(â â«&
4
ð ð . ðð )
š The Bank account (Money-market account) is such that:
š ððµ ð¡ = ð ð¡ . ðµ ð¡ . ðð¡ with ðµ ð¡ = 0 = 1
š ðµ ð¡ = exp(â«5
&
ð ð . ðð )
11
12. Luc_Faucheux_2021
A useful relationship - III
š Since we will most likely be looking at equations like:
š ðð ð¡ = âðð(ð¡). ðð¡ + ð. ðð
š ðð ð¡ = âðð(ð¡). ðð¡ + ð. ðð
š Which has a formal solution:
š ð ð¡6 = exp âðð¡6 . {exp ðð¡7 . ð ð¡7 + â«&8&7
&8&6
exp ðð¡ . ð. ([). ðð ð¡ }
š ð ð¡6 = exp âðð¡6 . {exp ðð¡7 . ð ð¡7 + â«&8&7
&8&6
exp ðð¡ . ð. ([). ðð ð¡ }
š So it looks like we will be taking exponentials of integrals of the Wiener process times a
function of time
12
13. Luc_Faucheux_2021
A useful relationship - IV
š We had in the simple case: ðŒ ð"1 & = ð
/!
!
&
š We would like to find a relation between:
š ðŒ exp[â«5
&
ð ð . ðð(ð )]
š And
š exp[â«5
& $
#
ð ð #. ðð ]
š Because that seems to work for ð ð = ð
š ðŒ exp[â«5
&
ð ð . ðð(ð )] = ðŒ exp[â«5
&
ð. ðð(ð )] = ðŒ exp[ð. ð ð¡ â ð 0 ] = ðŒ ð"1 &
š exp â«5
& $
#
ð ð #. ðð = exp â«5
& $
#
ð#. ðð = exp
$
#
ð#
â«5
&
ðð = exp
$
#
ð# ð¡ = ð
/!
!
&
13
14. Luc_Faucheux_2021
A useful relationship - V
š Letâs define:
š ð ð¡ = â«5
&
ð ð . ðð(ð ) â â«5
& $
#
ð ð #. ðð
š And : ð ð¡ = exp(ð ð¡ )
š Applying Ito lemma:
š ð ð ð¡6 â ð ð ð¡7 = â«&8&7
&8&6 -9
-:
. ([). ðð(ð¡) + â«&8&7
&8&6 $
#
.
-!9
-;! . ([). (ð¿ð)#
š In the âlimitâ of small [me increments, this can be wrien formally as the Ito lemma:
š ð¿ð =
-9
-;
. ([). ð¿ð +
$
#
.
-!9
-;! . (ð¿ð)#
š To the function: ð ð¥ = ð ð¡ = ð ð¡ = exp(ð¥ = ð ð¡ )
14
17. Luc_Faucheux_2021
A useful relationship - VIII
š Making the steps explicit here in order to convince ourselves of the validity (also when we
will want to expand this to a function ð ð ð , ð )
š ðŒ exp(â«5
&
ð ð . ðð(ð ) â â«5
& $
#
ð ð #. ðð ) = 1
š ðŒ exp â«5
&
ð ð . ðð ð . exp[â â«5
& $
#
ð ð #. ðð )] = 1
š exp[â â«5
& $
#
ð ð #. ðð )]. ðŒ exp â«5
&
ð ð . ðð ð = 1
š ðŒ exp â«5
&
ð ð . ðð ð = exp[â«5
& $
#
ð ð #. ðð )]
š This is indeed a beautiful relationship, and will be quite useful when dealing with interest-
rates modeling.
17
18. Luc_Faucheux_2021
A useful relationship - IX
š Under something called the âNovikov conditionâ (essentially none of the quantities diverge
to infinity and everything is well behaved)
š It can be shown that:
š exp(â«5
&
ð ð . ðð(ð ) â â«5
& $
#
ð ð #. ðð ) is a martingale
š Sometimes the above function is referred to the Doleans-Dade exponential in memory of
Catherine Doleans-Dade
š â° â«5
&
ð ð . ðð ð = exp(â«5
&
ð ð . ðð(ð ) â â«5
& $
#
ð ð #. ðð )
š In Finance it is related to the Radon-Nykodym derivative in the Girsanov theorem (in the
deck on numeraire and numeraire change)
š One moves from the historical measure â to the risk neutral measure â using the Radon-
Nykodym derivative:
š
=â
=â
= â° â«5
&
(
@ A )B
"
). ðð ð
18
19. Luc_Faucheux_2021
A useful relationship - X
š
=â
=â
= â° â«5
&
(
@ A )B
"
). ðð ð
š Where ð ð is the instantaneous risk-free rate, ð the asset drift and ð its volatility
š Just wanted to mention this here as we will see it in the deck on numeraire.
š Novikov is also very famous in financial markets as it is the name of a very fancy restaurant
in the Mayfair section of London always crowded with hedge fund managers and beautiful
models (not the interest rates model type). You might also get a âNovikov conditionâ
hanging there too long, but it is not related to the one we just mentionedâŠ.
19
20. Luc_Faucheux_2021
A useful relationship - XI
š In any case, for a function of time only, we have shown a very useful relationship when
dealing with interest rate models:
š ðŒ exp â«A85
A8&
ð ð . ðð ð = exp[â«A85
A8& $
#
ð ð #. ðð ]
20
21. Luc_Faucheux_2021
Letâs play a game.
Letâs see if you can spot the mistakes in the
next section
(*) Gilles Franchini found them under 2 minutes
21
23. Luc_Faucheux_2021
Another useful relationship for a Gaussian process
š From the Bachelier deck in the case of a Gaussian process we had the following:
š For the regular Gaussian â ð¥, ð¡ =
$
#C"!&
. exp(
);!
#"!&
) we have ðŒ =
$
#"!&
š ðŒD = â«)E
FE
ð)D;!
. ðð¥ =
C
D
š ðD ð¥ =
$
G0
. ð)D;!
is the normalized probability distribution
š < ð¥H > = â«)E
FE
ð¥H. ðD ð¥ . ðð¥
š < ð¥#H > = ðŒ)H. â(8$
(8H
(
$
#
+ ð â 1)
š < ð¥#HF$ > = 0
23
24. Luc_Faucheux_2021
Another useful relationship for a Gaussian process - II
š A couple of side notes first on ðŒD = â«)E
FE
ð)D;!
. ðð¥ =
C
D
š ðD ð¥ =
$
G0
. ð)D;!
is the normalized probability distribution
š < ð¥ > = â«)E
FE
ð¥. ðD ð¥ . ðð¥ = 0 because ð¥. ðD ð¥ is an odd function
š < ð¥H > = â«)E
FE
ð¥H. ðD ð¥ . ðð¥ = 0 because (ð¥H. ðD ð¥ ) is an odd function if ð is odd
24
25. Luc_Faucheux_2021
Kolmogorov equation: Bachelier thesis (page 35) IX
š ð ð¥, ð¡ =
I
&
. exp{â
CI!;!
&
} is already normalized
š We still need to solve for the value of ð»
š A couple of side notes first on ðŒD = â«)E
FE
ð)D;!
. ðð¥ =
C
D
š ðD ð¥ =
$
G0
. ð)D;!
is the normalized probability distribution
š < ð¥ > = â«)E
FE
ð¥. ðD ð¥ . ðð¥ = 0 because ð¥. ðD ð¥ is an odd function
š < ð¥H > = â«)E
FE
ð¥H. ðD ð¥ . ðð¥ = 0 because (ð¥H. ðD ð¥ ) is an odd function if ð is odd
25
26. Luc_Faucheux_2021
Kolmogorov equation: Bachelier thesis (page 35) X
š < ð¥# > = â«)E
FE
ð¥#. ðD ð¥ . ðð¥ =
C
D
. â«)E
FE
ð¥#. ð)D;!
. ðð¥
š Now:
-
-D
ð)D;!
= âð¥# ð)D;!
š So: < ð¥# > =
C
D
. â«)E
FE
ð¥#. ð)D;!
. ðð¥ =
C
D
. â«)E
FE )-
-D
ð)D;!
. ðð¥ =
C
D
.
)-
-D
[â«)E
FE
ð)D;!
. ðð¥]
š A little more formally:
š < ð¥# > =
)$
G0
.
-G0
-D
š Replacing ðŒD =
C
D
, we get < ð¥# > =
)$
1
0
.
-
1
0
-D
= â ðŒ.
-
-D
ðŒ â23
! =
$
#
. ðŒ â3
!. ðŒ â24
! =
$
#D
26
28. Luc_Faucheux_2021
Kolmogorov equation: Bachelier thesis (page 35) XII
š ðŒD = â«)E
FE
ð)D;!
. ðð¥ =
C
D
š ðD ð¥ =
$
G0
. ð)D;!
is the normalized probability distribution
š < ð¥H > = â«)E
FE
ð¥H. ðD ð¥ . ðð¥
š < ð¥#H > = ðŒ)H. â(8$
(8H
(
$
#
+ ð â 1)
š < ð¥#HF$ > = 0
š We will also look like Bachelier did at the positive part of the price distribution
š < (ð¥H|ð¥ > 0) > = â«5
FE
ð¥H. ðD ð¥ . ðð¥
28
29. Luc_Faucheux_2021
Kolmogorov equation: Bachelier thesis (page 35) XIII
š < ð¥#H > = ðŒ)H. â(8$
(8H
(
$
#
+ ð â 1)
š For the regular Gaussian â ð¥, ð¡ =
$
#C"!&
. exp(
);!
#"!&
) we have ðŒ =
$
#"!&
š A somewhat useful notation:
š ð! = â(8$
(8H
ð is the usual factorial
š ð!! = â(8$
(8H
ð is called the âdouble factorialâ and only includes in the product the terms that
have the SAME parity as ð
š In our specific case we can rewrite â(8$
(8H
(
$
#
+ ð â 1) as:
š â(8$
(8H
(
$
#
+ ð â 1) = â(8$
(8H
(
#()$
#
) = 2)H â(8$
(8H
(2ð â 1) = 2)H. 2ð â 1 âŒ
š < ð¥#H > = ðŒ)H. 2)H. 2ð â 1 âŒ
29
30. Luc_Faucheux_2021
Kolmogorov equation: Bachelier thesis (page 35) XIV
š < ð¥#H > = ðŒ)H. 2)H. 2ð â 1 âŒ
š In the case of the Gaussian, ðŒ =
$
#"!&
, so < ð¥#H > = (2ð# ð¡)H. 2)H. 2ð â 1 âŒ
š So : < ð¥#H > = (ð# ð¡)H. 2ð â 1 ⌠and < ð¥#HF$ > = 0
š Another cute way to express it is the following:
š < ð¥K > = (ð ð¡)K. ð â 1 ⌠if ð is even, 0 otherwise
š This is quite compact and beautiful
30
31. Luc_Faucheux_2021
Another useful relationship for a Gaussian process - III
š We put ourselves in the Gaussian case with 0 drift
š â ð¥, ð¡ =
$
#C"!&
. exp(
);!
#"!&
) we have ðŒ =
$
#"!&
š ðŒ ð = â«)E
FE
â ð¥, ð¡ . ð¥ . ðð¥ = 0 because â ð¥, ð¡ is an even function of ð¥
š ðŒ ð# = â«)E
FE
â ð¥, ð¡ . ð¥# . ðð¥ = ð# ð¡
š ðŒ ð#L = â«)E
FE
â ð¥, ð¡ . ð¥#L . ðð¥ = ð# ð¡ L. 2ð â 1 âŒ
š ðŒ ð#LF$ = â«)E
FE
â ð¥, ð¡ . ð¥#LF! . ðð¥ = 0
š Using the Taylor expansion we can get:
š exp
$
#
ðŒ ð# = âH85
H8E $
H!
.
$
#
ðŒ ð#
H
= âH85
H8E $
H!
.
"!&
#
H
31
32. Luc_Faucheux_2021
Another useful relationship for a Gaussian process - IV
š exp
$
#
ðŒ ð# = âH85
H8E $
H!
.
$
#
ðŒ ð#
H
= âH85
H8E $
H!
.
"!&
#
H
š ðŒ exp[ð] = â«)E
FE
â ð¥, ð¡ . exp(ð¥) . ðð¥
š ðŒ exp[ð] = â«)E
FE
â ð¥, ð¡ . âH85
H8E $
H!
. ð¥ H . ðð¥
š ðŒ exp[ð] = âH85
H8E $
H!
. â«)E
FE
â ð¥, ð¡ . ð¥H. ðð¥
š ðŒ exp[ð] = âH85
H8E $
H!
. ðŒ{ðH} and only the terms even in ð are non zero, so we can rewrite
using ð = 2ð
š ðŒ exp[ð] = âL85
L8E $
#L !
. ðŒ{ð#L}
š ðŒ exp[ð] = âL85
L8E $
#L !
. ð# ð¡ L. 2ð â 1 âŒ
32
37. Luc_Faucheux_2021
Another useful relationship for a Gaussian process - IX
š ðŒ exp ð = exp ðŒ ð . exp
$
#
ðŒ (ð(ð¡)â < ð >&)#
š In some textbooks you find the following notation:
š ðŒ (ð(ð¡)â < ð >&)# = ð[ð(ð¡)] for the Variance
š ðŒ ð = ð[ð(ð¡)] for the Mean
š ðŒ exp ð = exp ðŒ ð . exp
$
#
ðŒ (ð(ð¡)â < ð >&)#
š ðŒ exp ð = exp ð[ð(ð¡)] . exp
$
#
ð[ð(ð¡)]
š ðŒ exp ð = exp[ð] . exp
$
#
ð
š ðŒ ð: = ðU. ð
7
!
37
38. Luc_Faucheux_2021
Another useful relationship for a Gaussian process - X
š Note that in the Gaussian case we could also have explicitly derived the formula like we did
in the Langevin deck (couple of slides following)
38
39. Luc_Faucheux_2021
Some properties of the GBM - VIa
š ðŒ exp(ðð ð¡ ) = exp
"!
#
ð¡
š ðŒ ð"1 & = ð
/!
!
&
š We can also derive this one explicitly from the the fact that ð ð¡ ~ð(0, ð¡)
š ðŒ exp(ðð ð¡ ) = â«T8)E
T8FE
exp ððŠ . ð1 ðŠ, ð¡ . ððŠ
š And ð1 ðŠ, ð¡ = â ðŠ, ð¡ =
$
#C&
. exp(
)T!
#&
)
š ðŒ exp(ðð ð¡ ) = â«T8)E
T8FE
exp ððŠ .
$
#C&
. exp(
)T!
#&
) . ððŠ
š ðŒ exp(ðð ð¡ ) = â«T8)E
T8FE $
#C&
. exp ððŠ . exp(
)T!
#&
) . ððŠ
39
41. Luc_Faucheux_2021
Another useful relationship for a Gaussian process - XI
š Before we leave this chapter, there is something else we need to point out, as we will use it
when looking at the Radon-Nikodym derivative.
š ðŒ exp ð = exp ðŒ ð . exp
$
#
ðŒ (ð(ð¡)â < ð >&)#
š This also has to do with the Moment Generating function.
š The moment generating function of a variable ð is the function of the variable ð
š ð: ð = ðŒ exp ðð
š ð: ð = ðŒ exp[ðð] = âL85
L8E $
L!
. ðŒ [ðð]L = âL85
L8E W6
L!
. ðŒ [ð]L
š The cool thing about the Moment Generating function (if it exists, unless the distribution is
pathological) is that the n-th derivative is the n-th moment of the distribution.
š
=6
=W6 ð: ð |W85 = ðŒ [ð]L
41
44. Luc_Faucheux_2021
Another useful relationship for a Gaussian process - XIV
š ðŒ exp ðð = exp ð. ðð¡ . exp
$
#
ð# ð# ð¡
š Now letâs get a taste of the Radon-Nykodim theorem.
š ðŒ is the expectation associated to the random variable ð
š Letâs call it ðŒâ
š We now define a â measure equivalent to the â-measure, defined by :
š ðŒâ ð ð¡ ð 0 = ðŒâ{
=â
=â
ð(ð¡)|ð 0 }
š Suppose now that âout of nowhere) (Baxter p.71), we set the quantity
=â
=â
to be equal to:
š
=â
=â
= exp(âð. ð ð¡ â
$
#
ð#. ð¡)
š Where ð ð¡ is the regular Brownian motion under the â-measure (also sometimes called a
â-Brownian motion)
44
48. Luc_Faucheux_2021
Another useful relationship for a Gaussian process - XVIII
š So we get now:
š ðŒâ exp ð. ð(ð¡) = exp âððð¡ +
$
#
ð# ð¡
š But wait a second!
š We had started with:
š ðŒâ exp ðð = exp ð. ðð¡ . exp
$
#
ð# ð# ð¡
š ðŒâ exp ðð = exp
$
#
ð# ð¡
š So what the equation ðŒâ exp ð. ð(ð¡) = exp âððð¡ +
$
#
ð# ð¡ tells us is the following:
š The distribution of ð(ð¡) under the â-measure is ALSO a Normal distribution with mean
equal to âðð¡ and variance equal to (ð¡)
48
49. Luc_Faucheux_2021
Another useful relationship for a Gaussian process - XIX
š That is an awesome result that we need to ponder a little, and keep in the back of our mind
when doing the proper measure change through the CMG (Cameron-Martin-Girsanov)
theorem using the Radon-Nykodim derivative (not super rigorous at this point, but trying to
just get the jist of it)
š ð ð¡ is a â-Brownian motion, with a Normal distribution ð(0, ð¡)
š ð ð¡ is ALSO a â-Brownian motion, with a Normal distribution ð(âðð¡, ð¡)
š Letâs take a leap here and assume that what is true at time ð¡ is also true for all prior times,
by defining a drifted process:
š hð ð¡ = ð ð¡ + ðð¡
49
50. Luc_Faucheux_2021
Another useful relationship for a Gaussian process - XX
š ð ð¡ is a â-Brownian motion, with a Normal distribution ð(0, ð¡)
š ð ð¡ is ALSO a â-Brownian motion, with a Normal distribution ð(âðð¡, ð¡)
š hð ð¡ = ð ð¡ + ðð¡, is a â-Brownian motion, with a Normal distribution ð(0, ð¡)
š ðŒâ exp ðð (ð¡) = exp
$
#
ð# ð¡
š ðŒâ exp ð hð ð¡ = exp ððð¡ . exp
$
#
ð# ð¡
š ðŒâ exp ð. ð(ð¡) = exp âððð¡ +
$
#
ð# ð¡
š ðŒâ exp ð. hð(ð¡) = exp
$
#
ð# ð¡
š Note that all those are on the marginal distribution with â|ð 0 â
50
51. Luc_Faucheux_2021
Another useful relationship for a Gaussian process - XXI
š Where you need a little leap of faith is to also assume that what we did is also valid for times
prior to the terminal time.
š ðŒâ exp ð. hð(ð¡) = exp
$
#
ð# ð¡
š Which was really:
š ðŒâ exp ð. hð(ð¡) |ð 0 = exp
$
#
ð# ð¡
š By assuming: hð ð¡ = ð ð¡ + ðð¡, we are also implicitly assuming:
š ðŒâ exp ð. ( hð ð¡ â hð ð |ð ð = exp
$
#
ð#(ð¡ â ð )
š
=â
=â
= exp(âð. ð ð¡ â
$
#
ð#. ð¡)
51
52. Luc_Faucheux_2021
Another useful relationship for a Gaussian process - XXII
š Sometimes to make the notation easier to understand, we can use: ðŒ&
1
š To note the expected value at time ð¡ in the probability measure associated to the Brownian
motion ð(ð¡)
š hð ð¡ = ð ð¡ + ðð¡
š ðŒ&
1 exp ðð (ð¡)|ð 0 = exp
$
#
ð# ð¡
š ðŒ&
1 exp ð hð ð¡ |ð 0 = exp ððð¡ +
$
#
ð# ð¡
š ðŒ&
X1 exp ðð(ð¡) |ð 0 = exp âððð¡ +
$
#
ð# ð¡
š ðŒ&
X1
exp ð hð(ð¡) |ð 0 = exp
$
#
ð# ð¡
š
=â
=â
= exp(âð. ð ð¡ â
$
#
ð#. ð¡)
52
53. Luc_Faucheux_2021
Another useful relationship for a Gaussian process - XXIII
š Subject to some condition (most notably the Novikov condition), the results can be
extended from ð â ð(ð¡) (Baxter p.74).
š âIf ð(ð¡) is a â-Brownian motion and ð(ð¡) is a ð-previsible process satisfying the
boundedness condition ðŒâ exp
$
#
â«A85
A8&
ð ð #. ðð |ð 0 < â, then there exists a measure
â such that:
š â is equivalent to â
š
=â
=â
= exp(âð. ð ð¡ â
$
#
ð#. ð¡)
š Becomes:
š
=â
=â
= exp[â â«A85
A8&
ð(ð ). ðð(ð ) â
$
#
â«A85
A8&
ð ð #. ðð ]
š hð ð¡ = ð ð¡ + â«A85
A8&
ð(ð ). ðð
š hð ð¡ is a â-Brownian motion
53
54. Luc_Faucheux_2021
Another useful relationship for a Gaussian process - XXIV
š To quote Baxter, âwithin limits, drift is measure and measure is driftâ
š To quote Gilles Franchini âthe only thing we can really do in stochastic calculus is to calculate
expectations, so it would make sense that the only tools at our disposal are related to
changing the driftâ
š Note that here we showed that if we define a new Brownian motion as the original one plus
a drift, we recover an equivalent measure.
š It is a little more complicated to convince yourself that if you have a measure, ANY other
equivalent measure is such that the two Brownian motions associated to each measures are
only different by a drift:
š â«A85
A8&
ð(ð ). ðð = hð ð¡ â ð(ð¡)
š and that the Radon-Nykodim derivative is given by:
š
=â
=â
= exp[â â«A85
A8&
ð(ð ). ðð(ð ) â
$
#
â«A85
A8&
ð ð #. ðð ]
54
55. Luc_Faucheux_2021
Another useful relationship for a Gaussian process - XXV
š Another cool thing, the Radon-Nykodim derivative is a martingale under the â-measure
š
=â
=â
= exp[â â«A85
A8&
ð ð . ðð ð â
$
#
â«A85
A8&
ð ð #. ðð ]
š And we know from the first useful relationship that:
š ðŒ exp â«A85
A8&
ð ð . ðð ð = exp[â«A85
A8& $
#
ð ð #. ðð ], where ðŒ = ðŒ&
â = ðŒ&
1
š ðŒ&
â exp â«A85
A8&
ð ð . ðð ð |ð 0 = exp[â«A85
A8& $
#
ð ð #. ðð ]
š ðŒ&
â exp â«A85
A8&
ð ð . ðð ð â
$
#
â«A85
A8&
ð ð #. ðð |ð 0 = 1
š Just switching ð ð = âð(ð )
š ðŒ&
â exp â«A85
A8&
ð ð . ðð ð â
$
#
â«A85
A8&
ð ð #. ðð |ð 0 = 1
55
56. Luc_Faucheux_2021
Another useful relationship for a Gaussian process - XXVI
š ðŒ&
â exp â«A85
A8&
ð ð . ðð ð â
$
#
â«A85
A8&
ð ð #. ðð |ð 0 = 1
š
=â
=â
= exp[â â«A85
A8&
ð ð . ðð ð â
$
#
â«A85
A8&
ð ð #. ðð ]
š ðŒ&
â =â
=â
|ð 0 = 1
š Note that this should not be too surprising since the definition of the derivative is:
š ðŒâ ð ð¡ ð 0 = ðŒâ{
=â
=â
ð(ð¡)|ð 0 }
š We can replace ð ð¡ = 1 in the above definition and we will get:
š ðŒâ 1 ð 0 = 1 = ðŒâ{
=â
=â
|ð 0 }
š So we get: ðŒ&
â =â
=â
|ð 0 = 1
56
64. Luc_Faucheux_2021
Quick side note - VII
š In the stochastic calculus (ITO), the solution of the SDE:
š ðð ð¡ = ð ð¡ . ð(ð¡). ðð ð¡
š Is NOT the regular exponential that we are used to, but instead:
š ð ð¡ = ð 0 . exp(â«A85
A8&
ð ð . ðð ð â
$
#
â«A85
A8&
ð ð #. ðð )
š Sometimes the above function is referred to the Doleans-Dade exponential in memory of
Catherine Doleans-Dade, and because is it so useful and used
š â° â«5
&
ð ð . ðð ð = exp(â«5
&
ð ð . ðð(ð ) â â«5
& $
#
ð ð #. ðð )
š ð ð¡ = ð 0 . exp(â«A85
A8&
ð ð . ðð ð â
$
#
â«A85
A8&
ð ð #. ðð )
š ð ð¡ = ð 0 . â° â«5
&
ð ð . ðð ð
64
66. Luc_Faucheux_2021
Quick side note - IX
š The interesting thing is that:
š ðð ð¡ = ð ð¡ . ð ð¡ . [ . ðð ð¡
š Is driftless, and the solution of it is:
š ð ð¡ = ð 0 . exp(â«A85
A8&
ð ð . [ . ðð ð â
$
#
â«A85
A8&
ð ð #. ðð )
š Such that it is a martingale:
š ðŒ&
â ð ð¡ |ð 0 = ð(0)
š That would be another way to recover the useful relationship, is to use the property that a
driftless process is a martingale.
š This is the end of this quick note, but I wanted to point out the nice connection between a
process that is driftless and the fact that it is a martingale, in the case where we can have an
explicit solution of the SDE
66
67. Luc_Faucheux_2021
Quick side note - X
š There is an awful lot of complicated math to prove the equivalence, but very roughly, if the
Novikov condition is respected:
š ðŒ&
â exp(
$
#
â«A85
A8&
ð ð #. ðð )|ð 0 < â
š Then you have equivalence between driftless and martingale.
š Just like in Mario Kart, with the evil Wario, if the Novikov condition is not respected, then
the process becomes a wartingale
67
68. Luc_Faucheux_2021
Quick side note - XV
š In Finance you want to remove the drift (find the martingale)
š In Mario Kart, you want to control the drift especially around the corners
š I want to thank Gilles Franchini for pointing out how crucial the drift was in both situations
68
70. Luc_Faucheux_2021
Re-deriving Ho-Lee using the useful relationship - I
š We put ourselves in the formalism where we have not yet calibrated the Ho-Lee model
š We start with:
š ðð ð¡, ð¡, ð¡ = ð ð¡ . ðð¡ â ð. ([). ðð(ð¡)
š ð ð , ð , ð = ð 0,0,0 + â«Z85
Z8A
ð ð¢ . ðð¢ â â«Z85
Z8A
ð. ([). ðð(ð¢)
70
75. Luc_Faucheux_2021
Re-deriving Ho-Lee using the useful relationship - VI
š ð = â«A85
A8&
{â«Z85
Z8A
ð. ([). ðð(ð¢)}. ðð
75
s
s = t
u
s
s = t
u
ð = n
A85
A8&
ðð n
Z85
Z8A
ð. ([). ðð(ð¢) ð = n
Z85
Z8&
ðð(ð¢) n
A8Z
A8&
ð. ðð
77. Luc_Faucheux_2021
Re-deriving Ho-Lee using the useful relationship - VIII
š We now make use of the useful relationship to derive the mean and the variance of the
quantity: ð ð¡ = â«A85
A8&
ð ð , ð , ð . ðð
š Note that in the first part we did it explicitly, this is a little more general as a derivation
š Since we are after:
š ðð¶ 0,0, ð¡ = ðŒ&
â
exp(â â«A85
A8&
ð ð , ð , ð . ðð )|ð(0)
š And we know that:
š ðŒ exp ð = exp ðŒ ð . exp
$
#
ðŒ (ð(ð¡)â < ð >&)#
š ðŒ exp ð = exp ð[ð(ð¡)] . exp
$
#
ð[ð(ð¡)]
š ðŒ exp ð = exp[ð] . exp
$
#
ð
77
78. Luc_Faucheux_2021
Re-deriving Ho-Lee using the useful relationship - IX
š ð ð¡ = â«A85
A8&
ð ð , ð , ð . ðð
š ð(ð¡) = ð 0,0,0 . ð¡ + â«A85
A8&
â«Z85
Z8A
ð ð¢ . ðð¢ . ðð â ð. â«Z85
Z8&
ð¡ â ð¢ . [ . ðð ð¢
š ðŒ exp ð = exp ð[ð(ð¡)] . exp
$
#
ð[ð(ð¡)]
š Letâs first look at the mean (expected value of ð ð¡ )
š ð ð ð¡ = ðŒ ð(ð¡) = ðŒ&
â
ð(ð¡)|ð(0) to be fully explicit
š Note that we are operating in the risk neutral measure that is associated with our Brownian
motion ð(ð¢)
78
79. Luc_Faucheux_2021
Re-deriving Ho-Lee using the useful relationship - X
š ð(ð¡) = ð 0,0,0 . ð¡ + â«A85
A8&
â«Z85
Z8A
ð ð¢ . ðð¢ . ðð â ð. â«Z85
Z8&
ð¡ â ð¢ . [ . ðð ð¢
š The first two terms are deterministic:
š ðŒ ð 0,0,0 . ð¡ = ðŒ5
â
ð 0,0,0 . ð¡|ð(0) = ð 0,0,0 . ð¡
š ðŒ â«A85
A8&
â«Z85
Z8A
ð ð¢ . ðð¢ . ðð = ðŒ&
â
â«A85
A8&
â«Z85
Z8A
ð ð¢ . ðð¢ . ðð |ð(0) = â«A85
A8&
â«Z85
Z8A
ð ð¢ . ðð¢ . ðð
š The third term is stochastic:
š ðŒ ð. â«Z85
Z8&
ð¡ â ð¢ . [ . ðð ð¢ = ðŒ5
â
ð. â«Z85
Z8&
ð¡ â ð¢ . [ . ðð ð¢ |ð(0)
š Here we use the fact that for any trading strategy (function of time only), the ITO integral is a
martingale:
š ðŒ&
â
â«Z85
Z8&
ð(ð¢). [ . ðð ð¢ |ð(0) = 0 for any function ð(ð¢)
79
80. Luc_Faucheux_2021
Re-deriving Ho-Lee using the useful relationship - XI
š Here we use the fact that for any trading strategy (function of time only), the ITO integral is a
martingale:
š ðŒ&
â
â«Z85
Z8&
ð(ð¢). [ . ðð ð¢ |ð(0) = 0 for any function ð(ð¢)
š Letâs make sure that we are firmly convinced of that fact.
š We saw that in the stochastic calculus deck, but always worth looking at it again.
š As always, replace the integral by a limit of a sum (with the proper convention, LHS for ITO,
Middle for Strato,..) so that you can switch the Expectation operator and the sum operator
š â«Z85
Z8&
ð(ð¢). [ . ðð ð¢ = lim
GâE
â!85
!8G
ð ð ! . {ð ð !F$ â ð(ð !)}
š Letâs note ðŒ&
â
â«Z85
Z8&
ð(ð¢). [ . ðð ð¢ |ð(0) by ðŒ â«Z85
Z8&
ð(ð¢). [ . ðð ð¢ for sake of
simplicity of notation
80
81. Luc_Faucheux_2021
Re-deriving Ho-Lee using the useful relationship - XII
š â«Z85
Z8&
ð(ð¢). [ . ðð ð¢ = lim
GâE
â!85
!8G
ð ð ! . {ð ð !F$ â ð(ð !)}
š ðŒ â«Z85
Z8&
ð(ð¢). [ . ðð ð¢ = ðŒ lim
GâE
â!85
!8G
ð ð ! . {ð ð !F$ â ð(ð !)}
š ðŒ â«Z85
Z8&
ð(ð¢). [ . ðð ð¢ = lim
GâE
â!85
!8G
ðŒ{ð ð ! . {ð ð !F$ â ð(ð !)}}
š ðŒ â«Z85
Z8&
ð(ð¢). [ . ðð ð¢ = lim
GâE
â!85
!8G
ð ð ! . ðŒ{{ð ð !F$ â ð(ð !)}}
š ðŒ â«Z85
Z8&
ð(ð¢). [ . ðð ð¢ = lim
GâE
â!85
!8G
ð ð ! . 0
š ðŒ â«Z85
Z8&
ð(ð¢). [ . ðð ð¢ = 0
š Note that this would ALSO be true in the Stratonovitch calculus, because it is a function of
time only
š â«Z85
Z8&
ð(ð¢). â . ðð ð¢ = lim
GâE
â!85
!8G
ð
A"FA"<3
#
. {ð ð !F$ â ð(ð !)}
81
82. Luc_Faucheux_2021
Re-deriving Ho-Lee using the useful relationship - XIII
š ðŒ â«Z85
Z8&
ð(ð¢). [ . ðð ð¢ = 0
š ðŒ â«Z85
Z8&
ð(ð¢). â . ðð ð¢ = 0
š This would be different for a function of the stochastic driver (also a self replicating strategy)
š â«Z8&7
Z8&6
ð ð(ð¢) . (â). ðð(ð¢) = â«Z8&7
Z8&6
ð ð(ð¢) . ([). ðð(ð¢) +
$
#
â«Z8&7
Z8&6
ðâ² ð(ð¢) . ðð¢
š ðŒ â«Z85
Z8&
ð ð ð¢ . ([). ðð(ð¢) = 0
š ðŒ â«Z85
Z8&
ð ð(ð¢) . (â). ðð(ð¢) â 0
š ðŒ â«Z85
Z8&
ð ð(ð¢) . (â). ðð(ð¢) = ðŒ
$
#
â«Z85
Z8&
ðâ² ð(ð¢) . ðð¢
82
83. Luc_Faucheux_2021
Re-deriving Ho-Lee using the useful relationship - XIV
š ðŒ â«Z85
Z8&
ð ð ð¢ . ([). ðð(ð¢) = 0
š ðŒ â«Z85
Z8&
ð ð(ð¢) . (â). ðð(ð¢) â 0
š ðŒ â«Z85
Z8&
ð ð(ð¢) . (â). ðð(ð¢) = ðŒ
$
#
â«Z85
Z8&
ðâ² ð(ð¢) . ðð¢
š Suppose that we take the simple function ð ð€ = ð€
š Remember we try to stick to the notation where we take lower case for regular calculus
variable and upper case for stochastic variable
š We just write ð ð(ð¢) and ðâ² ð(ð¢) for sake of simplicity
š But following Baxter, we should really write more precisely:
š ð ð€ = ð(ð¢) and
-9(Y)
-Y
|Y81(Z)
83
85. Luc_Faucheux_2021
Re-deriving Ho-Lee using the useful relationship - XVI
š If you recall what we had from the stochastic calculus deck:
š Can you integrate lâð ?
85
86. Luc_Faucheux_2021
Re-deriving Ho-Lee using the useful relationship - XVII
š We had derived then
š Within the ITO convention
š â«&85
&84
ð. ([). ðð =
:(4)!
#
â
$
#
â«&85
&84
1. ðð¡ or â«&85
&84
ð. ðð =
:(4)!
#
â
$
#
ð
š Within the STRATANOVITCH convention
š â«&85
&84
ð. â . ðð = â«&85
&84
ð. ([). ðð +
$
#
ð =
:(4)!
#
â
$
#
ð +
$
#
ð =
:(4)!
#
š STRATANOVITCH as expected follows in a formal manner the usual rules of calculus
š With our current notations
š â«Z85
Z8&
ð(ð¢). ([). ðð(ð¢) =
1(&)!
#
â
$
#
ð¡
š â«Z85
Z8&
ð(ð¢). (â). ðð(ð¢) =
1(&)!
#
86
87. Luc_Faucheux_2021
Re-deriving Ho-Lee using the useful relationship - XVIII
š â«Z85
Z8&
ð(ð¢). ([). ðð(ð¢) =
1(&)!
#
â
$
#
ð¡
š â«Z85
Z8&
ð(ð¢). (â). ðð(ð¢) =
1(&)!
#
š So now taking the expectations and knowing that for the Brownian motion we have the
usual expectation:
š ðŒ ð(ð¡) = 0
š ðŒ ð(ð¡)# = ð¡
š We then recover in a very consistent manner:
š ðŒ â«Z85
Z8&
ð(ð¢). ([). ðð(ð¢) = ðŒ
1(&)!
#
â
$
#
ð¡ = ðŒ
1(&)!
#
â
$
#
ð¡ =
$
#
ð¡ â
$
#
ð¡ = 0
š ðŒ â«Z85
Z8&
ð(ð¢). (â). ðð(ð¢) = ðŒ
1(&)!
#
=
$
#
ð¡
87
88. Luc_Faucheux_2021
Re-deriving Ho-Lee using the useful relationship - XIX
š This is sometimes a very useful trick that you can use
š The ITO integral is a martingale for the measure associated with the Brownian motion
š The expectation is then 0
š ðŒ â«Z85
Z8&
[ððððð¡âððð ðð ð(ð¢)]. ([). ðð(ð¢)
š If the [ððððð¡âððð ðð ð(ð¢)] is a function ð ð(ð¢) with a well behaved first derivative, letâs
call it by the notation
-9(Y)
-Y
|Y81(Z) = ðâ²(ð ð¢ )
š You know the relationship between the ITO and STRATANOVITCH integral:
š â«Z85
Z8&
ð ð(ð¢) . (â). ðð(ð¢) = â«Z85
Z8&
ð ð(ð¢) . ([). ðð(ð¢) +
$
#
â«Z85
Z8&
ðâ² ð(ð¢) . ðð¢
š You can take the expectations on both sides
š ðŒ â«=>?
=>'
ð ð(ð¢) . (â). ðð(ð¢) = ðŒ â«=>?
=>'
ð ð(ð¢) . ([). ðð(ð¢) + ðŒ
.
-
â«=>?
=>'
ðâ² ð(ð¢) . ðð¢
88
89. Luc_Faucheux_2021
Re-deriving Ho-Lee using the useful relationship - XX
š ðŒ â«=>?
=>'
ð ð(ð¢) . (â). ðð(ð¢) = ðŒ â«=>?
=>'
ð ð(ð¢) . ([). ðð(ð¢) + ðŒ
.
-
â«=>?
=>'
ðâ² ð(ð¢) . ðð¢
š ðŒ â«Z85
Z8&
ð ð(ð¢) . (â). ðð(ð¢) = 0 + ðŒ
$
#
â«Z85
Z8&
ðâ² ð(ð¢) . ðð¢
š ðŒ â«Z85
Z8&
ð ð(ð¢) . (â). ðð(ð¢) = ðŒ
$
#
â«Z85
Z8&
ðâ² ð(ð¢) . ðð¢
š And you are left with somewhat easier expressions to deal with
š IN PARTICULAR, you can rely on the fact that you can use the regular rules of âNewtonianâ
calculus (remember, only in a formal manner) within the Startanovitch calculus
š So that makes computing â«Z85
Z8&
ð ð(ð¢) . (â). ðð(ð¢) easier
š Which makes it easier to also computes:
š â«Z85
Z8&
ð ð(ð¢) . ([). ðð(ð¢) = â«Z85
Z8&
ð ð(ð¢) . (â). ðð(ð¢) â
$
#
â«Z85
Z8&
ðâ² ð(ð¢) . ðð¢
89
90. Luc_Faucheux_2021
Re-deriving Ho-Lee using the useful relationship - XXI
š OK, back to the problem at hand here:
š ðð ð¡, ð¡, ð¡ = ð ð¡ . ðð¡ â ð. ([). ðð(ð¡)
š ð ð¡, ð¡, ð¡ = ð 0,0,0 + â«Z85
Z8&
ð ð¢ . ðð¢ â â«Z85
Z8&
ð. ([). ðð(ð¢)
š ð ð ð¡, ð¡, ð¡ = ðŒ ð ð¡, ð¡, ð¡ = ðŒ&
â
ð ð¡, ð¡, ð¡ |ð(0) to be fully explicit
š ðŒ ð ð¡, ð¡, ð¡ = ðŒ ð 0,0,0 + â«Z85
Z8&
ð ð¢ . ðð¢ â â«Z85
Z8&
ð. ([). ðð(ð¢)
š ðŒ ð ð¡, ð¡, ð¡ = ð 0,0,0 + â«Z85
Z8&
ð ð¢ . ðð¢ â ðŒ â«Z85
Z8&
ð. ([). ðð(ð¢)
š You can use the fact that the ITO integral is a martingale or explicitly write
š â«Z85
Z8&
ð. ([). ðð(ð¢) = ð. ð(ð¡)
90
91. Luc_Faucheux_2021
Re-deriving Ho-Lee using the useful relationship - XXII
š ðŒ â«Z85
Z8&
ð. ([). ðð(ð¢) = 0 because it is a ITO integral of a self-financing trading strategy
š Or:
š ðŒ â«Z85
Z8&
ð. ([). ðð(ð¢) = ðŒ ð. ð(ð¡) = 0
š In the specific case of the Brownian motion
š ðŒ ð ð¡, ð¡, ð¡ = ð[ð ð¡, ð¡, ð¡ ] = ð 0,0,0 + â«Z85
Z8&
ð ð¢ . ðð¢
š For the variance:
š ðŒ (ð (ð¡)â < ð >&)# = ð[ð (ð¡)] for the Variance
š ð ð ð¡, ð¡, ð¡ = ðŒ (ð (ð¡, ð¡, ð¡) â ð[ð ð¡, ð¡, ð¡ ])#
91
97. Luc_Faucheux_2021
Re-deriving Ho-Lee using the useful relationship - XXXIII
š Or we can be more pedestrian and recheck once again our hopefully firmly grounded
understanding of ITO integrals by going back once again to the definition of the ITO integral
as a limit of a sum, using the LHS (Left Hand Side) convention for where the function to be
integrated is evaluated
š â«Z85
Z8&
ð ð¢ . ([). ðð ð¢ = lim
GâE
â!85
!8G
ð ð ! . {ð ð !F$ â ð(ð !)}
š We will leave to the pure math guys the job of coming up with all the pathological cases
where a regular well behaved mesh does not work for regular well behaved functions
š ð ð(ð¡) = ðŒ (ð. â«Z85
Z8&
ð¡ â ð¢ . [ . ðð ð¢ )#
š ð ð(ð¡) = ð#. ðŒ (â«Z85
Z8&
ð(ð¢). [ . ðð ð¢ )# with ð ð¢ = (ð¡ â ð¢)
97
102. Luc_Faucheux_2021
Re-deriving Ho-Lee using the useful relationship - XXXVIII
š All right, almost there !
š We are after:
š ðð¶ 0,0, ð¡ = ðŒ&
â
exp(â â«A85
A8&
ð ð , ð , ð . ðð )|ð(0) = ðŒ exp(â â«A85
A8&
ð ð , ð , ð . ðð )
š ðð¶ 0,0, ð¡ = ðŒ exp(âð ð¡
š ðð¶ 0,0, ð¡ is the current bond prices (also referred to as the current term structure)
š We now can use our useful relationship:
š ðŒ ð: = ð ðŒ{:}. ð
7[A]
! = ðU[:]. ð
7[A]
!
š A last little twist because of the minus sign:
š ðŒ ð): = ð ðŒ{):}. ð
7[2A]
! = ð)U[:]. ð
7[A]
!
š (you can convince yourself of it by doing ð â âð)
102
112. Luc_Faucheux_2021
Bond prices dynamics using mean and variance
š We can apply the same trick to recover the dynamics of Bond prices.
š The math is a little more complicated because instead of integrating from 0 to ð¡, we will be
now integrating from ð¡ to ð¡!
š But in essence it will be the same
š We will calculate something that is the exponential of a stochastic process
š We will then compute the expectation of the exponential by using the useful relationship
š Letâs have a couple of slides to remind us about the expectations and how we are going to
use it
112
115. Luc_Faucheux_2021
Bond prices dynamics using mean and variance - IV
š We have for ease of notation the rolling numeraire:
š ðµ ð¡ = exp(â«A85
A8&
ð ð , ð , ð . ðð
š We can rewrite the previous section as:
š ðð¶ 0,0, ð¡ = ðŒ&
â
exp(â â«A85
A8&
ð ð , ð , ð . ðð )|ð(0)
š ðð¶ 0,0, ð¡ = ðŒ&
â $
2(&)
|ð(0)
š ðð¶ 0,0, ð¡! = ðŒ&"
â $
2(&")
|ð(0) and remember that ðµ 0 = 1
š Letâs convince ourselves that we also have the following relation:
š ðð¶ ð¡, ð¡, ð¡! = ðŒ&"
â 2(&)
2(&")
|ð(ð¡) = ðŒ&"
â
exp(â â«A8&
A8&"
ð ð , ð , ð . ðð )|ð(ð¡)
š If we do, then we are in business.
115
116. Luc_Faucheux_2021
Bond prices dynamics using mean and variance - V
š ðð ð¡, ð¡, ð¡ = ð ð¡ . ðð¡ â ð. ([). ðð(ð¡)
š ð ð¡, ð¡, ð¡ = ð 0,0,0 + â«Z85
Z8&
ð ð¢ . ðð¢ â â«Z85
Z8&
ð. ([). ðð(ð¢)
š ð ð ð¡, ð¡, ð¡ = ðŒ ð ð¡, ð¡, ð¡ = ð 0,0,0 + â«Z85
Z8&
ð ð¢ . ðð¢
š ð ð ð¡, ð¡, ð¡ = ðŒ (ð (ð¡, ð¡, ð¡) â ð[ð ð¡, ð¡, ð¡ ])# = ð# ð¡
š We already know that we calibrated the model to the ðð¶ 0,0, ð¡ = ðŒ&
â $
2(&)
|ð(0)
š ð ð¡ =
-
-&
ð 0, ð¡, ð¡ + ð#. ð¡
š We integrated from 0 to ð¡, we will be now integrating from ð¡ to ð¡!
š We had : ð ð¡ = â«A85
A8&
ð ð , ð , ð . ðð
š We now will calculate ð ð¡, ð¡! = â«A8&
A8&"
ð ð , ð , ð . ðð
116
122. Luc_Faucheux_2021
Bond prices dynamics using mean and variance - XI
š ð ð¡, ð¡! = â«A8&
A8&"
{ð ð¡, ð¡, ð¡ + ð 0, ð , ð â ð (0, ð¡, ð¡) +
"!
#
. (ð # âð¡#) â â«Z8&
Z8A
ð. ([). ðð(ð¢) }. ðð
š â«A8&
A8&"
{ð ð¡, ð¡, ð¡ }. ðð = ð ð¡, ð¡, ð¡ . (ð¡! â ð¡)
š â«A8&
A8&"
{ð 0, ð¡, ð¡ }. ðð = ð 0, ð¡, ð¡ . (ð¡! â ð¡)
š â«A8&
A8&"
{
"!
#
. (ð # âð¡#)}. ðð =
"!
~
. ð¡! â ð¡ #. (ð¡! + 2ð¡)
š â«A8&
A8&"
{ð 0, ð , ð }. ðð = â«A8&
A8&"
{ð 0, ð , ð }. ðð , we leave this one as is for now
š And we use our good old friend Guido Fubini on the last term:
š â«A8&
A8&"
{â«Z8&
Z8A
ð. ([). ðð(ð¢)}. ðð
122
123. Luc_Faucheux_2021
Bond prices dynamics using mean and variance - XII
š â«A8&
A8&"
{â«Z8&
Z8A
ð. ([). ðð(ð¢)}. ðð
š We just have to make sure that we are careful about the variables because they are not the
same ones we had on our previous graph
š Before we were starting from 0, we now start the integral at ð¡
123
124. Luc_Faucheux_2021
Bond prices dynamics using mean and variance - XIII
š ð = â«A8&
A8&"
{â«Z8&
Z8A
ð. ([). ðð(ð¢)}. ðð
124
s
ð = ð¡!
u
s
u
ð = ð¡ ð = ð¡!ð = ð¡
125. Luc_Faucheux_2021
Bond prices dynamics using mean and variance - XIV
š ð = â«A8&
A8&"
{â«Z8&
Z8A
ð. ([). ðð(ð¢)}. ðð
125
s
ð = ð¡!
u
s
u
ð = ð¡ ð = ð¡!ð = ð¡
ð = n
A8&
A8&"
{ n
Z8&
Z8A
ð. ([). ðð(ð¢)}. ðð ð = n
Z8&
Z8&"
{ n
A8Z
A8&"
ð. ðð }. ([). ðð(ð¢)
139. Luc_Faucheux_2021
Bond prices dynamics using mean and variance - XXVIII
š We have derived:
š ðð¶ ð¡, ð¡, ð¡! =
*+ 5,5,&"
*+ 5,5,&
. exp{â(ð ð¡, ð¡, ð¡ â ð 0, ð¡, ð¡ ). ð¡! â ð¡ â
"!
#
. ð¡! â ð¡ #. ð¡}
š We can compare to what we derived in the first part of the deck:
š ðð¶ ð¡, ð¡, ð¡! = ðð¶ 0, ð¡, ð¡! . exp{â[ð ð¡, ð¡, ð¡ â ð 0, ð¡, ð¡ ]. ð¡! â ð¡ â
"!
#
ð¡ ð¡! â ð¡ #}
š Yep, we still got itâŠwe ended up mot messing up too much in the derivation !!
š We can once again put in evidence the affine property of the Ho-Lee model
š This is a neat way to derive the dynamics of Bond prices using:
š Fubini theorem
š Isometry property of the ITO integral
š Useful relationship
139
141. Luc_Faucheux_2021
The deflated Zeros
š In some textbooks (Bjork for example), a very useful quantity is defined:
š The deflated Zeros
š The Zeros are the usual Zero Coupon Bonds: ðð¶ ð¡, ð¡, ð¡!
š By the way when I redo all those slides I will just use one letter ð ð¡, ð¡, ð¡! , not quite sure why
I started using ðð¶ instead of ð
š I might even start now to start getting used to it.
š The other quantity is the rolling numeraire, or money market account:
š ðµ ð¡ = exp(â«A85
A8&
ð ð , ð , ð . ðð )
š The deflated zeros are defined as:
š {ðð¶ ð¡, ð¡, ð¡! = |ð ð¡, ð¡, ð¡! =
*+ &,&,&"
2(&)
=
* &,&,&"
2(&)
141
142. Luc_Faucheux_2021
The deflated Zeros - II
š {ðð¶ ð¡, ð¡, ð¡! = |ð ð¡, ð¡, ð¡! =
*+ &,&,&"
2(&)
=
* &,&,&"
2(&)
š The cool thing about the deflated Zeros is that in the Risk neutral measure they are
martingales, and so their SDE is driftless:
š ð |ð ð¡, ð¡, ð¡! = 0. ðð¡ + ð â* ð¡, ð¡! . [ . ðð(ð¡)
š Letâs re-derive that just for sake of consistency.
š ðµ ð¡ = exp(â«A85
A8&
ð ð , ð , ð . ðð )
š ððµ ð¡ = ð ð¡, ð¡, ð¡ . ðµ ð¡ . ðð¡
š In the risk neutral measure we have for the Zeros:
š
=*+ &,&,&"
*+ &,&,&"
= ð ð¡, ð¡, ð¡ . ðð¡ + ð ð¡, ð¡!, ð¡! . ([). ðð ð¡
š With for Ho-Lee: ð ð¡, ð¡!, ð¡! = ð. (ð¡! â ð¡)
142