SlideShare a Scribd company logo
1 of 215
Download to read offline
Luc_Faucheux_2020
Stochastic Calculus - Part I
Integrals - Ito lemma – SDEs and SIEs
1
Luc_Faucheux_2020
What is this class, and what it is not
¨ Not a formal, so please interrupt if any question
¨ More of a pragmatic approach on how to approach stochastic calculus
¨ I have tried as much as possible to alert when there is something to be careful about
¨ I have also tried as much as possible to be as rigorous as possible, without getting lost in the
notations, or being too formal just for the sake of being formal
¨ Those notes are more “notes of a practitioner”, and by no means I would dare to hope to
substitute a robust course in stochastic calculus
¨ Those slides originated from a class I taught in 2018 at the hedge fund DRW to their first
year associate class, 40 students or so from various backgrounds. The class was a general
Fixed-Income class over 8 full days. It was intense, exhilarating, and kept me on my toes the
whole time
¨ Starting from the usual Ito lemma as in most textbooks (Hull), play with it for a while, then
start again from the proper formulation of the stochastic integrals
2
Luc_Faucheux_2020
How those slides came about
¨ Those slides originated from a class I taught in 2018 at the hedge fund DRW in the great city
of Chicago. I have modified them and added to them over the past couple of years.
¨ The class was composed of 40 students or so with various backgrounds, ranging from
computer science students with almost no background in Finance, to recent graduates of
Masters in Math, to some graduates of the prestigious MSCF (Master of Science in
Computational Finance) at Carnegie Melon University under the guidance of Steven Shreve,
to some who had almost no mathematical background in options, bond math, random
processes but had advanced degrees in Economy, so it was rather a tricky bunch
¨ The class was a general Fixed-Income class over 8 full days. It was intense, exhilarating, and
kept me on my toes the whole time. It pushed me to realize what I had not understood
about stochastic processes for 20 years or so, because I never bothered to ask the “what if”
questions and took a lot for granted out of sheer intellectual laziness
¨ The textbook we used was Hull so you will see pages reference to this book, as we used it in
class as a starting point to further explorations of the derivative pricing theory.
3
Luc_Faucheux_2020
How those slides came about - II
¨ It is often said that the only person who learns anything out of a class is the teacher.
¨ I hope that my DRW students got something out of it.
¨ But I know for a fact that without those two weeks in Chicago teaching, I would not have
gotten those slides off the ground, and most of the materials would still be in disparate
pieces of papers flying around my desk.
¨ Hopefully the end result is not complete garbage, at least I know that I greatly enjoyed
putting those together.
¨ So, even if the end result is not up to your standards, I am extremely grateful to DRW for
having given me the opportunity to teach the associate class that summer of 2018, and even
more so grateful to the students who during those two weeks, took me out of my comfort
zone, and forced me to confront what I knew and what I realized I was rather ignorant of.
¨ When you are teaching in front of a bunch of super smart people with different backgrounds
for 8 days straight, there is no hiding behind the curtain
¨ So thanks again to DRW and the 2018 associate class ! I miss you guys.
4
Luc_Faucheux_2020
How those slides came about - III
¨ Also for those of you who have spent more than 5 minutes with me, you will have noticed
that I bring the Ito-Stratonovitch controversy a lot.
¨ First of all that is sort of a hobby of mine, ever since my PhD thesis (Appendix B)
¨ Second, I have found it to be quite enlightening, because everyone takes Ito for granted, and
then you ask the question “what if”, and that forces you to make sure that your
understanding of Ito was solid. So I am using the example of Stratonovitch to really test the
fact that my understanding of ITO is solid
¨ Apologies on the Powerpoint format, there are battles with Microsoft that you cannot win
(including the fact that the Solver is a Macro and not a function, as opposed to some of the
earlier spreadsheets like Wingz)
¨ At times it might feel like going down the rabbit hole, or going up the river to meet Kurtz,
but (at least in my experience), I have found those discursions to be useful.
¨ So this is not really something great about Stratonovitch, it is using Stratonovitch to
convince ourselves that we understand ITO
5
Luc_Faucheux_2020
How those slides came about - IV
¨ Again quite frankly those notes are more for me (sorry)
¨ I had noticed that I had a number of handwritten notes flying around my desk, and when
needed at times I would just rederive from scratch rather then finding the right one.
¨ So this is trying to put in one place, in a format that I can copy/past easily, most of what I
had to go through and still use. I have tried (and failed at many places) to keep the notation
consistent
¨ Again, this is from a “practitioner” point of view, I have tried to be rigorous when I felt it was
needed, and be less so when I felt that this was a detail that was not needed and at times
would obscure the intuition.
¨ Notations are hard to keep consistent. I have also find at times that the right notation can
be illuminating, whereas the full explicit one can be cumbersome. So I have tried to adapt
the notation to what was needed to grasp the concept, and at times be more careful about
it to make sure that we do not fall into a trap.
¨ I guess Godel knew that, the right notation can change completely the proof….
6
Luc_Faucheux_2020
If you really want to master stochastic calculus.
¨ The uncontested bible in the field of stochastic calculus for Finance. Quite dense and
concise. It sometimes take me 40 pages to understand what Steven Shreve does in 2 lines.
7
Luc_Faucheux_2020
If you really want to master stochastic calculus - II
¨ An absolute wonderful short book. The notations at times are infuriating, but an absolute
must read
8
Luc_Faucheux_2020
If you really want to master stochastic calculus - III
¨ If you are like me coming from a Physics background, this book is still relevant today, a
testament to the genius of Van Kampen
9
Luc_Faucheux_2020
If you really want to master stochastic calculus - IV
¨ You cannot ignore this book. Every sentence carries meaning, and is worth reading time and
time again.
10
Luc_Faucheux_2020
If you really want to master stochastic calculus - V
¨ Another wonderful short book. The logic is clear, concise and beautiful. The exercises are
worth going through. From one of the most respected options traders in the field. He also
has quite a ferocious appetite for chocolate cakes.
11
Luc_Faucheux_2020
If you really want to master stochastic calculus - VI
¨ Quite applied. The appendix on SDEs is rather beautiful, and follows Mikosch. Great for
practical applications in the field of Fixed-Income
12
Luc_Faucheux_2020
If you really want to master stochastic calculus - VII
¨ What not to say about this book? An absolute gem. The 1900 Ph.D. thesis of Louis
Bachelier (in French!) with an amazing translation by Davis and Etheridge, and some great
chapters about the history of modern finance. Bachelier did it all, 80 years or so before
everyone else
13
Luc_Faucheux_2020
If you really want to master stochastic calculus - VIII
¨ I could not but not add this one here. It is a movie about the life of Vincent (Wolfgang)
Doblin (Doebling) who essentially discovered Ito calculus at least 5 to 10 years before Ito in
circumstances so incredible that they made a movie out of his life. Ito lemma and calculus is
now usually referred to as Ito-Doblin lemma and calculus in recognition of Vincent’s
incredible accomplishments
14
Luc_Faucheux_2020
If you really want to master stochastic calculus - IX
¨ Amazing book if you are coming from a Physics background
15
Luc_Faucheux_2020
If you really want to master stochastic calculus - X
¨ Found this book as I was almost finished with those slides, and thought about throwing
them away, because this book has pretty much anything you want. Pretty heavy on the
operators formalism, which is quite elegant once you get used to it, but that it a step that
you need to go through
16
Luc_Faucheux_2020
What is so hard about Stochastic Calculus?
¨ It is quite recent
¨ It is quite cumbersome
¨ It is not intuitive
¨ It is incomplete
¨ No one really knows how to do it.
¨ This presentation is trying to strike a balance between being practical and being rigorous, so
apologies for the many terms in “”, whereas a rigorous math class will define what terms like
“stable” or “got to 0” or “go to infinity” or “converge” much more exactly. What we will say
is “almost” rigorous and works in practice 99% of the time
¨ This is more of a practitioner’s point of view on how to use (or not) stochastic calculus, as it
relates to finance and Physics
17
Luc_Faucheux_2020
Stochastic Calculus is quite recent
¨ Geometry ~ -6,000 BC (navigating looking at the stars, buildings, ships,…)
¨ Fractions (music with scale, buildings,,)
¨ Probabilities (~1,600 AD), Pascal triangle, combinatory analysis
¨ Calculus (Newton, Leibniz, finite differences) (~1,600 AD) (we started shooting canons long
range, Electricity, magnets, how do they work? The ICP is still asking)
¨ Taylor Expansion (1720)
¨ Brown (1890), Wiener (1940), Bachelier (1900), Einstein (1905),Langevin (1908),
Doblin(1940), Ito (1950), Feynman-Kac (1950), Stratonovitch (1966), Black-Sholes (1972)
¨ So let’s go a little easy on ourselves, shall we?
¨ Quantum Stochastic Calculus (1980), random processes (diffusion) in fractal geometries
18
Luc_Faucheux_2020
Stochastic calculus is a French thing (except Gauss)
¨ Black-Sholes might have gotten a Nobel prize in 1972, but Louis Bachelier did it all in his
Ph.D. thesis in 1900 (almost)
¨ Ito might have been known until recently for the Ito calculus and Ito lemma, but Vincent
Doelin wrote it all while on the Ardennes front in World War I. This is now being recognized
and some textbooks use the term “Ito-Doeblin” instead of “Ito”
¨ Paul Langevin also essentially wrote the textbook on SDEs
¨ And for the math, all you need is Laplace, Fourier, Cauchy
¨ Taylor expansion is the only non-French, but it is essentially L’Hospital rule, so
again…French…
¨ Note: L’Hopital rule is very powerful and often overlooked.
¨ If two functions 𝑓(𝑥) and 𝑔(𝑥) and are differentiable, and lim
!→#
[
$%(!)
(%(!)
] exists, then
¨ lim
!→#
[
$(!)
((!)
] = lim
!→#
[
$%(!)
(%(!)
]
19
Luc_Faucheux_2020
The Ph.D. thesis of Louis Bachelier
¨ Reading the original thesis (both in French if you can and the excellent translation by Mark
Davis and Alison Etheridge) is humbling.
¨ Without a strong well-developed theory of stochastic calculus (Ito lemma) that only came
about in the 1960s or so
¨ Without a strong theoretical footing of what is a numeraire and how to price a derivative in
the risk-neutral probability associated to that numeraire (Pliska 1980 or so)
¨ Without yet the strong connection between PDE (Partial Differential Equations) and SDE
(Stochastic Differential Equations) that really came about from the Feynman-Kac formula
(1950 roughly)
¨ Louis Bachelier managed to not only built a theory of option pricing that is nowadays
coming back in fashion with a vengeance, but perusing through the rather short thesis, one
cannot but be amazed at the breadth of his genius, but also at his attention to details.
Bachelier at times go through numerical examples with the same precision and clarity of
thoughts that he displays in the other more theoretical parts of his thesis.
20
Luc_Faucheux_2020
Stochastic Calculus is not intuitive
¨ Regular calculus has usually to do with ”things that are around some other things”
¨ Taylor expansion and derivatives (expansion around a value, local derivatives at or around a
point)
¨ Finite difference for integrating functions on an axis or a path (keep following the path in a
continuous fashion)
¨ Stochastic Calculus tries to address the problem of dealing with “things around things that
are not there”. What do I mean ?
¨ Take the coin flipping problem (Head is +1, Tail is -1). The coin is either head or tail (+1 or -
1), never anything else. And yet we will try to calculate expansions, derivations, integrations
of functions around the mean or average (0), which is NOT a possible state of the coin.
¨ Without stating the obvious or oversimplifying, this is the crux of the problem with
stochastic calculus, and also that we are so used to usual calculus that we take a lot of things
for granted
21
Luc_Faucheux_2020
Stochastic processes are “non-differentiable”
¨ A stochastic process essentially “flips a coin” at each point in time.
¨ A “regular” process, meaning it is differentiable, would have a unique tangent for every
point in time. If 𝑋 𝑡 is differentiable, there is a unique
)*(+)
)+
¨ The stochastic process does NOT have a unique tangent, because, using the coin flip idea,
and being somewhat liberal with scaling, at each point in time, 𝑋 𝑡 goes to either {𝑋 𝑡 +
𝛿𝑋} or {𝑋 𝑡 − 𝛿𝑋} with some probability (50% in the simplest case, or driftless case).
¨ The tangent is NOT the average (0) of the two possible tangents.
¨ So in essence for a stochastic process, writing something like
)*(+)
)+
is meaningless
¨ In stochastic processes, the only real thing that we can do is integrals, almost never
differential calculus (not surprising as it is not differentiable)
¨ NOTE that non-differentiable does NOT mean not smooth or not continuous
22
Luc_Faucheux_2020
Stochastic calculus is usually self-similar
¨ We will go over this in more details, but essentially the simplest stochastic process is the
Wiener process or also called the standard Brownian motion. We will start with it
¨ 𝑊 𝑡 has a normal 𝑁(0, 𝑡) distribution function
¨ 𝑊 𝑡 − 𝑠 has obviously the same normal 𝑁(0, 𝑡 − 𝑠) distribution function
¨ {𝑊 𝑡 − 𝑊(𝑠)} has ALSO the same 𝑁(0, 𝑡 − 𝑠) distribution
¨ Note that they are NOT the same, writing 𝑊 𝑡 − 𝑊 𝑠 = 𝑊(𝑡 − 𝑠) is obviously wrong
but you will find sometimes the notation:
¨ 𝑊 𝑡 − 𝑊 𝑠 ≝ 𝑊(𝑡 − 𝑠), meaning distributional identity, NOT pathwise identity
¨ The simple Brownian motion is self-similar with coefficient (1/2)
¨ 𝑇, 𝑊 𝑡- , 𝑇, 𝑊 𝑡. , . . , 𝑇, 𝑊 𝑡/ ≝ 𝑊 𝑇𝑡- , 𝑊 𝑇𝑡. , . . , 𝑊 𝑇𝑡/ , with 𝐻 = 1/2
¨ This is sometimes used to numerically construct Brownian motion.
¨ If you “scale” up the time scale by a factor 𝑇, the space scale gets only magnified by a factor
𝑇.
23
Luc_Faucheux_2020
Stochastic Calculus is cumbersome
¨ Usual knowledge and tricks of calculus do not apply anymore
¨ Chain rule does not work
¨ Functional derivation does not work : 𝑑 𝑙𝑛𝑆 ≠ ( ⁄𝑑𝑆 𝑆) !
¨ Integration and especially derivations are not well defined
¨ Usual calculus 𝑊(𝑡), when (𝛿𝑡) →0, (𝛿𝑊)~𝛿𝑡, (𝛿𝑊).~(𝛿𝑡).
¨ Stochastic calculus we still have (𝛿𝑊) → 0, BUT WE ALSO HAVE (𝛿𝑊).~𝛿𝑡 so higher orders
are mixed together
¨ Coin toss (+1, -1). Average is 0, variance scales linearly with the number of flips
¨ Can you think of processes where variance goes to 0 and we need to go to the next order?
¨ Can you think of a process where (𝛿𝑊). scales maybe as (𝛿𝑡)0, where 𝑈 < 1 ?
¨ Also, if you think about it, when 𝛿𝑊is stochastic, somehow (𝛿𝑊). is now deterministic, or
at least has a deterministic component, the inverse is NOT true
24
Luc_Faucheux_2020
Always better to integrate than to differentiate
¨ If we have a stochastic process 𝑊(𝑡), we would like to work with functions 𝑓(𝑊) (please
note that those functions are usually well behaved, meaning differentiable and such,
without going into too much math)
¨ 𝑓(𝑊) is differentiable in 𝑊
¨ 𝑊(𝑡) is NOT differentiable in 𝑡
¨ But we are really dealing with 𝑓(𝑊 𝑡 ) so we would like to write things like
¨ 𝛿𝑓 =
)$
)1
.
)1
)+
. 𝛿𝑡 which is one way to write the traditional “chain rule”
¨ In integral form, the chain rule would read something like:
¨ 𝑓 𝑊 𝑡 − 𝑓 𝑊 0 = ∫234
23+ )$
)1
.
)1
)2
. 𝑑𝑠
¨ The crux of the issue in stochastic calculus is that we do not know what
)1
)2
means, and so
we should NOT expect to be able to rely on the usual chain rule
25
Luc_Faucheux_2020
The whole Ito-Stratanovitch thing
¨ Essentially, ITO breakthrough was to find a way to define:
¨ 𝑓 𝑊 𝑡 − 𝑓 𝑊 0 = ∫234
23+ )$
)1
.
)1
)2
. 𝑑𝑠 = ∫234
23+ )$
)1
. 𝑑𝑊
¨ ITO invented the field of stochastic integrals
¨ The essence of it is that 𝑑𝑊(𝑡) is a “jump”
¨ ITO (1950) defines the ITO integral as a limit of a sum where the value of
)$
)1
is taken “before
the jump”. A consequence of this convention is that the usual rules of calculus (chain rule,
Leibniz rule,..) do not apply. For example 𝑑 𝑙𝑛𝑆 ≠ ( ⁄𝑑𝑆 𝑆) ! On the other hand the ITO
integral has the nice property to be a martingale (zero expected value)
¨ Stratanovitch (STRATO) defines the STRATO integral as a limit of a sum where the value of
)$
)1
is taken “in the middle of the jump”. A rather intriguing consequence is that in STRATO
calculus the usual rules of calculus are (formally!) respected. 𝑑 𝑙𝑛𝑆 = ( ⁄𝑑𝑆 𝑆). HOWEVER
the STRATO integral is NOT a martingale.
26
Luc_Faucheux_2020
The whole Ito-Stratanovitch thing - II
¨ So we will see how to not get confused between the two equally valid interpretations of the
integrals (from a really theoretical point of view mathematicians prefer ITO because it is
defined over a wider range of functions than STRATO, but really nothing we should concern
ourselves at this point).
¨ This will take us some time, first going over the Riemann integral in regular calculus
¨ Then defining the ITO
¨ Then looking at the relationship between ITO and STRATO integral
¨ Then explicitly proving the ITO lemma, then the equivalent STRATO lemma
¨ From then we look at the SDE (SIE), and we try to map the relations between a specific SDE
and its associated PDE for the PDF
¨ SDE: Stochastic Differential Equation
¨ SIE: Stochastic Integral Equation
¨ PDE: Partial Differential Equation (that is usual regular calculus, but not easy by any means)
¨ PDF: Probability Density Function
27
Luc_Faucheux_2020
The whole Ito-Stratanovitch thing - III
¨ So apologies in advance if those slides feel pedestrian at time, but unfortunately in order to
be somewhat rigorous without losing the intuition, and in order to convince ourselves that
we are on somewhat firm ground to justify what we write (without having a 100% rigorous
mathematical proof), we have to walk before we run.
¨ Alternatively, you could be a genius like Vincent Doelin and bypass the entire theory of
stochastic integrals and express everything as a Brownian time change, 40 years or so before
everyone else, while fighting WWII in the Ardennes as a radio operator.
¨ If I have time, part V of those decks will be on that
¨ Also the ITO-STRATO controversy in the 1990s was linked to the concept of thermal ratchets,
Brownian motors, biological motors, hence quite a few articles on the subject.
¨ This is also linked to an individual that physicists refer to as the Maxwell’s demon
¨ In deck III we will revisit this unsavory character, who prompted me spending a lot of time
on ITO-STRATO over my life and as a PhD student.
28
Luc_Faucheux_2020
The whole Ito-Stratanovitch thing - IV
¨ The Maxwell demon putting the moves on Mr. Tompkins fiancée and trying to impress her
with his tennis skills
29
Luc_Faucheux_2020
The whole Ito-Stratanovitch thing - V
¨ A more common representation of the demon
30
Luc_Faucheux_2020
The whole Ito-Stratanovitch thing - VI
¨ There are actually applications in Finance of the Maxell demon, known as the Parrondo
paradox.
¨ Two trading strategies (PM at a hedge fund) on average lose money (B and C, blue and green
line)
¨ 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
31
Luc_Faucheux_2020
Just a taste of how powerful Vincent Doelin was
¨ SDE have usually the form:
¨ 𝑑𝑋 𝑡 = 𝑎 𝑋 𝑡 , 𝑡 . 𝑑𝑡 + 𝑏 𝑋 𝑡 , 𝑡 . 𝑑𝑊, which really should always be written as SIE:
¨ 𝑋 𝑡5 − 𝑋 𝑡6 = ∫+3+6
+3+5
𝑑𝑋 𝑡 = ∫+3+6
+3+5
𝑎 𝑋 𝑡 , 𝑡 . 𝑑𝑡) + ∫+3+6
+3+5
𝑏 𝑋 𝑡 , 𝑡 . 𝑑𝑊(𝑡)
¨ The whole theory of ITO is trying to define what exactly is : ∫+3+6
+3+5
𝑏 𝑋 𝑡 , 𝑡 . 𝑑𝑊(𝑡)
¨ What Doelin did essentially was to say: hey I do not need to define this integral, which is
subject to the exact convention of “where to take the value of 𝑏 𝑋 𝑡 , 𝑡 before, during or
after the jump 𝑑𝑊(𝑡)”, and run into all sort of Ito-Stratanovitch confusion, because I am a
genius, whatever that integral is, it is equal to :
¨ ∫+3+6
+3+5
𝑏 𝑋 𝑡 , 𝑡 . 𝑑𝑊(𝑡) ≝ 𝑊(∫+3+6
+3+5
𝑏 𝑋 𝑡 , 𝑡 .. 𝑑𝑡) which is perfectly well defined.
¨ Boom. Microphone drop.
¨ And by the way he burnt most of his research before killing himself to avoid capture by the
Germans, so who knows what else he had discovered….
32
Luc_Faucheux_2020
ITO and DOELIN comparison in 2 pictures
33
Luc_Faucheux_2020
Ito-Doblin and stochastic calculus is..
¨ “First and foremost defined as an integral calculus”
¨ Really the only thing that works is integrating stochastic processes.
¨ There are theories of differentiations for stochastic variables
¨ Malliavin calculus (1980)
¨ One day I will try to understand what that actually means. I still have no idea. But I should
try after all Malliavin is also French
¨ So even the Ito lemma as we know it is really better expressed in integral form
¨ However most textbooks do present it in “differential” form like a Taylor expansion, or even
deal with SDE quite liberally. This is sometimes for ease of notations and we will fall into
that pattern also. Note that this is a FORMAL equivalence to regular calculus equations like
Taylor expansion. Taking those literally leads to mistakes, as we will demonstrate
34
Luc_Faucheux_2020
Why are PDEs so important, and why SDEs are terrifying
¨ PDEs are Partial Differential Equations
)!7(!,+)
)!! =
)7(!,+)
)+
¨ Usually in the context of stochastic calculus they will appear for the PDF (Probability Density
Function) of the stochastic variable 𝑋 (like a Gaussian), 𝑃(𝑥, 𝑡) or for functions of that
variable (call payoff or 𝐶(𝑋, 𝜎, 𝐾, 𝑇) for example)
¨ PDEs are well defined and well known (since 1700), tons of knowledge on how to deal with
those (Navier-Stokes in fluid mechanic, Maxwell equations in electro-magnetism, Fokker-
Planck, Feynman-Kac,..)
¨ Once you know the PDE, you know ALL the moments of the distribution in the case of a PDE
¨ PDEs are complete, SDEs are incomplete
¨ SDEs are Stochastic Differential equations 𝑑𝑆 = 𝜇. 𝑆. 𝑑𝑡 + 𝜎. 𝑆. 𝑑𝑊
¨ No one really knows how to deal with them, especially if the volatility is a function of the
stochastic variable (and it always is, or when you look at functions of X)
¨ But sometimes you can get rid of a SDE and use a related PDE (that was the trick that Black
Sholes discovered and got a Nobel prize for), or Dupire equation
¨ Also stay away from SDE, always look for the PDE
35
Luc_Faucheux_2020
The structure of those slides
¨ We will first start with the usual textbook (Hull) that presents essentially what are Ito SDEs
and Ito lemma being a regular Taylor expansion just making sure that we go high enough to
keep all the terms linear in time and linear in the stochastic driver
¨ So in some ways, in the usual calculus 𝑊(𝑡), when (𝛿𝑡) →0, (𝛿𝑊)~𝛿𝑡, (𝛿𝑊).~(𝛿𝑊).
¨ Stochastic calculus we still have (𝛿𝑊) → 0, BUT WE ALSO HAVE (𝛿𝑊).~𝛿𝑡 so higher orders
are mixed together, but if we keep the right terms we should be fine
¨ This is usually where most textbooks in finance stops
¨ We will show by using something called Stratonovitch convention, that treating the SDEs and
the Taylor expansion the way we would do in regular calculus is wrong.
¨ In fact the Ito lemma and the Ito SDEs are just a formal manner to write Integrals and SIE,
which is really the only thing that you can do in stochastic calculus
¨ Remember, you always read that a stochastic process is NOT differentiable, yet somehow we
all proceed happy to write Stochastic DIFFERENTIAL Equations, and writing Ito lemma as a
Taylor expansion, without really thinking twice about it
36
Luc_Faucheux_2020
The structure of those slides - II - Integrals
¨ We then take a couple of steps back and go over a review of the regular integrals (Riemann)
in the regular calculus
¨ We extend this to the realm of stochastic calculus
¨ We show that unlike the regular case, the point taken in the partition buckets (the mesh)
actually matters. One convention is to take the starting point (ITO). Another one is to take
the middle point (Stratonovitch)
¨ We then derive the relationships between the Ito and the Strato integral
¨ At this point there is no finance theory but it will be worth keeping in mind that the Ito
integral will have the property of being a martingale (zero expected value)
¨ Note that things like “partition” or ”mesh” will not be super rigorously defined, but we leave
that to mathematicians, we use those concepts assuming that they are well defined, and we
can check that indeed they are over a certain range of “non-pathological” functions or
geometries
37
Luc_Faucheux_2020
The structure of those slides - III - Lemma
¨ From what we learned from looking at those integrals, we then look at the issue around
making some operations on those integrals. This is where we look at Ito lemma
¨ Ito lemma is crucial in finance
¨ We will see that formally we can write the lemma, and perform operations on functions and
integral, in a formal manner that looks like regular calculus, with the exceptions that you
have to go “one more up” in any kind of derivations or Taylor expansion, in order to capture
ALL the terms linear in time and linear in the stochastic driver
¨ In doing so the “usual” rules of calculus (derivations, integrations, Leibniz,..) are no longer
true in stochastic calculus using the Ito interpretation of the integral
¨ We will compare this with the Strato integral
¨ We will show that the ”usual” rules of calculus are still formally present in Strato calculus
(remember this is a formal analogy). This can be useful when explicitly solving equations
¨ We then show the relationship between the Ito and the Strato lemma
38
Luc_Faucheux_2020
The structure of those slides - IV - SDE
¨ From what we learned looking at the integrals and how to manipulate them, we can try to
look at what would be SDE, Stochastic Differential Equations
¨ Remember though, that just the same way we can formally write Ito and Strato lemma in
“differential” form for ease of notation, those are ALWAYS simpler way to write down what
would really be equations dealing with integrals
¨ In stochastic calculus, I do not know what a differentiation would actually mean
¨ All I can do really is to integrate
¨ Like the integral and the lemma, we will show the relationship between the Ito SDE and the
Strato SDE (or really more exactly the Ito SIE and the Strato SIE), and introduce the so-called
“drift” between the two representations
¨ SIE: Stochastic Integral Equations
39
Luc_Faucheux_2020
The structure of those slides - V - PDE
¨ From the SIE, we then explore the correspondence between the SIE and the PDE (Partial
Differential Equation) for the PDF (Probability Distribution Function) of the stochastic
variable.
¨ PDE are just part of the regular calculus, there is no issue there
¨ SIE and SDE depends on the interpretation (Ito or Strato).
¨ We will then by extension look at PDE that would correspond to a specific SDE under Ito,
and similarly under Strato.
¨ This is where it can get a little confusing (or even more confusing that it already is)
¨ We will revisit our old friend the Maxwell demon and see why it was so confusing in the
1990s when applied to biological concepts of thermal ratchets or Brownian motors
¨ Because a PDE is “exact” (once you know the PDE, and if you can solve it you have the PDF,
so you have exactly all the moments of the stochastic variable) whereas an SDE only has the
first two moments in most cases, and is also subject to interpretation, it is worth keeping in
mind that fact: a PDE is not subject to interpretation. An SDE is subject to interpretation,
and needs to be treated with great caution
40
Luc_Faucheux_2020
The structure of those slides - VI - PDE
¨ We will look at some cases of PDEs from the world of Physics in order to gain some more
intuition on what is a firm ground to stand on (PDE), and a somewhat more recent and still a
little shaky one (SDE)
¨ This will be a somewhat indirect introduction to a fundamental theorem that links the world
of “regular” well known calculus of PDE (400 years in the making) with the more recent one
having to do with probability and stochastic calculus (only 100 years in the making), and still
very new.
¨ This year Abel prize was given to pioneers of the ergodic theory, that essentially to crudely
oversimplify, try to find the solutions of PDEs by using properties of the associated SDEs
¨ We can then go back to Black-Sholes with a renewed belief in how justified we are in our
derivation, in particular we tried to “get away” using simple Taylor expansions and just
keeping some higher orders, we showed however how wrong it can get very quickly.
41
Luc_Faucheux_2020
¨ Since we will first fall into the trap of treating an SDE the way we would do usual calculus, a
number of slides in this deck are WRONG. It is sometimes more useful to learn from
mistakes than to follow a magistral correct demonstration.
¨ So in order to not make the whole deck a complete garbage of my random wrong ramblings
on stochastic calculus while going up the river to meet colonel Kurtz, I have put a stamp
“WRONG” across those slides.
¨ So do not read those slides at face value, but know that those are WRONG.
¨ It might be an interesting exercise for you to convince yourself why those slides are wrong.
¨ Again, I left them there because I think it is quite instructive to identify the fault in a
derivation.
¨ We could have kept all the slides in ITO calculus, but it is quite enlightening to also look at
STRATANOVITCH calculus, if only to convince ourselves that we really understand ITO (If we
understand the difference between two things, then most likely that increases our
knowledge about each of those things)
A note of caution
42
Luc_Faucheux_2020
Why this whole thing about Ito calculus? Hull textbook
¨ Hull – White chapters 13, 14 and 15
¨ People got excited about stock prices trading as a percentage (people expect a “return”),
p.306, and so what mattered was the return of the stock 𝑆, or ⁄∆𝑆 𝑆
¨ So then they started writing things like : 𝑑𝑆 = 𝜇. 𝑆. 𝑑𝑡 + 𝜎. 𝑆. 𝑑𝑊, (p.307)
¨ And then they got stuck, because 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊, where 𝑏 𝑥, 𝑡 is a function
of the stochastic variable 𝑥, is not something we know how to deal with (p.306, and no, it is
NOT a “small approximation” as they claim)
¨ So you need to use a ”guess” on how to deal with 𝑏 𝑥, 𝑡 , which is why it is called a “lemma”
¨ Ito (1951) guessed that you can write 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊
as ∆𝑥 = 𝑎 𝑥, 𝑡 . ∆𝑡 + 𝑏 𝑥, 𝑡 . ∆𝑊
or δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥, 𝑡 . 𝛿𝑊
¨ That seems like a good guess but then the rules of calculus are no longer applicable, you can
barely derive without making a mistake, and forget about trying to integrate (p. 311)
¨ Now you get this weird thing where 𝑑 𝑙𝑛𝑆 = ( ⁄𝑑𝑆 𝑆) − ( ⁄𝜎. 2). 𝑑𝑡, (p.312)
43
Luc_Faucheux_2020
Why chose Ito then?
44
¨ Someone else made a “better” guess, Stratonovitch (1966)
¨ Strato guessed that you can write 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊
as ∆𝑥 = 𝑎 𝑥, 𝑡 . ∆𝑡 + 𝑏 𝑥 + Δ𝑥/2, 𝑡 . ∆𝑊
or δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥 + 𝛿𝑥/2, 𝑡 . 𝛿𝑊
¨ Within Strato’s convention, the usual rules of calculus FORMALLY apply 𝑑 𝑙𝑛𝑆 = ( ⁄𝑑𝑆 𝑆),
and the chain rule is verified in the usual manner
¨ So this is super confusing
¨ We will look at both Ito and Strato, and understand how to go from the SDE (Stochastic
Differential Equation) to the PDE (Partial Differential Equation) and the correct PDF
(Probability Density Function).
¨ We will go over the binomial trees and binomial distribution, and its limit the Gaussian
distribution (HW chapter 13, p.296-299)
¨ We will show why the Gaussian distribution is so common and so central to everything
(central limit theorem)
Luc_Faucheux_2020
Actually it is not called Ito, but Ito-Doeblin
¨ It was discovered quite recently (2000) that Vincent Doeblin, born Wolfgang Doeblin, Ph.D.
at 23, and drafted in the French army in 1938, while posted in the Ardennes as a phone
operator, essentially worked out Ito calculus on his own and sent his results to the Academie
des Sciences “sous plis scelle”.
¨ He shot himself after burning the rest of his notes when the German army was about to
advance and take over his positions
¨ So wo knows what else was in his notes? Malliavin calculus maybe ?
45
Luc_Faucheux_2020
Ito and Stratonovitch “Taylor” expansion
¨ Ito (1951) guessed that you can write 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊
as δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥, 𝑡 . 𝛿𝑊
¨ If 𝐹(𝑥) a function of x, the corresponding SDE as a result of Taylor expansion in ITO is:
¨ 𝛿𝐹 =
)9
)!
. 𝛿𝑥 +
-
.
.
)!9
)!! . 𝑏.. 𝛿𝑡
¨ That is the celebrated Ito lemma, or how the chain rule gets modified in Ito stochastic
calculus
¨ Strato guessed that you can write 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊
as δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥 + 𝛿𝑥/2, 𝑡 . 𝛿𝑊
¨ So you think that you could write something like this: the SDE for 𝐹(𝑥) in Stratonovitch
convention is:
¨ 𝛿𝐹 =
)9
)!
. 𝛿𝑥 +
-
.
.
)!9
)!! . 𝑏.. 𝛿𝑡 +
-
.
.
)9
)!
.
)5
)!
. 𝑏. 𝛿𝑡
46
Luc_Faucheux_2020
Ito and Stratonovitch “Taylor” expansion - II
¨ We would like to write something like this for Stratonovitch, following the rule of expanding
to the second order and keeping the terms linear in time and linear in the stochastic driver
(so essentially in a way use Ito calculus in the Stratonovitch convention)
¨ Bear in mind that this is absolutely wrong
¨ It is quite insightful to go through it though and see where it is wrong
¨ So we have a function 𝐹 of the stochastic variable 𝑥, the function 𝐹 in itself is nothing weird,
it is a regular function that we assume to be differentiable 𝐹(𝑥)
¨ Ito calculus tells us that if we want to look at the variations of that function in terms of the
stochastic variable 𝑥, assuming a process 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊
¨ You can formally write within Ito calculus: δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥, 𝑡 . 𝛿𝑊
¨ 𝛿𝐹 =
)9
)!
. 𝛿𝑥 +
-
.
.
)!9
)!! . 𝑏.. 𝛿𝑡
47
Luc_Faucheux_2020
Ito and Stratonovitch “Taylor” expansion - III
¨ More	exactly
¨ 𝛿𝐹 =
)9
)!
. 𝛿𝑥 +
-
.
.
)!9
)!! . 𝛿𝑥.
¨ And : δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥, 𝑡 . 𝛿𝑊
¨ So keeping the terms linear in time and in the stochastic driver
¨ 𝛿𝑥. = 𝑏.. 𝛿𝑡
¨ And we get the usual Ito formula: 𝛿𝐹 =
)9
)!
. 𝛿𝑥 +
-
.
.
)!9
)!! . 𝑏.. 𝛿𝑡
¨ The	“canonical”	example	is	usually	𝐹 𝑥 = 𝐿𝑛(𝑥),	with	
)9
)!
=
-
!
,	and	
)!9
)!! =
:-
!!,	and	with	
the	stochastic	process:	δ𝑥 = 𝑥. 𝛿𝑊,	so	𝑏 = 𝑥,	and	𝑏. = 𝑥.
¨ And	so	we	get:	𝛿 𝐿𝑁 𝑥 =
-
!
. 𝛿𝑥 +
-
.
.
:-
!! . 𝑏.. 𝛿𝑡 =
-
!
. 𝑥. 𝛿𝑊 +
-
.
.
:-
!! . 𝑥.. 𝛿𝑡
48
Luc_Faucheux_2020
Ito and Stratonovitch “Taylor” expansion - IV
¨ Or again:
¨ 𝛿 𝐿𝑁 𝑥 = 𝛿𝑧 −
-
.
. 𝛿𝑡 =
;!
!
−
-
.
. 𝛿𝑡
¨ Whereas the “usual rule of calculus would read : 𝛿 𝐿𝑁 𝑥 =
;!
!
¨ That is the usual example in most textbooks, especially in Finance
¨ NOW comes the weird little Stratanovitch trick, where we assume that we can write
something like :
¨ Strato guessed that you can write 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊
as δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥 + 𝛿𝑥/2, 𝑡 . 𝛿𝑊
¨ So again we follow the “Ito” rule of calculus by expanding in linear terms in time and in the
stochastic driver
49
Luc_Faucheux_2020
Ito and Stratonovitch “Taylor” expansion - V
¨ 𝛿𝐹 =
)9
)!
. 𝛿𝑥 +
-
.
.
)!9
)!! . 𝛿𝑥.
¨ And : δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥 + 𝛿𝑥/2, 𝑡 . 𝛿𝑊
¨ Or: δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥, 𝑡 . 𝛿𝑊 +
;!
.
.
)5
)!
. 𝛿𝑊, which keeping only the terms linear in time
and in the stochastic driver 𝛿𝑊, leads to:
¨ δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥, 𝑡 . 𝛿𝑊 +
-
.
.
)5
)!
. 𝑏. 𝛿𝑡
¨ And so: 𝛿𝐹 =
)9
)!
. 𝛿𝑥 +
-
.
.
)!9
)!! . 𝛿𝑥.
¨ 𝛿𝐹 =
)9
)!
. 𝛿𝑥 +
-
.
.
)!9
)!! . 𝑏.. 𝛿𝑡 +
-
.
.
)9
)!
.
)5
)!
. 𝑏. 𝛿𝑡
50
Luc_Faucheux_2020
Ito and Stratonovitch “Taylor” expansion - VI
¨ So we think that within the Ito rules of calculus (doing a Taylor expansion and keeping only
the terms linear in time and the stochastic driver) but following the Stratonovitch
convention we can write something like:
¨ 𝛿𝐹 =
)9
)!
. 𝛿𝑥 +
-
.
.
)!9
)!! . 𝑏.. 𝛿𝑡 +
-
.
.
)9
)!
.
)5
)!
. 𝑏. 𝛿𝑡
¨ The	“canonical”	example	again	is	𝐹 𝑥 = 𝐿𝑛(𝑥),	with	
)9
)!
=
-
!
,	and	
)!9
)!! =
:-
!!,	and	with	the	
stochastic	process:	δ𝑥 = 𝑥. 𝛿𝑊,	so	𝑏 = 𝑥,	and	𝑏. = 𝑥.,	and	
)5
)!
= 1
¨ So	we	get:
¨ 𝛿 𝐿𝑁 𝑥 =
-
!
. 𝛿𝑥 −
-
.
.
-
!! . 𝛿𝑡 +
-
.
.
-
!
. 1. 𝑥. 𝛿𝑡
¨ The last two terms cancel out and we get: 𝛿 𝐿𝑁 𝑥 =
-
!
. 𝛿𝑥
¨ Which is the usual result in ”regular” (non stochastic) calculus, and we think that we now
understood stochastic calculus because textbooks are telling us that the usual rules of
calculus are preserved in Strato
51
Luc_Faucheux_2020
Ito and Stratonovitch “Taylor” expansion - VII
¨ So far we seem to be pretty happy because we found that for the canonical example, the
“usual rules of calculus” are preserved when using the Stratonovitch convention within the
Ito calculus (meaning that we are using formally a Taylor expansion, making sure to keep all
the terms linear in time and in the stochastic driver, i.e. the Ito rule of calculus, but assuming
that we are taking the “mid-point” of the jump for functions multiplying this jump, i.e. what
we think to be what the Stratonovitch convention is)
¨ We will show that nothing could be more wrong
¨ Both Ito and Stratonovitch are calculus on their own on the same footing
¨ We just cannot do the Taylor expansion within the Stratonovitch framework
¨ Not super obvious, took me a long time to be confused about it, the point is to always go
back to the fact that in stochastic calculus the only thing that you can write is an integral,
and sometimes for ease of notation we write something that looks like a Taylor expansion or
and SDE. But this is only a formal way of writing, and the fact that it looks like regular
calculus does NOT allow us to use those equations “as is”
52
Luc_Faucheux_2020
Ito and Stratonovitch “Taylor” expansion – VII - a
¨ So for example, when in Ito calculus we write the chain rule as:
¨ 𝛿𝐹 =
)9
)1
. 𝛿𝑊 +
-
.
.
)!9
)1! . 𝛿𝑊.
¨ What we are really writing, (and what we should always write from time to remind us, since
we do not have enough paper and ink to write it all the time at every step) is:
¨ 𝑓 𝑊 𝑡5 − 𝑓 𝑊 𝑡6 = ∫+3+6
+3+5 )$
)*
. ([). 𝑑𝑊(𝑡) +
-
.
∫+3+6
+3+5 )!9
)1! (𝑊 𝑡 ). 𝑑𝑡
¨ In the ”limit” of small time increments, this can be written formally as the Ito lemma:
¨ 𝛿𝑓 =
)$
)*
. 𝛿𝑊 +
-
.
.
)!9
)1! . 𝛿𝑡
¨ We will go over it once we rebuild our knowledge of stochastic calculus around the integral
¨ Here we are following mostly Thomas Mikosch “Elementary Stochastic Calculus with Finance
in View”, a wonderful little book that is at times frustrating for some of the weird notations
that he uses.
53
Luc_Faucheux_2020
Ito and Stratonovitch “Taylor” expansion - VIII
¨ A very quick manner to realize how wrong we are is to use another example:
¨ 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊
¨ with 𝑏 𝑥, 𝑡 = σ. 𝑥 ,
)5
)!
= 𝜎 , 𝑏. = 𝜎.. 𝑥. and	 𝑎 𝑥, 𝑡 = 0,
¨ 𝐹 𝑥 = 𝑥.,
)9
)!
= 2𝑥 ,
)!9
)!! = 2
¨ Using ITO, 𝛿𝐹 =
)9
)!
. 𝛿𝑥 +
-
.
.
)!9
)!! . 𝑏.. 𝛿𝑡 = 2𝑥. 𝛿𝑥 + 𝜎.. 𝛿𝑡
¨ Using Strato, 𝛿𝐹 =
)9
)!
. 𝛿𝑥 +
-
.
.
)!9
)!! . 𝑏.. 𝛿𝑡 +
-
.
.
)9
)!
.
)5
)!
. 𝑏. 𝛿𝑡
¨ We then get: 𝛿𝐹 = 2𝑥. 𝛿𝑥 + 𝜎.. 𝛿𝑡 +
-
.
. 2𝑥. σ. σ. 𝑥. 𝛿𝑡 = 2𝑥. 𝛿𝑥 + 2.𝜎.. 𝛿𝑡
¨ So neither what we call Ito and what we call Strato do follow the usual rule of calculus. That
should be a sign that we did something very wrong when we loosely used the Taylor
expansion and applied it to the Stratonovitch case, because textbooks are telling us that the
usual rules of calculus (chain rule) are ”preserved” (similar in form) in Strato
54
Luc_Faucheux_2020
Ito and Stratonovitch “Taylor” expansion - IX
¨ The answer is not completely obvious in identifying what we did wrong
¨ Conversely, someone could say it is absolutely obvious, because stochastic process are NOT
differentiable, and so any kind of Taylor expansion is wrong
¨ As a note, even though in most textbooks in Finance the Ito lemma is expressed as a Taylor
expansion using the formal rules of calculus, the rule is that with stochastic processes you
can NEVER differentiate (at least in a manner that makes sense and is safe), you can ONLY
integrate
¨ And so a more formally correct formulation of the Ito Lemma is:
¨ 𝐹 𝑊 𝑡 = 𝑇 − 𝐹 𝑊 𝑡 = 𝑆 = ∫+3<
+3= )9
)1
. 𝑑𝑊 +
-
.
∫+3<
+3= )!9
)1! . 𝑑𝑡
¨ Or also:
¨ ∫+3<
+3=
𝑑𝐹(𝑊 𝑡 ) = ∫+3<
+3= )9
)1
. 𝑑𝑊(𝑡) +
-
.
∫+3<
+3= )!9
)1! . 𝑑𝑡
55
Luc_Faucheux_2020
Ito and Stratonovitch “Taylor” expansion - X
¨ What is then the correct formulation of Stratonovitch lemma?
¨ Is that?
¨ 𝐹 𝑥 𝑡 = 𝑇 − 𝐹 𝑥 𝑡 = 𝑆 = ∫+3<
+3= )9
)!
. 𝑑𝑥 + ∫+3<
+3=
[
-
.
)!9
)!! . 𝑏. +
-
.
.
)9
)!
.
)5
)!
. 𝑏]. 𝑑𝑡
¨ Or:
¨ ∫+3<
+3=
𝑑𝐹(𝑥 𝑡 ) = ∫+3<
+3= )9
)!
. 𝑑𝑥 + ∫+3<
+3=
[
-
.
)!9
)!! . 𝑏. +
-
.
.
)9
)!
.
)5
)!
. 𝑏]. 𝑑𝑡
¨ Or to identify the distinctions between the two:
¨ STRATO(∫+3<
+3=
𝑑𝐹(𝑥 𝑡 ))=ITO(∫+3<
+3=
𝑑𝐹(𝑥 𝑡 ))+
-
.
∫+3<
+3= )9
)!
.
)5
)!
. 𝑏. 𝑑𝑡
¨ This is still wrong, as we mixed Taylor expansion using the usual rules of calculus (Ito rules)
with something completely different. We will now show what is the correct way to look at it
using integrals.
56
Luc_Faucheux_2020
Simple rules of stochastic calculus (cheat sheet)
¨ NEVER EVER EVER work with processes where the volatility is a function of the stochastic
variable
¨ If you do or have no choice, transform it into a constant volatility equation (Hull p. 320), or
find a way to set the volatility term to 0 (p.330), or give up and come up with something
different (SABR model)
¨ If you are working in Finance (and ”discrete processes) -> use ITO
¨ If you are working with a DIGITAL computer -> use ITO
¨ If you are working with an ANALOG computer -> use STRATO
¨ If you are working in physics, and you do not like to break the time invariance, and you are
also not that smart, so you want the usual rules of calculus -> use STRATO
¨ In ALL cases, especially when working with the SDE, and discrete computer simulations,
ALWAYS check that your drift has not been “polluted” by the variable diffusion (“spurious”
drift, Ryter 1980)
57
Luc_Faucheux_2020
58
Building our knowledge of stochastic
calculus around the integral
Luc_Faucheux_2020
We have to start from the basics
¨ Because clearly using ITO lemma as a Taylor expansion where we keep certain terms and
using still the usual rules of calculus is wrong.
¨ As we hinted at it, we should never write something that looks like a differential but always
as an integral.
¨ Let’s now go through that derivation, and start with the usual Riemann integral in the usual
calculus
¨ We then show that extending that concept to a stochastic variable is not well defined, and
needs a convention, or interpretation of the integral. ITO is one interpretation, STRATO is
another one.
¨ We show the correspondence between the two interpretations.
¨ Extending the concept of the integral being a limit of sums subject to an interpretation, we
then derive the ITO lemma as well as the STRATO lemma
59
Luc_Faucheux_2020
“Regular” Riemann integrals (definite integrals)
¨ Riemann integrals are the regular integrals
¨ Interval [a,b] on regular ”continuous” variable t
¨ N sections of width (𝑏 − 𝑎)/𝑁, left side 𝐿>, right side 𝑅>, and middle 𝑀>
¨ The Riemann integral ∫+36
+35
𝑓 𝑡 . 𝑑𝑡 is the limit when (𝑁 → ∞) of the Riemann sums
¨ LEFT Riemann Sum: 𝐿𝑅𝑆 =
5:6
/
∑>3-
>3/
𝑓(𝐿>)
¨ RIGHT Riemann Sum: 𝑅𝑅𝑆 =
5:6
/
∑>3-
>3/
𝑓(𝑅>)
¨ MIDDLE Riemann Sum: 𝑀𝑅𝑆 =
5:6
/
∑>3-
>3/
𝑓(𝑀>)
¨ SOMETHING Riemann Sum: 𝑆𝑅𝑆 =
5:6
/
∑>3-
>3/
𝑓(𝑆>) where 𝑆> is somewhere in the section
indexed by k, we could also define irregular partitions or “mesh” if we like
¨ All those different sums converge to the same integral
60
Luc_Faucheux_2020
61
ITO and STRATO integrals in the
simple case W(t)
Luc_Faucheux_2020
Stochastic Integrals
¨ When not integrating over a “regular” continuous variable t but over a stochastic variable X,
those sums do NOT converge to the same value. What does that even look like?
¨ So first of all we would like to write something like this ∫*3*6
*3*5
𝑓 𝑋 . 𝑑𝑋
¨ The problem is that this is not well defined, what is the path over 𝑋 that we will integrate
over? Remember 𝑋 is really 𝑋(𝑡) and is stochastic (jumps all over the place)
¨ So really the only integration we can do is over 𝑡
¨ So we are looking for something like this: ∫+3+6
+3+5
𝑓 𝑋 𝑡 . 𝑑𝑋(𝑡)
¨ When 𝑡 increases by a small amount 𝛿𝑡, 𝑋(𝑡) jumps by a small amount δ𝑋(𝑡)
¨ So, breaking the time interval into 𝑁 sections of small time increment 𝛿𝑡 =
(+5:+6)
/
¨ We looking at something like : SOMETHING = ∑>3-
>3/
𝑓(𝑋(𝑡>)). [𝑋(𝑡>?-) − 𝑋(𝑡>)]
62
Luc_Faucheux_2020
Ito Integral
¨ That something is the ITO sum, which converges to the ITO integral. Note that at this point it
is just how we define it
¨ ∫+3+6
+3+5
𝑓 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) = lim
/→@
{∑>3-
>3/
𝑓(𝑋(𝑡>)). [𝑋(𝑡>?-) − 𝑋(𝑡>)]}
¨ Note in the integral the usual (.) is replaced by ([) to explicitly indicate ITO convention
¨ This is to indicate that we take for 𝑓(𝑋(𝑡>) the value of 𝑓 𝑋 BEFORE the jump
¨ This is to be compared to the ITO lemma where 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊
gets written in discrete manner as ∆𝑥 = 𝑎 𝑥, 𝑡 . ∆𝑡 + 𝑏 𝑥, 𝑡 . ∆𝑊
or δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥, 𝑡 . 𝛿𝑊
¨ Note also that to be fairly rigorous there are a lot of conditions that need to be verified for
that “SOMETHING” to converge in a well defined manner to a well defined function
63
Luc_Faucheux_2020
Stratonovitch Integral
¨ Similar to the Stratonovitch lemma, where 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊
gets written in the discrete manner as ∆𝑥 = 𝑎 𝑥, 𝑡 . ∆𝑡 + 𝑏 𝑥 + Δ𝑥/2, 𝑡 . ∆𝑊
or δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥 + 𝛿𝑥/2, 𝑡 . 𝛿𝑊
¨ The Stratonovitch integral is defined as:
¨ ∫+3+6
+3+5
𝑓 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = lim
/→@
{∑>3-
>3/
𝑓 [𝑋(𝑡> + 𝑋(𝑡>?-)]/2). [𝑋(𝑡>?-) − 𝑋(𝑡>)]}
¨ If ITO was the “left side Riemann”, Stratonovitch is the “middle Riemann”
¨ Surprise surprise, they do NOT converge to the same value
¨ Notice in the integral the usual (.) is replaced by (∘) to indicate that we take the middle point
64
Luc_Faucheux_2020
Reverse ITO integral
¨ Similarly we can also define a Reverse ITO (OTI?) integral as
¨ ∫+3+6
+3+5
𝑓 𝑋 𝑡 . (]). 𝑑𝑋(𝑡) = lim
/→@
{∑>3-
>3/
𝑓(𝑋(𝑡>?-)). [𝑋(𝑡>?-) − 𝑋(𝑡>)]}
¨ Notice in the integral the usual (.) is replaced by (]) to indicate that we take the right side
¨ The Reverse Ito lemma would then expand 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊
in the discrete manner as ∆𝑥 = 𝑎 𝑥, 𝑡 . ∆𝑡 + 𝑏 𝑥 + Δ𝑥, 𝑡 . ∆𝑊
or δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥 + 𝛿𝑥, 𝑡 . 𝛿𝑊
¨ This would be equivalent as “reading into the future” because we would take the value after
the jump, but we do not know yet what the jump will be
¨ Stratonovitch is also sometimes said to be “similar”, as in you need to know the jump ahead
of time to evaluate the function, even though you do not know the jump ahead of time
¨ ITO is well adapted to finance and to “Martingales” (you do not know the future, expected
value is the current value). The technical term is that the ITO integral is “non-anticipating”
65
Luc_Faucheux_2020
Couple of notes here
¨ It would seem that ITO would be well suited for processes that are “discontinuous” or
“discrete” in nature (a very poor choice of words on my part), like:
¨ Discrete computer simulation (on a digital computer)
¨ Finance, gambling, games of chance
¨ Radioactive decay (number of particles is discrete and the rate only depends on the previous
state)
¨ On the other hand STRATO seems to be better suited for “continuous” processes like most
processes in Physics and Biology (diffusion, advection, chemotaxis, Brownian motors,..)
¨ Part of the confusion will arise when trying to model in a discrete fashion a “continuous”
process
66
Luc_Faucheux_2020
Conversion between ITO and Stratonovitch
¨ ∫+3+6
+3+5
𝑓 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = lim
/→@
{∑>3-
>3/
𝑓 [𝑋(𝑡> + 𝑋(𝑡>?-)]/2). [𝑋(𝑡>?-) − 𝑋(𝑡>)]}
¨ ∫+3+6
+3+5
𝑓 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = lim
/→@
{∑>3-
>3/
𝑓 𝑋(𝑡> +
[*(+"#$):*(+")]
.
). [𝑋(𝑡>?-) − 𝑋(𝑡>)]}
¨ ∫+3+6
+3+5
𝑓 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = lim
/→@
{∑>3-
>3/
𝑓 𝑋(𝑡> ). [𝑋(𝑡>?-) − 𝑋(𝑡>)]} +
lim
/→@
{∑>3-
>3/
𝑓′ 𝑋(𝑡> .
[*(+"#$):*(+")]
.
). [𝑋(𝑡>?-) − 𝑋(𝑡>)]}
¨ ∫+3+6
+3+5
𝑓 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) =
∫+3+6
+3+5
𝑓 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) + lim
/→@
{∑>3-
>3/
𝑓′ 𝑋(𝑡> . (
[*(+"#$):*(+")]
.
). [𝑋(𝑡>?-) − 𝑋(𝑡>)]}
¨ 𝑆𝑇𝑅𝐴𝑇𝑂(𝑓) = 𝐼𝑇𝑂(𝑓) + lim
/→@
{∑>3-
>3/
𝑓′ 𝑋(𝑡> . (
[*(+"#$):*(+")]
.
). [𝑋(𝑡>?-) − 𝑋(𝑡>)]}
67
Luc_Faucheux_2020
Conversion between ITO and Stratonovitch 2
¨ 𝑆𝑇𝑅𝐴𝑇𝑂(𝑓) = 𝐼𝑇𝑂(𝑓) + lim
/→@
{∑>3-
>3/
𝑓′ 𝑋(𝑡> . (
[*(+"#$):*(+")]
.
). [𝑋(𝑡>?-) − 𝑋(𝑡>)]}
¨ NOW this is a limit of a Riemann sum, because (𝛿𝑋).~𝛿𝑡 and is now deterministic (in the
case of a simple Wiener process for 𝑋 = 𝑊)
¨ In the more general case of 𝑑𝑋 𝑡 = 𝑎 𝑋 𝑡 , 𝑡 . 𝑑𝑡 + 𝑏 𝑋 𝑡 , 𝑡 . 𝑑𝑊(𝑡) we will show that
¨ (𝛿𝑋).~𝑏 𝑋 𝑡 , 𝑡 . 𝛿𝑡 but for now let’s keep 𝑋 𝑡 = 𝑊(𝑡) the simple Brownian motion
¨ That Riemann sum converges to the definite Riemann integral
-
.
∫+3+6
+3+5
𝑓′ 𝑊 𝑡 . 𝑑𝑡
¨ ∫+3+6
+3+5
𝑓 𝑊 𝑡 . (∘). 𝑑𝑊(𝑡) = ∫+3+6
+3+5
𝑓 𝑊 𝑡 . ([). 𝑑𝑊(𝑡) +
-
.
∫+3+6
+3+5
𝑓′ 𝑊 𝑡 . 𝑑𝑡
¨ 𝑆𝑇𝑅𝐴𝑇𝑂(𝑓) = 𝐼𝑇𝑂 𝑓 +
-
.
∫+3+6
+3+5
𝑓′ 𝑊 𝑡 . 𝑑𝑡
¨ 𝑆𝑇𝑅𝐴𝑇𝑂(𝑓) = 𝐼𝑇𝑂 𝑓 +
-
.
. 𝑅𝐼𝐸𝑀𝐴𝑁𝑁(𝑓%)
68
Luc_Faucheux_2020
69
ITO lemma and
Stratonovitch lemma
Luc_Faucheux_2020
Chain Rule (Ito lemma)- I
¨ ITO integral:
¨ ∫+3+6
+3+5
𝑓 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) = lim
/→@
{∑>3-
>3/
𝑓(𝑋(𝑡>)). [𝑋(𝑡>?-) − 𝑋(𝑡>)]}
¨ Another way to express it is the following:
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = ∑>3-
>3/
{𝑓(𝑋(𝑡>)) − 𝑓(𝑋(𝑡>:-))}
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim
/→@
∑>3-
>3/
{𝑓(𝑋(𝑡>)) − 𝑓(𝑋(𝑡>:-))}
¨ As long as we have defined a reasonable “mesh” for the sequence of 𝑋(𝑡>)
¨ 𝑓(𝑋(𝑡>)) = 𝑓(𝑋(𝑡>:-)) +
)$
)*
. ([). 𝛿𝑋 +
-
.
.
)!$
)!! . ([). (𝛿𝑋).
¨ Where we use the ([) notation to reflect the “left-hand” Riemann.
¨
)$
)*
. ([). 𝛿𝑋 =
)$
)*
𝑋 = 𝑋(𝑡>:- ]. [𝑋(𝑡>) − 𝑋(𝑡>:-)]
70
Luc_Faucheux_2020
Chain Rule (Ito lemma)- II
¨
-
.
.
)!9
)!! . ([). 𝛿𝑋 . =
-
.
.
)!$
)!! [𝑋 = 𝑋(𝑡>:-)]. [𝑋(𝑡>) − 𝑋(𝑡>:-)].
¨
)$
)*
. ([). 𝛿𝑋 =
)$
)*
𝑋 = 𝑋(𝑡>:- ]. [𝑋(𝑡>) − 𝑋(𝑡>:-)]
¨ 𝑓(𝑋(𝑡>)) = 𝑓(𝑋(𝑡>:-)) +
)$
)*
. ([). 𝛿𝑋 +
-
.
.
)!$
)!! . ([). (𝛿𝑋).
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim
/→@
∑>3-
>3/
{𝑓(𝑋(𝑡>)) − 𝑓(𝑋(𝑡>:-))}
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim
/→@
∑>3-
>3/
{
)$
)*
. ([). 𝛿𝑋 +
-
.
.
)!$
)!! . ([). (𝛿𝑋).}
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = ∫+3+6
+3+5 )$
)*
. ([). 𝑑𝑋(𝑡) + ∫+3+6
+3+5 -
.
.
)!$
)!! . ([). (𝛿𝑋).
¨ In the ”limit” of small time increments, this can be written formally as the Ito lemma:
¨ 𝛿𝑓 =
)$
)*
. 𝛿𝑋 +
-
.
.
)!$
)!! . (𝛿𝑋).
71
Luc_Faucheux_2020
Chain Rule (Strato lemma)- I
¨ We will now expand around the middle point
¨ We still have:
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = ∑>3-
>3/
{𝑓(𝑋(𝑡>)) − 𝑓(𝑋(𝑡>:-))}
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim
/→@
∑>3-
>3/
{𝑓(𝑋(𝑡>)) − 𝑓(𝑋(𝑡>:-))}
¨ We now write it a little differently by looking at expanding around the middle point:
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim
/→@
∑>3-
>3/
{𝑓(𝑋(𝑡>)) − 𝑓(𝑋(𝑡>:-))}
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim
/→@
∑>3-
>3/
𝑓 𝑋(𝑡> − 𝑓 ∘ + 𝑓 ∘ − 𝑓(𝑋(𝑡>:-))}
¨ Where for sake of clarity we noted : 𝑓 ∘ = 𝑓
*(+" ? *(+"%$
.
72
Luc_Faucheux_2020
Chain Rule (Strato lemma)- II
¨ We then have the two following expansions of the function around the middle point:
¨ 𝑓(𝑋(𝑡>)) = 𝑓(∘) +
)$
)*
. (∘).
;*
.
+
-
.
.
)!$
)!! . (∘). (
;*
.
).
¨ 𝑓(𝑋(𝑡>:-)) = 𝑓(∘) −
)$
)*
. (∘).
;*
.
+
-
.
.
)!$
)!! . (∘). (
;*
.
).
¨ 𝛿𝑋 = 𝑋(𝑡>) − 𝑋(𝑡>:-)
¨ 𝑋(𝑡>) −
*(+")?*(+"%$)
.
=
;*
.
¨
*(+")?*(+"%$)
.
− 𝑋(𝑡>:-) =
;*
.
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim
/→@
∑>3-
>3/
𝑓 𝑋(𝑡> − 𝑓 ∘ + 𝑓 ∘ − 𝑓(𝑋(𝑡>:-))}
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 =
lim
/→@
{∑>3-
>3/ )$
)*
. ∘ .
;*
.
+
-
.
.
)!$
)!! . ∘ .
;*
.
.
− (−
)$
)*
. (∘).
;*
.
+
-
.
.
)!$
)!! . (∘). (
;*
.
).)}
73
Luc_Faucheux_2020
Chain Rule (Strato lemma)- III
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 =
lim
/→@
{∑>3-
>3/ )$
)*
. ∘ .
;*
.
+
-
.
.
)!$
)!! . ∘ .
;*
.
.
− (−
)$
)*
. (∘).
;*
.
+
-
.
.
)!$
)!! . (∘). (
;*
.
).)}
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim
/→@
{∑>3-
>3/ )$
)*
. ∘ .
;*
.
+ −(−
)$
)*
. (∘).
;*
.
)}
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim
/→@
{∑>3-
>3/ )$
)*
. ∘ . 𝛿𝑋}
¨ And so defining the integral with the Stratonovitch convention:
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = ∫+3+6
+3+5 )$
)*
. (∘). 𝑑𝑋(𝑡)
¨ In the ”limit” of small time increments, this can be written formally as the Strato lemma:
¨ 𝛿𝑓 =
)$
)*
. ∘ . 𝛿𝑋
¨ Please note that we are keeping the notation : ∘
74
Luc_Faucheux_2020
Chain Rule (Strato lemma)- IV
¨ Using the Stratonovitch definition of the stochastic integral, we can write :
¨ 𝛿𝑓 =
)$
)*
. ∘ . 𝛿𝑋
¨ This is the usual chain rule
¨ So formally in some textbooks, you will see the following statement:
¨ “We do not mean that the Stratonovitch stochastic integral is a classical (Riemann) integral.
We only claim that the corresponding chain rules have a similar structure”. Mikosh p127
¨ It is important to note that BOTH the Ito and the Stratonovitch integrals are defined in a
mathematically correct manner.
¨ Ito rules of calculus are not the usual ones but the Ito integral is a martingale
¨ Strato rules of calculus are the usual ones (FORMALLY) but the Strato integral is NOT a
martingale
¨ We still have to review how we treat SDE and PDE in those frameworks.
75
Luc_Faucheux_2020
Chain Rule (Ito and Strato lemma)- I
¨ The crux of the matter is which interpretation of an integral do you want to use:
¨ Because lim
/→@
{∑>3-
>3/ )$
)*
. ∘ . 𝛿𝑋}
¨ And
¨ lim
/→@
∑>3-
>3/
{
)$
)*
. ([). 𝛿𝑋}
¨ Do NOT converge to the same value, unlike a regular Riemann integral
¨ In fact we just showed that
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim
/→@
{∑>3-
>3/ )$
)*
. ∘ . 𝛿𝑋}
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim
/→@
∑>3-
>3/
{
)$
)*
. ([). 𝛿𝑋 +
-
.
.
)!9
)!! . ([). (𝛿𝑋).}
76
Luc_Faucheux_2020
Chain Rule (Ito and Strato lemma)- II
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim
/→@
{∑>3-
>3/ )$
)*
. ∘ . 𝛿𝑋}
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim
/→@
∑>3-
>3/
{
)$
)*
. ([). 𝛿𝑋 +
-
.
.
)!9
)!! . ([). (𝛿𝑋).}
¨ We formally define in integral form:
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = ∫+3+&
+3+'
𝑑𝐹(𝑥 𝑡 )
¨ And ITO:
¨ ∫+3+6
+3+5
𝐹 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) = lim
/→@
{∑>3-
>3/
𝐹(𝑋(𝑡>)). [𝑋(𝑡>?-) − 𝑋(𝑡>)]}
¨ And STRATO:
¨ ∫+3+6
+3+5
𝐹 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = lim
/→@
{∑>3-
>3/
𝐹 [𝑋(𝑡> + 𝑋(𝑡>?-)]/2). [𝑋(𝑡>?-) − 𝑋(𝑡>)]}
77
Luc_Faucheux_2020
Chain Rule (Ito and Strato lemma)- III
¨ We then have for the chain rule using the ITO interpretation of the integral:
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim
/→@
∑>3-
>3/
{
)$
)*
. ([). 𝛿𝑋 +
-
.
.
)!9
)!! . ([). (𝛿𝑋).}
¨ ∫+3+&
+3+'
𝑑𝐹(𝑋 𝑡 ) = ∫+3+6
+3+5 )$
)*
. ([). 𝑑𝑋(𝑡) +
-
.
∫+3+6
+3+5 )!9
)!! 𝑋 𝑡 . (𝑑𝑋).
¨ Note that all integrals are on : ∫+3+&
+3+'
()
¨ HOWEVER, the increment of integrands are different, 𝑑𝐹, 𝑑𝑋 and 𝑑𝑡
¨ BOTH 𝐹 and 𝑋 are functions of 𝑡 ultimately, 𝐹(𝑋 𝑡 ) and 𝑋(𝑡)
¨ So only in a formal manner we write:
¨ 𝛿𝑓 =
)$
)*
. 𝛿𝑋 +
-
.
.
)!9
)!! . (𝛿𝑋)., which is the celebrated Ito lemma (chain rule)
78
Luc_Faucheux_2020
Chain Rule (Ito and Strato lemma)- IV
¨ 𝛿𝑓 =
)$
)*
. 𝛿𝑋 +
-
.
.
)!9
)!! . 𝑏.. 𝛿𝑡, which is the celebrated Ito lemma (chain rule)
¨ We somehow convinced ourselves at first that this was just using regular Taylor expansion
but just keeping higher order terms, in particular the second order in 𝑋, because:
¨ 𝛿𝑋. 𝛿𝑋~𝑏.. 𝛿𝑡
¨ But even though it looks formally the same, the only rigorous manner in which to write it is:
¨ ∫+3+&
+3+'
𝑑𝐹(𝑋 𝑡 ) = ∫+3+6
+3+5 )$
)*
. ([). 𝑑𝑋(𝑡) +
-
.
∫+3+6
+3+5 )!9
)!! 𝑋 𝑡 . 𝑏 𝑋 𝑡 , 𝑡 .. 𝑑𝑡
¨ With the ITO interpretation of the integral:
¨ ∫+3+6
+3+5
𝐹 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) = lim
/→@
{∑>3-
>3/
𝐹(𝑋(𝑡>)). [𝑋(𝑡>?-) − 𝑋(𝑡>)]}
79
Luc_Faucheux_2020
Chain Rule (Ito and Strato lemma)- V
¨ We then have for the chain rule using the STRATO interpretation of the integral:
¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim
/→@
{∑>3-
>3/ )$
)*
. ∘ . 𝛿𝑋}
¨ ∫+3+&
+3+'
𝑑𝐹(𝑋 𝑡 ) = ∫+3+6
+3+5 )$
)*
. ∘ . 𝑑𝑋(𝑡)
¨ Note that all integrals are on : ∫+3+&
+3+'
()
¨ HOWEVER, the increment of integrands are different, 𝑑𝐹, 𝑑𝑋 and 𝑑𝑡
¨ BOTH 𝐹 and 𝑋 are functions of 𝑡 ultimately, 𝐹(𝑋 𝑡 ) and 𝑋(𝑡)
¨ So only in a formal manner we write:
¨ 𝛿𝑓 =
)$
)*
. 𝛿𝑋, which is the celebrated STRATO lemma (chain rule)
80
Luc_Faucheux_2020
Chain Rule (Ito and Strato lemma)- VI
¨ 𝛿𝑓 =
)$
)*
. 𝛿𝑋, which is the celebrated STRATO lemma (chain rule)
¨ HOWEVER, the only rigorous manner in which to write it is:
¨ ∫+3+&
+3+'
𝑑𝐹(𝑋 𝑡 ) = ∫+3+6
+3+5 )$
)*
. ∘ . 𝑑𝑋(𝑡)
¨ With the STRATO interpretation of the integral:
¨ ∫+3+6
+3+5
𝐹 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = lim
/→@
{∑>3-
>3/
𝐹 [𝑋(𝑡> + 𝑋(𝑡>?-)]/2). [𝑋(𝑡>?-) − 𝑋(𝑡>)]}
81
Luc_Faucheux_2020
Chain Rule (Ito and Strato lemma)- VII
¨ So sometimes we will just stick to the notation:
¨ Ito lemma:
¨ 𝛿𝑓 =
)$
)*
. ([). 𝛿𝑋 +
-
.
.
)!9
)!! . ([). 𝛿𝑋.
¨ Just to remind us that we are using stochastic calculus and that it is NOT a usual product
¨ Similarly when using STRATO we will sometimes use:
¨ 𝛿𝑓 =
)$
)*
. ∘ . 𝛿𝑋
¨ To remind us that even if it looks like the regular chain rule, we are in the stochastic world
where things are a little weird.
¨ Again the only rigorous manner to deal with those is to always go back to the integrals, and
the interpretation of is as a limit of a sum, which is rigorous.
82
Luc_Faucheux_2020
Chain Rule (Ito and Strato lemma)- VIII
¨ Note that if the function 𝐹(𝑋 𝑡 ) has an explicit dependency on time 𝐹(𝑋 𝑡 , 𝑡)
¨ Then for the time part the regular chain rule applies and we will obtain terms of the
expression:
)9
)+
. 𝑑𝑡 in BOTH ITO and STRATO
¨ Because integrating over time is a normal Riemann integral, and both sums converge to the
same limit
¨ ∫+3+6
+3+5 )9
)+
. ∘ . 𝑑𝑡 = ∫+3+6
+3+5 )9
)+
. [ . 𝑑𝑡 = ∫+3+6
+3+5 )9
)+
. 𝑑𝑡
83
Luc_Faucheux_2020
84
Leibniz rule
Luc_Faucheux_2020
Leibniz rule - I
¨ In most textbooks, it is usually also presented as
¨ Hey do a Taylor expansion
¨ Just make sure to keep the higher order terms
¨ And you good
¨ Surprisingly it is formally the same expression
¨ Even more surprisingly, if you were to be working in a Strato world, textbooks would say:
¨ Hey just use regular calculus
¨ Since we have seen now that stochastic calculus can be tricky, let’s spend some time on
convincing ourselves that we can adapt the Leibniz rule in stochastic calculus (it will also be
super useful when changing numeraires in the Numeraire deck)
85
Luc_Faucheux_2020
Leibniz rule - II
¨ It will be easier to do it once we have a more general expression for the ITO and STRATO
integrals, so for now we will just state them (Leibniz rule for first order)
¨ Leibniz rule in the REGULAR calculus:
¨ 𝛿 𝑓𝑔 = 𝑔
)$
)*
. 𝛿𝑋 + 𝑓
)(
)*
. 𝛿𝑋
¨ Leibniz rule in the ITO calculus:
¨ 𝛿(𝑓𝑔) = 𝑔
)$
)*
. ([). 𝛿𝑋 + 𝑔
-
.
.
)!$
)*! . ([). 𝛿𝑋. + 𝑓
)(
)*
. ([). 𝛿𝑋 + 𝑓
-
.
.
)!(
)*! . ([). 𝛿𝑋. +
)(
)*
.
)$
)*
. ([). 𝛿𝑋.
¨ Leibniz rule in the STRATO calculus:
¨ 𝛿 𝑓𝑔 = 𝑔
)$
)*
. ∘ . 𝛿𝑋 + 𝑓
)(
)*
. ∘ . 𝛿𝑋
86
Luc_Faucheux_2020
87
A simple example
Integrating X
Luc_Faucheux_2020
A worked out example to gain some intuition
¨ ∫+3+6
+3+5
𝑓 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) = lim
/→@
{∑>3-
>3/
𝑓(𝑋(𝑡>)). [𝑋(𝑡>?-) − 𝑋(𝑡>)]}
¨ This is the definition of the Ito integral
¨ Let’s try with the simple case of the function 𝑓(𝑋(𝑡>)) = 𝑋(𝑡>)
¨ The Riemann sum is: 𝑆/ = ∑>3-
>3/
𝑓(𝑋(𝑡>)). [𝑋(𝑡>?-) − 𝑋(𝑡>)]
¨ For this specific case: 𝑆/ = ∑>3-
>3/
𝑋(𝑡>). [𝑋(𝑡>?-) − 𝑋(𝑡>)]
88
Luc_Faucheux_2020
In the “regular” case – Riemann integral
¨ ∫!3!6
!3!5
𝑓 𝑥 . 𝑑𝑥 = lim
/→@
{∑>3-
>3/
𝑓(𝑥>). [(𝑥>?-) − (𝑥>)]}
¨ This is the usual “triangle” representation
89
Luc_Faucheux_2020
In the “regular” case – Riemann integral - II
¨ In the regular case,
¨ (𝑥>?-) − (𝑥>) = 𝛿𝑥 =
!':!&
/
=
;*
/
¨ Another way to look at it, is 𝑥 = 𝑡, so at every point in time, δ𝑥 = 𝛿𝑡
¨ In particular, δ𝑥. = δ𝑡.
90
Xi i+1i-1
F(X) to integrate
Luc_Faucheux_2020
In the stochastic case, we cannot draw that picture
¨ We cannot write (𝑥>?-) − (𝑥>) = 𝛿𝑥 =
!':!&
/
=
;*
/
¨ If anything, what we can write is (𝑥>?-) − (𝑥>) = 𝛿𝑥> = ±𝛿𝑥
¨ And so ∑(𝑥>?-) − (𝑥>) = ∑ 𝛿𝑥> = 0
¨ And ∑[(𝑥>?-) − (𝑥>)].= ∑ 𝛿𝑥>
.
= ∑ 𝛿𝑥. = 𝑁. 𝛿𝑥.
91
Xi i+1i-1
Luc_Faucheux_2020
Another way to think about the difference
¨ The regular case, the horizontal axis is the regular non-stochastic variable
¨ Any integration of function is the regular Riemann integral
¨ In the stochastic case, the horizontal axis becomes time, and the vertical is the stochastic
variable X, and we will need to integrate a function of it over the vertical axis, but whereas
time is regular, a stochastic process cannot be such that each step is just the interval divided
by the number of steps
92
X(t) is stochastic
Time t is regular
F(X(t))tointegrate
Luc_Faucheux_2020
The regular case revisited
¨ ∫!3!6
!3!5
𝑓 𝑥 ([)𝑑𝑥 = lim
/→@
{∑>3-
>3/
𝑓(𝑥>). [(𝑥>?-) − (𝑥>)]}
¨ ∫!3!6
!3!5
𝑓 𝑥 ([)𝑑𝑥 = lim
/→@
∑>3-
>3/
𝑓(𝑥>). 𝛿𝑥> = lim
/→@
!':!&
/
. ∑>3-
>3/
𝑓(𝑥>)
¨ In the case where 𝑓(𝑥>) = 𝑥> = 𝑘. 𝛿𝑥 = 𝑘.
!':!&
/
¨ ∫!3!6
!3!5
𝑥([)𝑑𝑥 = lim
/→@
!':!&
/
.
!':!&
/
. ∑>3-
>3/
𝑘 = lim
/→@
!':!&
/
.
!':!&
/
.
/(/?-)
.
¨ ∫!3!6
!3!5
𝑥([)𝑑𝑥 = (𝑥5 − 𝑥6). lim
/→@
-
/
.
-
/
.
/(/?-)
.
¨ Using 𝑥6 = 0 without losing any generality,
¨ ∫!34
!3*
𝑥([)𝑑𝑥 =
-
.
𝑋., which is the usual result
93
Luc_Faucheux_2020
The regular case revisited – II – Strato convention
¨ ∫!3!6
!3!5
𝑓 𝑥 (∘)𝑑𝑥 = lim
/→@
{∑>3-
>3/
𝑓(
!"#$?!"
.
). [(𝑥>?-) − (𝑥>)]}
¨ ∫!3!6
!3!5
𝑓 𝑥 (∘)𝑑𝑥 = lim
/→@
∑>3-
>3/
𝑓(
!"#$?!"
.
). 𝛿𝑥> = lim
/→@
!':!&
/
. ∑>3-
>3/
𝑓(
!"#$?!"
.
)
¨ In the case where𝑓(
!"#$?!"
.
) =
!"#$?!"
.
= (𝑘 +
-
.
). 𝛿𝑥 = (𝑘 +
-
.
).
!':!&
/
¨ ∫!3!6
!3!5
𝑥(∘)𝑑𝑥 = lim
/→@
!':!&
/
.
!':!&
/
. ∑>3-
>3/
(𝑘 +
-
.
) = lim
/→@
!':!&
/
.
!':!&
/
. [
/ /?-
.
+
/
.
]
¨ ∫!3!6
!3!5
𝑥(∘)𝑑𝑥 = (𝑥5 − 𝑥6). lim
/→@
-
/
.
-
/
. [
/ /?-
.
+
/
.
]
¨ Using 𝑥6 = 0 without losing any generality,
¨ ∫!34
!3*
𝑥(∘)𝑑𝑥 =
-
.
𝑋., which is the usual result
94
Luc_Faucheux_2020
The regular case revisited – III
¨ So in the regular case, the Riemann integral gives the same result, irrespective of where
inside the small interval we pick the value of the function
¨ ∫!34
!3*
𝑥([)𝑑𝑥 =
-
.
𝑋. = ∫!34
!3*
𝑥(∘)𝑑𝑥
¨ In essence, this is because the terms ∑>3-
>3/
. 𝛿𝑥> scale like 1, and so any terms in ∑>3-
>3/
. 𝛿𝑥>
.
will go to 0 in the limit 𝑁 → ∞
¨ The “quadratic variation” ∑>3-
>3/
. 𝛿𝑥>
.
does not add up to any finite number in the limit.
¨ In the case of a stochastic variable, the terms ∑>3-
>3/
. 𝛿𝑥> scale like 0, because 𝛿𝑥> = ±𝛿𝑥,
and the terms like ∑>3-
>3/
. 𝛿𝑥>
.
will actually scale like time.
¨ The “quadratic variation” ∑>3-
>3/
. 𝛿𝑥>
.
is said to scale linearly with time (some textbooks use
the expression additive with time) and will not “disappear” to 0 in the limit.
95
Luc_Faucheux_2020
Back to the worked-out example
¨ The Riemann sum is: 𝑆/ = ∑>3-
>3/
𝑋(𝑡>). [𝑋(𝑡>?-) − 𝑋(𝑡>)]
¨ We can expand: [𝑋(𝑡>?-) − 𝑋(𝑡>)].= 𝑋(𝑡>?-). + 𝑋(𝑡>). − 2. 𝑋(𝑡>). 𝑋(𝑡>?-)
¨ And: 𝑋(𝑡>). [𝑋(𝑡>?-) − 𝑋(𝑡>)] = 𝑋(𝑡>). 𝑋(𝑡>?-) − 𝑋(𝑡>).
¨ And : 𝑋(𝑡>). [𝑋(𝑡>?-) − 𝑋(𝑡>)] =
-
.
. [𝑋(𝑡>?-). + 𝑋(𝑡>). − [𝑋(𝑡>?-) − 𝑋(𝑡>)].−2. 𝑋(𝑡>).]
¨ Or: 𝑋(𝑡>). [𝑋(𝑡>?-) − 𝑋(𝑡>)] =
-
.
. [𝑋(𝑡>?-). − 𝑋(𝑡>).] −
-
.
. [𝑋(𝑡>?-) − 𝑋(𝑡>)].
¨ So: 𝑆/ = ∑>3-
>3/
𝑋(𝑡>). [𝑋(𝑡>?-) − 𝑋(𝑡>)] =
-
.
. 𝑋(𝑡/?-). −
-
.
∑>3-
>3/
[𝑋(𝑡>?-) − 𝑋(𝑡>)].
¨ The first term is the expected
*!
.
¨ The second term would usually disappear when the variable X is regular, because the sum
would scale as 𝑁. (
-
/
).~
-
/
which will tend to 0 when 𝑁 → ∞
96
Luc_Faucheux_2020
Back to the worked-out example - II
¨ 𝑆/ = ∑>3-
>3/
𝑋(𝑡>). [𝑋(𝑡>?-) − 𝑋(𝑡>)] =
-
.
. 𝑋(𝑡/?-). −
-
.
∑>3-
>3/
[𝑋(𝑡>?-) − 𝑋(𝑡>)].
¨ We define 𝑄/ = ∑>3-
>3/
[𝑋(𝑡>?-) − 𝑋(𝑡>)].
¨ We can use here the simple “binary” assumption that [𝑋(𝑡>?-) − 𝑋(𝑡>)].= (𝑡>?-−𝑡>)
¨ Note that this is capturing the essence of the matter
¨ A more accurate and general way to go about this would be to assume the following:
¨ {𝑋(𝑡>?-) − 𝑋(𝑡>)} follows the 𝑁(0, (𝑡>?-−𝑡>)) distribution
¨ We would then estimate the expected value of 𝑄/
¨ 𝐸 𝑄/ = ∑>3-
>3/
𝐸[𝑋(𝑡>?-) − 𝑋(𝑡>)]. = ∑>3-
>3/
(𝑡>?-−𝑡>) = 𝑡/?-
¨ We would have to show then that: lim
/→@
𝑄/ = 𝐸[𝑄/]
¨ This opens up the can of worms of how you define convergence (convergence in
distribution, in probability, almost sure convergence, Lp-convergence). Will try to
incorporate later, but trying to keep it simple to not hide the intuition behind notations
97
Luc_Faucheux_2020
What did we get?
¨ ∫+34
+3=
𝑓 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) = lim
/→@
{∑>3-
>3/
𝑓(𝑋(𝑡>)). [𝑋(𝑡>?-) − 𝑋(𝑡>)]}
¨ In the simple case 𝑓 𝑋(𝑡> = 𝑋(𝑡>)
¨ ∫+34
+3=
𝑋(𝑡). ([). 𝑑𝑋(𝑡) =
-
.
𝑋. −
-
.
𝑇
¨ A couple of notes:
¨ Ito integral does NOT recover the usual rules of calculus
¨ Ito integral is called a martingale, meaning if X is a martingale (E[X]=0), then the Ito integral
of a function f(X) is also a martingale
¨ 𝐸 𝑋 = 0 and 𝐸[∫+34
+3=
𝑋 𝑡 . [). 𝑑𝑋 𝑡 = 𝐸
-
.
𝑋. −
-
.
𝑇 = 𝐸
-
.
𝑋. −
-
.
𝑇 = 0
¨ So 𝐸 𝑋. = 𝑇
¨ We recover the usual variance of the Brownian motion
98
Luc_Faucheux_2020
A couple more notes
¨ Let’s recall the relationship between the Ito and Stratonovitch integral (we can also work it
out from the Riemann sum)
¨ ∫+3+6
+3+5
𝑓 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = ∫+3+6
+3+5
𝑓 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) +
-
.
∫+3+6
+3+5
𝑓′ 𝑋 𝑡 . 𝑑𝑡
¨ Here, 𝑓 𝑋 = 𝑋 so 𝑓% 𝑋 = 1
¨ ∫+34
+3=
𝑓 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = ∫+34
+3=
𝑓 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) +
-
.
∫+34
+3=
1. 𝑑𝑡
¨ And we have: ∫+34
+3=
𝑋(𝑡). ([). 𝑑𝑋(𝑡) =
-
.
𝑋. −
-
.
𝑇
¨ So ∫+34
+3=
𝑓 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) =
-
.
𝑋.
¨ This is the usual rule of calculus
¨ The Stratonovitch integral preserves the usual rule of calculus
99
Luc_Faucheux_2020
100
What does the usual exponential even
mean in stochastic calculus?
The ITO exponential
Luc_Faucheux_2020
A couple more notes - II
¨ The Stratonovitch integral is NOT a martingale (this makes sense since taking the mid point
introduces correlation, and thus the terms in the product are NOT independent)
¨ Because the Ito lemma does NOT recover the usual rules of calculus, what a function
actually means has to be at times redefined.
¨ For example, we all know the exponential function 𝑓 𝑡 = exp(𝑡)
¨ It is such that : 𝑓% 𝑡 = 𝑓(𝑡)
¨ However we would not expect this function to preserve the same property when dealing
with a stochastic variable (another way to say it is that this function is convex, and so the
Jensen inequality will introduce a convexity correction, see the lecture on options)
¨ We know that Stratonovitch will preserve the rules of calculus
¨ So ∫+34
+3=
𝑒𝑥𝑝 𝑋 𝑡 . ∘ . 𝑑𝑋 𝑡 = exp(𝑋(𝑇))
¨ And ∫+34
+3=
𝑓 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = ∫+34
+3=
𝑓 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) +
-
.
∫+34
+3=
𝑓′ 𝑋 𝑡 . 𝑑𝑡
101
Luc_Faucheux_2020
The Ito exponential function
¨ ∫+34
+3=
𝑓 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = 𝑒𝑥𝑝 𝑋 𝑇 = ∫+34
+3=
𝑓 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) +
-
.
∫+34
+3=
𝑓′ 𝑋 𝑡 . 𝑑𝑡
¨ 𝑒𝑥𝑝 𝑋 𝑇 = ∫+34
+3=
𝑓 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) +
-
.
∫+34
+3=
𝑒𝑥𝑝 𝑋 𝑡 . 𝑑𝑡
¨ Since the second term is always positive, we have
¨ ∫+34
+3=
𝑒𝑥𝑝 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) <> 𝑒𝑥𝑝 𝑋 𝑇
¨ So clearly under Ito, the usual exponential function does not verify 𝐹% 𝑋 = 𝐹(𝑋)
¨ For that reason, the convention is to rename some of the functions by specifying “Ito” in
front of them
¨ This can be confusing at times
102
Luc_Faucheux_2020
The Ito exponential function -II
¨ Ito lemma has : 𝛿𝐹 =
)9
)*
. 𝛿𝑋 +
-
.
.
)!9
)*! . 𝛿𝑋. +
)9
)+
. 𝛿𝑡
¨ With 𝐹 𝑋 = exp 𝑋 , 𝐹% 𝑋 = exp 𝑋 , 𝐹%% 𝑋 = exp(𝑋)
¨ We get 𝛿𝐹 = exp 𝑋 . 𝛿𝑋 +
-
.
. exp 𝑋 . 𝛿𝑋. = exp(𝑋). 𝛿𝑋 +
-
.
. exp(𝑋). 𝛿𝑡
¨ With 𝐹 𝑋, 𝑡 = exp(𝑋 −
+
.
) ,
)9
)*
= exp 𝑋 −
+
.
,
)!9
)*! = exp 𝑋 −
+
.
,
)9
)+
=
:-
.
. exp(𝑋 −
+
.
)
¨ We get 𝛿𝐹 = exp 𝑋 −
+
.
. 𝛿𝑋 +
-
.
. exp 𝑋 −
+
.
. 𝛿𝑋. −
-
.
. exp 𝑋 −
+
.
. 𝛿𝑡
¨ Or: 𝛿𝐹 = exp 𝑋 −
+
.
. 𝛿𝑋 = 𝐹 𝑋, 𝑡 . 𝛿𝑋
¨ So you will see sometimes the function 𝐹 𝑋, 𝑡 = exp 𝑋 −
+
.
being referred to as the Ito-
exponential, whereas the regular exponential is of course 𝐹 𝑋 = exp 𝑋
103
Luc_Faucheux_2020
The Ito exponential function -III
¨ So this is another indication that we should be careful with using regular functions and some
of their properties.
¨ In regular calculus, the exponential function 𝑓 𝑥 = exp(𝑥) is such that : 𝑓% 𝑥 = 𝑓(𝑥)
¨ In ITO calculus, the “ITO exponential function” that still verifies: 𝐹% 𝑋(𝑡) = 𝐹(𝑋(𝑡)) is given
by: 𝐹 𝑋(𝑡), 𝑡 = exp 𝑋(𝑡) −
+
.
¨ In STRATO calculus, the “STRATO exponential function” that still verifies: 𝐹% 𝑋(𝑡) =
𝐹(𝑋(𝑡)) is given by: 𝐹 𝑋(𝑡), 𝑡 = exp 𝑋(𝑡)
¨ Note that we are somehow lucky that those functions are still local (i.e. depends only on the
value of 𝑋(𝑡) and 𝑡. It is not clear that we could not have ended up with a more
complicated function that depends on the history or the path ∫+34
+3=
𝑋 𝑡 . 𝑑𝑡 for example.
104
Luc_Faucheux_2020
A nice little summary
Name of Integral RIEMANN ITO STRATONOVITCH
Type Deterministic Stochastic Stochastic
Points of integration Doesn’t matter LEFT MIDDLE
Martingale NO YES NO
Non-Anticipating NO YES NO
Usual Calculus Rule YES NO YES(*)
Main Usage Everywhere Finance Physics
105
¨ (*) ONLY from a formal point of view. This is still a stochastic integral and a nasty beast not
to be messed with “à la légère”
Luc_Faucheux_2020
106
Back to trying to “integrate” X
Luc_Faucheux_2020
Another example: can you integrate X ?
¨ Somewhat cultural side note: In French “to integrate” (or “integrer”) is the same word to
perform a mathematical integration or to be accepted in a school.
¨ One of the most prestigious schools is called “Polytechnique” or “X” because of the logo on
their hats of two crossed swords (it was created and is still technically a military school, a
little like West Point here, but starts around the junior college level)
¨ So anyways, after high school, there are special schools called “classes preparatoires” where
you just study for the entrance exam to those advanced schools (“hautes ecoles”), with
names like X, Ecole Normale Superieure (Normal Superieure school) or others with names
that are usually tied to their original concentration: Ponts et Chaussees (bridges and
sidewalks), Mines (mining), Supelec (Superior Electricity), and so on
¨ Usually students spend around 2 years studying in “classes preparatoires” before taking the
entrance exam (everything is an exam, there is no application, you take the exam, you get
ranked and then the school offers you a spot in the order of ranking).
¨ So you get the idea, the question is how many years does it take you to integrate X
¨ 50 years later, people will still remember what “halfs” they were, or they will lie about it
107
Luc_Faucheux_2020
Another example: can you integrate X ? - II
¨ If you are a genius and if it takes you only one year to integrate X, the student is called a ”one
half” because: ∫*34
*3-
𝑋. 𝑑𝑋 = [
*!
.
]*34
*3-
=
-
.
¨ I have heard of “one-half” I have never met one.
¨ This is not the same as “half-bloods” from Harry Potter, I have never met one of those either
¨ If you are a decent student and if it takes you two years to integrate X, the student is called a
”three half” because: ∫*3-
*3.
𝑋. 𝑑𝑋 = [
*!
.
]*3-
*3.
=
C
.
−
-
.
=
D
.
¨ (I was a 3/2, but I did not integrate X, my father did, my brother did, I am the black sheep of
the family)
¨ If you are a decent student and if it takes you three years to integrate X, the student is called a
”five half” because: ∫*3.
*3D
𝑋. 𝑑𝑋 = [
*!
.
]*3.
*3D
=
E
.
−
C
.
=
F
.
¨ If it takes you four years to integrate X, the student is called a ”seven half” because:
∫*3D
*3C
𝑋. 𝑑𝑋 = [
*!
.
]*3D
*3C
=
-G
.
−
E
.
=
H
.
¨ Being called a “seven half” is an insult
108
Luc_Faucheux_2020
Another example: can you integrate X ? - III
¨ Oh also if you integrate 𝑋 not in 𝑑𝑋 but in 𝑑𝑡, you are called “endette”, or in debt
¨ OK, so that was a little digression
¨ Let’s get back to the matter at hands.
¨ If 𝑋 is not stochastic (also called deterministic), this is the usual calculus that we are used to,
and so:
¨ ∫ 𝑋. 𝑑𝑋 =
*!
.
+ 𝐶
¨ What happens if 𝑋 is now a stochastic variable?
¨ The truth is that I have never met anyone who can integrate a stochastic variable (I have also
never met a “one half”), and so usually people always treat the problem from the other way,
you start with a guess of what the integral is, you use ITO lemma and then you check in an
iterative manner
109
Luc_Faucheux_2020
Another example: can you integrate X ? - IV
¨ If we have a function 𝐹(𝑋), Ito lemma is the following:
¨ Ito (1951) guessed that you can write 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑧 as δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 +
𝑏 𝑥, 𝑡 . 𝛿𝑧
¨ If 𝐹(𝑥) a funcyon of x, the corresponding SDE as a result of “Taylor” expansion in ITO is:
¨ Bear in mind that this is really not a Taylor expansion, it is just ITO lemma that looks like a
Taylor expansion where you kept some terms and not others
¨ 𝛿𝐹 =
)9
)!
. 𝛿𝑥 +
-
.
.
)!9
)!! . 𝑏.. 𝛿𝑡
¨ 𝛿𝐹 =
)9
)*
. 𝛿𝑋 +
-
.
.
)!9
)*! . 𝛿𝑋.
¨ Let’s use 𝐹 𝑋 = 𝑋. as a good guess
¨ 𝐹 𝑋 = 𝑋.,
)9
)*
= 2𝑋,
)!9
)*! = 2, we then get: 𝛿 𝑋. = 2𝑋. 𝛿𝑋 +
-
.
. 2. 𝛿𝑋.
110
Luc_Faucheux_2020
Another example: can you integrate X ? - V
¨ 𝛿 𝑋. = 2𝑋. 𝛿𝑋 +
-
.
. 2. 𝛿𝑋.
¨ In the case of the Geometric Brownian motion (Hull): 𝑑𝑋 = 𝜎𝑋𝑑𝑊
¨ 𝛿 𝑋. = 2𝑋. 𝛿𝑋 +
-
.
. 2. (𝜎𝑋). 𝛿𝑡
¨ 𝛿 𝑋. = 2𝑋. 𝛿𝑋 + (𝜎𝑋). 𝛿𝑡
¨ And so: ∫ 𝑋. 𝑑𝑋 = ∫{
; *!
.
−
-
.
. (𝜎𝑋). 𝛿𝑡}
¨ ∫ 𝑋. 𝑑𝑋 = [
*!
.
] −
-
.
∫(𝜎𝑋). 𝛿𝑡
¨ With the usual convention that 𝑋 𝑡 = 0 = 0, in the ITO convention:
¨ ∫+34
+3=
𝑋. 𝑑𝑋 =
*(=)!
.
−
-
.
∫+34
+3=
(𝜎𝑋). 𝛿𝑡
111
Luc_Faucheux_2020
Another example: can you integrate X ? - VI
¨ Things are a little different in the Stratonovitch convention:
¨ Strato guessed that you can write 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊
as δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥 + 𝛿𝑥/2, 𝑡 . 𝛿𝑊
¨ The SDE for 𝐹(𝑥) in Stratonovitch convention is (using what is known as ITO calculus):
¨ 𝛿𝐹 =
)9
)!
. 𝛿𝑥 +
-
.
.
)!9
)!! . 𝑏.. 𝛿𝑡 +
-
.
.
)9
)!
.
)5
)!
. 𝑏. 𝛿𝑡
¨ We know this is wrong, but let’s just illustrate how wrong it is:
¨ Rewriting it without the drift term we get:
¨ 𝛿𝑋 = 𝑏 𝑋 +
;*
.
, 𝑡 . 𝛿𝑊 = 𝑏 𝑋 . 𝛿𝑊 +
-
.
.
)5
)!
. 𝛿𝑋. 𝛿𝑊
¨ 𝛿𝑋 = 𝑏 𝑋 +
;*
.
, 𝑡 . 𝛿𝑊 = 𝑏 𝑋 . 𝛿𝑊 +
-
.
.
)5
)!
. 𝑏 𝑋 . 𝛿𝑡
¨ 𝛿𝑋 = 𝑏 𝑋 . 𝛿𝑊 +
-
.
.
)5
)!
. 𝑏 𝑋 . 𝛿𝑡 in Stratonovitch while Ito had: 𝛿𝑋 = 𝑏 𝑋 . 𝛿𝑊
112
Luc_Faucheux_2020
Another example: can you integrate X ? - VII
¨ Redoing it wrong just to convince ourselves of how wrong it is:
¨ 𝛿𝑋 = 𝑏 𝑋 ∘ 𝛿𝑊 = 𝑏 𝑋 +
;*
.
, 𝑡 . 𝛿𝑊 = 𝑏 𝑋 . 𝛿𝑊 +
-
.
.
)5
)!
. 𝑏 𝑋 . 𝛿𝑡
¨ 𝛿𝐹 =
)9
)*
. 𝛿𝑋 +
-
.
.
)!9
)*! . 𝑏.. 𝛿𝑡 +
-
.
.
)9
)*
.
)5
)*
. 𝑏. 𝛿𝑡
¨ 𝛿𝐹 =
)9
)*
. 𝛿𝑋 +
-
.
.
)!9
)*! . 𝛿𝑋.
¨ 𝛿𝐹 =
)9
)*
. 𝑏 𝑋 +
;*
.
. 𝛿𝑊 +
-
.
.
)!9
)*! . 𝑏 𝑋 +
;*
.
, 𝑡
.
. 𝛿𝑡
¨ 𝛿𝐹 =
)9
)*
. 𝑏 𝑋 . 𝛿𝑊 +
)9
)*
.
)5
)*
.
;*
.
. 𝛿𝑊 +
-
.
.
)!9
)*! . 𝑏 𝑋 +
;*
.
, 𝑡
.
. 𝛿𝑡
113
Luc_Faucheux_2020
Another example: can you integrate X ? - IX
¨ We get:
¨ 𝛿𝐹 =
)9
)*
. 𝑏 𝑋 . 𝛿𝑊 +
)9
)*
.
)5
)*
.
;*
.
. 𝛿𝑊 +
-
.
.
)!9
)*! . 𝑏 𝑋 +
;*
.
.
. 𝛿𝑡
¨ And :
¨ 𝑏 𝑋 +
;*
.
.
= 𝑏 𝑋 . + 2. 𝑏 𝑋 .
)5
)*
.
;*
.
¨ 𝛿𝐹 =
)9
)*
. 𝑏 𝑋 . 𝛿𝑊 +
)9
)*
.
)5
)*
.
;*
.
. 𝛿𝑊 +
-
.
.
)!9
)*! . 𝑏 𝑋 .. 𝛿𝑡 +
-
.
.
)!9
)*! . 2𝑏 𝑋 .
)5
)*
.
;*
.
𝛿𝑡
¨ Keeping only the terms in order 1 in 𝛿𝑍 and 𝛿𝑡
¨ 𝛿𝐹 =
)9
)*
. 𝑏 𝑋 . 𝛿𝑊 +
-
.
.
)!9
)*! . 𝑏 𝑋 .. 𝛿𝑡 +
-
.
.
)9
)*
.
)5
)*
. 𝑏. 𝛿𝑡
114
Luc_Faucheux_2020
Another example: can you integrate X ? - X
¨ 𝛿𝐹 =
)9
)*
. 𝑏 𝑋 . 𝛿𝑊 +
-
.
.
)!9
)*! . 𝑏 𝑋 .. 𝛿𝑡 +
-
.
.
)9
)*
.
)5
)*
. 𝑏. 𝛿𝑡
¨ In the case of the simple Geometric Brownian motion (Hull): 𝑑𝑋 = 𝜎𝑋𝑑𝑊
¨ Using the guess 𝐹 𝑋 = 𝑋.
¨ 𝛿(𝑋.) = 2𝑋. 𝜎𝑋. 𝛿𝑊 +
-
.
. 2. (𝜎𝑋). 𝛿𝑡𝛿𝑡 +
-
.
. 2𝑋. 𝜎. 𝜎𝑋. 𝛿𝑡
¨ 𝛿(𝑋.) = 2𝑋. 𝛿𝑋 + 2. (𝜎𝑋).. 𝛿𝑡
¨ In Ito we had : 𝛿 𝑋. = 2𝑋. 𝛿𝑋 +
-
.
. 2. (𝜎𝑋).. 𝛿𝑡
¨ So using ITO: ∫+34
+3=
𝑋. 𝑑𝑋 =
*(=)!
.
−
-
.
∫+34
+3=
(𝜎𝑋).. 𝛿𝑡
¨ Using Stratonovitch: ∫+34
+3=
𝑋. 𝑑𝑋 =
*(=)!
.
− ∫+34
+3=
(𝜎𝑋).. 𝛿𝑡
115
Luc_Faucheux_2020
Another example: can you integrate X ? - Xa
¨ If we were to look at the simple Brownian motion case:
¨ 𝑑𝑋 = 𝑑𝑊
¨ ITO: 𝛿 𝑋. = 2𝑋. 𝛿𝑋 +
-
.
. 2. 𝛿𝑋.
¨ 𝛿 𝑋. = 2𝑋. 𝛿𝑋 + 𝛿𝑡
¨ And so: ∫ 𝑋. 𝑑𝑋 = ∫{
; *!
.
−
-
.
. 𝛿𝑡}
¨ ∫ 𝑋. 𝑑𝑋 = [
*!
.
] −
-
.
∫ 𝛿𝑡
¨ With the usual convention that 𝑋 𝑡 = 0 = 0, in the ITO convention:
¨ ∫+34
+3=
𝑋. 𝑑𝑋 =
*(=)!
.
−
-
.
𝑇
116
Luc_Faucheux_2020
Another example: can you integrate X ? - Xb
¨ ∫+34
+3=
𝑋. 𝑑𝑋 =
*(=)!
.
−
-
.
𝑇
¨ Note that we know this to be consistent since the ITO integral is a martingale (or a trading
strategy)
¨ We will go over this again at more length but we have:
¨ 𝔼 ∫+34
+3=
𝑋. 𝑑𝑋 = 0 = 𝔼
* = !
.
− 𝔼
-
.
𝑇 = 𝔼
* = !
.
−
-
.
𝑇
¨ And we recover the usual dispersion expectation for the simple Brownian motion:
¨ 𝔼 𝑋 𝑇 . = 𝑇
117
Luc_Faucheux_2020
Another example: can you integrate X ? - Xc
¨ If we were to look at the simple Brownian motion case:
¨ 𝑑𝑋 = 𝑑𝑊 𝑏 𝑋 = 1 so
)5
)*
= 0
¨ STRATO: 𝛿𝐹 =
)9
)*
. 𝑏 𝑋 . 𝛿𝑊 +
-
.
.
)!9
)*! . 𝑏 𝑋 .. 𝛿𝑡 +
-
.
.
)9
)*
.
)5
)*
. 𝑏. 𝛿𝑡
¨ 𝛿 𝑋. = 2𝑋. 𝛿𝑋 + 𝛿𝑡 This is exactly the same as ITO
¨ So we get the strange result that in BOTH ITO and STRATO we will recover for the simple
Brownian motion:
¨ ∫+34
+3=
𝑋. 𝑑𝑋 =
*(=)!
.
−
-
.
𝑇
¨ This is clearly wrong, and once again the error was in “using a STRATO convention in the ITO
calculus”, and not even that, as we took “ITO calculus” as “doing regular Taylor expansions
like in regular calculus and just keeping some terms and neglecting some other terms”
118
Luc_Faucheux_2020
Another example: can you integrate X ? - XI
¨ HOWEVER, remember the formula in the simple Wiener case : 𝑋 = 𝑊
¨ 𝑆𝑇𝑅𝐴𝑇𝑂(𝑓) = 𝐼𝑇𝑂 𝑓 +
-
.
. 𝑅𝐼𝐸𝑀𝐴𝑁𝑁(𝑓%)
¨ In the case of 𝑓 = 𝑋, 𝑓% = 1
¨ 𝑆𝑇𝑅𝐴𝑇𝑂 𝑋 = 𝐼𝑇𝑂 𝑋 +
-
.
. 𝑅𝐼𝐸𝑀𝐴𝑁𝑁(1) with
¨ 𝑅𝐼𝐸𝑀𝐴𝑁𝑁 1 = ∫+34
+3=
1. 𝑑𝑡 = 𝑇
¨ So Ito : ∫+34
+3=
𝑋. 𝑑𝑋 =
*(=)!
.
−
-
.
∫+34
+3=
1. 𝛿𝑡 or ∫+34
+3=
𝑋. 𝑑𝑋 =
*(=)!
.
−
-
.
𝑇
¨ Stratonovitch: ∫+34
+3=
𝑋 ∘ 𝑑𝑋 = ∫+34
+3=
𝑋. 𝑑𝑋 +
-
.
𝑇 =
*(=)!
.
−
-
.
𝑇 +
-
.
𝑇 =
*(=)!
.
¨ Startonovitch as expected follows in a formal manner the usual rules of calculus
119
Luc_Faucheux_2020
Another example: can you integrate X ? - XII
¨ So clearly we failed to integrate X, because we found the following wrong results:
¨ In the Geometric Brownian motion case, Strato does not match formally the usual rules of
calculus
¨ In the simple Brownian motion case, BOTH ITO and STRATO returns the same result
¨ This is not possible as we know that the ITO and STRATO integrals are different and we know
what the difference is (
-
.
. 𝑅𝐼𝐸𝑀𝐴𝑁𝑁)
¨ But again, apologies if so far the mistake was obvious, but NEVER rely on usual rules of
calculus, or think that Stratanovitch is just “taking the middle point” and that you can still
use ITO lemma or Taylor expansion in a nested manner. If you choose STRATO you CANNOT
use ITO calculus, you have to stay in STRATO calculus, and vice versa
¨ Also note that ITO lemma is very generic in (𝛿𝑋). but when choosing a specific expression
for𝑋(𝑡) as a function of the Brownian motion (Wiener process 𝑊(𝑡)), we end up with very
different expressions, some tractable, some not so much
120
Luc_Faucheux_2020
Another example: can you integrate X ? - XIII
¨ The right way to do it is of course:
¨ ITO: 𝛿𝑓 =
)$
)*
. ([). 𝛿𝑋 +
-
.
.
)!9
)!! . ([). 𝛿𝑋.
¨ The results we obtained in the preceding slides for ITO are still correct
¨ STRATO: 𝛿𝑓 =
)$
)*
. ∘ . 𝛿𝑋
¨ And then we do get in the case of the simple Brownian motion: 𝑑𝑋 = 𝑑𝑊
¨ 𝐹 𝑋 = 𝑋.
¨ 𝛿𝑓 = 2𝑋. ∘ . 𝛿𝑊
¨ ∫+34
+3=
𝑋. ∘ . 𝑑𝑋 =
-
.
∫+34
+3=
2𝑋. ∘ . 𝛿𝑊 =
-
.
∫+34
+3=
𝑑 𝑋. = [
*!
.
]+34
+3=
¨ As expected if STRATO follows formally the usual rules of calculus
121
Luc_Faucheux_2020
Quick notes
¨ Usually when dealing with an SDE, people will try
¨ 1) get back to an SDE where the stochastic scaling does not depend on the stochastic
variable. For example, when dealing with 𝑑𝑆 = 𝜇. 𝑆. 𝑑𝑡 + 𝜎. 𝑆. 𝑑𝑊, the natural road to
explore is writing something like
I<
<
= 𝜇. 𝑑𝑡 + 𝜎. 𝑑𝑊, and then try to ”define” what
I<
<
is.
¨ It seems intuitive that
I<
<
should have something to do with 𝑑(ln 𝑆 ), so we would love to
write something like 𝑑(𝑙𝑛 𝑆 ) = 𝜇. 𝑑𝑡 + 𝜎. 𝑑𝑊, but it turns out that is not quite right,
because 𝑙𝑛 𝑆 is convex as a function of 𝑆, and 𝑆 is stochastic, not deterministic
¨ So always re-derive ITO lemma to make sure you are not dropping terms in the “Taylor
expansion”
¨ 2) try to get rid of the term in from of the 𝑑𝑧, because then you are not dealing with a SDE
anymore, but with a PDE. This is what Black Sholes did by adding a “delta hedge”, and off to
a Nobel prize they went
122
Luc_Faucheux_2020
Quick notes - II
¨ We also see the famous “convexity adjustment” that we looked at in the Options deck.
¨ 𝑙𝑛 𝑆 is negatively convex as a function of the stochastic variable
¨ So from an intuition point of view as we saw
¨ < 𝑙𝑛 𝑆 > ≠ 𝑙𝑛 < 𝑆 >
¨ Actually we know that : < 𝑙𝑛 𝑆 > = 𝔼(𝑙𝑛 𝑆 ) ≤ 𝑙𝑛 𝔼(𝑆) = 𝑙𝑛 < 𝑆 >
¨ So it would make sense that around the point 𝔼(𝑆) the function 𝑙𝑛 𝑆 would not behave as
the “regular” function 𝑙𝑛 𝔼(𝑆)
¨ Note that the distinction though, the convexity adjustment was coming from integrating
over the possible outcomes at a given time of the stochastic variable. Here we are
integrating over the time (over the path over time of the stochastic variable). They are
related nonetheless.
¨ Obviously if you have the expression that resulted from integrating over the path, you can
now use it to integrate over the distribution
123
Luc_Faucheux_2020
Quick notes - III
¨ Just to make it explicit.
¨ ∫+3+&
+3+'
𝑑𝐹(𝑋 𝑡 ) = ∫+3+6
+3+5 )9
)*
. ([). 𝑑𝑋(𝑡) +
-
.
∫+3+6
+3+5 )!9
)!! 𝑋 𝑡 . (𝑑𝑋).
¨ 𝛿𝐹 =
)9
)*
. 𝛿𝑋 +
-
.
.
)!9
)!! . (𝛿𝑋)., which is the celebrated Ito lemma (chain rule)
¨ 𝐹 𝑋 = ln(𝑋),
)9
)*
= 1/𝑋,
)!9
)*! = −1/𝑋.
¨ In the case of the simple Brownian motion: 𝑑𝑋 = 𝑑𝑊
¨ 𝛿(𝑙𝑛𝑊) =
-
1
. 𝛿𝑊 −
-
.
.
-
1! . 𝛿𝑡 this is quite ugly to deal with
¨ In the case of the Geometric Brownian motion (Hull) 𝑑𝑋 = 𝑋. 𝑑𝑊
¨ 𝛿 𝑙𝑛𝑋 =
-
*
. 𝑋. 𝛿𝑊 −
-
.
.
-
*! . 𝑋.. 𝛿𝑡 = 𝛿𝑊 −
-
.
. 𝛿𝑡 this is quite nice and easy to deal with
124
Luc_Faucheux_2020
Quick notes - IV
¨ So for the log function and in the case of a geometric Brownian motion, we have
¨ 𝛿 𝑙𝑛𝑋 = 𝛿𝑊 −
-
.
. 𝛿𝑡
¨ Or more rigorously:
¨ ∫+3+&
+3+'
𝑑𝐿𝑁(𝑋 𝑡 ) = 𝐿𝑛 𝑋 𝑡5 − 𝐿𝑛(𝑋 𝑡6 ) =
¨ 𝐿𝑛 𝑋 𝑡5 − 𝐿𝑛 𝑋 𝑡6 = ∫+3+6
+3+5
𝑑𝑊 𝑡 −
-
.
∫+3+6
+3+5
𝑑𝑡 = 𝑊 𝑡5 − 𝑊 𝑡6 −
-
.
(𝑡5 − 𝑡6)
¨ One could be tempted to identify
-
.
. 𝛿𝑡 as a convexity adjustment (which it is in some way,
since if the function was not convex, i.e.
)!9
)*! = 0, this term would not appear), but it is not
exactly the one we are dealing with in the option deck, as that one also depends on the
specific distribution being used.
125
Luc_Faucheux_2020
Quick notes - V
¨ Still rather crudely (but rigorous, as we will show in section V on the GBM)
¨ 𝑑𝑋 = 𝑋. 𝑑𝑊 careful, this is not the same as writing 𝑋 𝑡 = exp(𝑊 𝑡 )
¨ 𝔼 𝑋 𝑡5 = 𝑋 𝑡6
¨ 𝐿𝑛 𝑋 𝑡5 − 𝐿𝑛 𝑋 𝑡6 = ∫+3+6
+3+5
𝑑𝑊 𝑡 −
-
.
∫+3+6
+3+5
𝑑𝑡 = 𝑊 𝑡5 − 𝑊 𝑡6 −
-
.
(𝑡5 − 𝑡6)
¨ 𝔼 𝐿𝑛 𝑋 𝑡5 − 𝐿𝑛 𝑋 𝑡6 = 𝔼{𝑊 𝑡5 − 𝑊 𝑡6 −
-
.
(𝑡5 − 𝑡6)}
¨ 𝔼 𝐿𝑛 𝑋 𝑡5 − 𝐿𝑛 𝑋 𝑡6 = 𝔼 −
-
.
𝑡5 − 𝑡6 = −
-
.
(𝑡5 − 𝑡6)
¨ 𝔼 𝐿𝑛 𝑋 𝑡5 = 𝔼 𝐿𝑛 𝑋 𝑡6 −
-
.
𝑡5 − 𝑡6 = 𝐿𝑛 𝑋 𝑡6 −
-
.
(𝑡5 − 𝑡6)
¨ 𝔼 𝐿𝑛 𝑋 𝑡5 = 𝐿𝑛 𝔼 𝑋 𝑡5 −
-
.
(𝑡5 − 𝑡6)
¨ This is indeed the convexity adjustment (again makes sense, it is the second order)
126
Luc_Faucheux_2020
Quick notes - VI
¨ Just to make it explicit.
¨ ∫+3+&
+3+'
𝑑𝐹(𝑋 𝑡 ) = ∫+3+6
+3+5 )9
)*
. ([). 𝑑𝑋(𝑡) +
-
.
∫+3+6
+3+5 )!9
)!! 𝑋 𝑡 . (𝑑𝑋).
¨ 𝛿𝐹 =
)9
)*
. 𝛿𝑋 +
-
.
.
)!9
)!! . (𝛿𝑋)., which is the celebrated Ito lemma (chain rule)
¨ 𝐹 𝑋 = 𝑋.,
)9
)*
= 2𝑋,
)!9
)*! = 2
¨ In the case of the simple Brownian motion: 𝑑𝑋 = 𝑑𝑊
¨ 𝛿(𝑊.) = 2𝑊. 𝛿𝑊 +
-
.
. 2. 𝛿𝑡
¨ In the case of the Geometric Brownian motion (Hull) 𝑑𝑋 = 𝑋. 𝑑𝑊
¨ 𝛿 𝑋. = 2𝑋. 𝑋. 𝛿𝑊 +
-
.
. 2. 𝑋.. 𝛿𝑡 = 2 𝑋. 𝛿𝑊 + (𝑋.). 𝛿𝑡
127
Luc_Faucheux_2020
Quick notes - VII
¨ So for the square function and in the case of a simple Brownian motion, we have
¨ 𝛿(𝑊.) = 2𝑊. 𝛿𝑊 +
-
.
. 2. 𝛿𝑡
¨ Or more rigorously:
¨ ∫+3+&
+3+'
𝑑𝑊. = 𝑊.(𝑡5) − 𝑊.(𝑡6) =
¨ 𝑊. 𝑡5 − 𝑊. 𝑡6 = ∫+3+6
+3+5
2𝑊 𝑡 . [ . 𝑑𝑊 𝑡 + ∫+3+6
+3+5
𝑑𝑡
¨ 𝑊. 𝑡5 − 𝑊. 𝑡6 = ∫+3+6
+3+5
2𝑊 𝑡 . ([). 𝑑𝑊 𝑡 + (𝑡5 − 𝑡6)
128
Luc_Faucheux_2020
Quick notes - VIII
¨ Still rather crudely (but rigorous, as we will show in section V on the GBM)
¨ 𝑑𝑋 = 𝑑𝑊
¨ 𝔼 𝑊 𝑡5 = 𝑊 𝑡6
¨ 𝑊. 𝑡5 − 𝑊. 𝑡6 = ∫+3+6
+3+5
2𝑊 𝑡 . ([). 𝑑𝑊 𝑡 + 𝑡5 − 𝑡6
¨ 𝔼 𝑊. 𝑡5 − 𝑊. 𝑡6 = 𝔼{∫+3+6
+3+5
2𝑊 𝑡 . ([). 𝑑𝑊 𝑡 + 𝑡5 − 𝑡6 }
¨ 𝔼 𝑊. 𝑡5 − 𝑊. 𝑡6 = 𝔼 𝑡5 − 𝑡6 = (𝑡5 − 𝑡6)
¨ 𝔼 𝑊. 𝑡5 = 𝔼 𝑊. 𝑡6 + 𝑡5 − 𝑡6 = 𝑊. 𝑡6 + (𝑡5 − 𝑡6)
¨ 𝔼 𝑊. 𝑡5 = 𝔼 𝑊 𝑡5
. + (𝑡5 − 𝑡6)
¨ This is indeed the convexity adjustment (again makes sense, it is the second order)
¨ Since the square function is positively convex we would expect the convexity adjustment to
be positive
129
Luc_Faucheux_2020
130
From Ordinary Differential Equations
to SDE and SIE
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i
Lf 2020 stochastic_calculus_ito-i

More Related Content

Similar to Lf 2020 stochastic_calculus_ito-i

AliceVision : pipeline de reconstruction 3D open source
AliceVision : pipeline de reconstruction 3D open sourceAliceVision : pipeline de reconstruction 3D open source
AliceVision : pipeline de reconstruction 3D open sourceOpen Source Experience
 
Script For Perfect Presentation
Script For Perfect PresentationScript For Perfect Presentation
Script For Perfect PresentationAlan Doherty
 
#design The Brave New World: an owner’s manual. Chapter next: on mass-educa...
#design The Brave New World:  an owner’s manual. Chapter next:  on mass-educa...#design The Brave New World:  an owner’s manual. Chapter next:  on mass-educa...
#design The Brave New World: an owner’s manual. Chapter next: on mass-educa...Stefano Mirti
 
#design the brave new world: an owner’s manual (revised version)
#design the brave new world: an owner’s manual (revised version)#design the brave new world: an owner’s manual (revised version)
#design the brave new world: an owner’s manual (revised version)Whoami_edu
 
Charlie munger art of stock picking
Charlie munger   art of stock pickingCharlie munger   art of stock picking
Charlie munger art of stock pickingThugy Dee
 
ASSESSMENT CHECKLIST FOR DISCUSSION POSTS Content .docx
ASSESSMENT CHECKLIST FOR  DISCUSSION POSTS  Content .docxASSESSMENT CHECKLIST FOR  DISCUSSION POSTS  Content .docx
ASSESSMENT CHECKLIST FOR DISCUSSION POSTS Content .docxfredharris32
 
A Lesson on Elementary, Worldly Wisdom As It Relates To Invest.docx
A Lesson on Elementary, Worldly Wisdom As It Relates To Invest.docxA Lesson on Elementary, Worldly Wisdom As It Relates To Invest.docx
A Lesson on Elementary, Worldly Wisdom As It Relates To Invest.docxrock73
 
Everything you always wanted to know about psychology and technical communica...
Everything you always wanted to know about psychology and technical communica...Everything you always wanted to know about psychology and technical communica...
Everything you always wanted to know about psychology and technical communica...Chris Atherton @finiteattention
 
Night Essays.pdf
Night Essays.pdfNight Essays.pdf
Night Essays.pdfDamaris Tur
 
Ten Commandments for Good Teaching expla
Ten Commandments for Good Teaching explaTen Commandments for Good Teaching expla
Ten Commandments for Good Teaching explasouls2destroyer
 
A Lesson on Elementary, Worldly Wisdom As It Relates To Invest.docx
A Lesson on Elementary, Worldly Wisdom As It Relates To Invest.docxA Lesson on Elementary, Worldly Wisdom As It Relates To Invest.docx
A Lesson on Elementary, Worldly Wisdom As It Relates To Invest.docxransayo
 
Sawyer mathematicians delight
Sawyer mathematicians delightSawyer mathematicians delight
Sawyer mathematicians delightJuan Luis Cruz
 
Introduction About Yourself Essay Examples Sitedoct
Introduction About Yourself Essay Examples SitedoctIntroduction About Yourself Essay Examples Sitedoct
Introduction About Yourself Essay Examples SitedoctJeff Brooks
 

Similar to Lf 2020 stochastic_calculus_ito-i (20)

AliceVision : pipeline de reconstruction 3D open source
AliceVision : pipeline de reconstruction 3D open sourceAliceVision : pipeline de reconstruction 3D open source
AliceVision : pipeline de reconstruction 3D open source
 
Script For Perfect Presentation
Script For Perfect PresentationScript For Perfect Presentation
Script For Perfect Presentation
 
Slides
SlidesSlides
Slides
 
Teaching that Sticks
Teaching that SticksTeaching that Sticks
Teaching that Sticks
 
#design The Brave New World: an owner’s manual. Chapter next: on mass-educa...
#design The Brave New World:  an owner’s manual. Chapter next:  on mass-educa...#design The Brave New World:  an owner’s manual. Chapter next:  on mass-educa...
#design The Brave New World: an owner’s manual. Chapter next: on mass-educa...
 
#design the brave new world: an owner’s manual (revised version)
#design the brave new world: an owner’s manual (revised version)#design the brave new world: an owner’s manual (revised version)
#design the brave new world: an owner’s manual (revised version)
 
Charlie munger art of stock picking
Charlie munger   art of stock pickingCharlie munger   art of stock picking
Charlie munger art of stock picking
 
Illustration And Example Essay
Illustration And Example EssayIllustration And Example Essay
Illustration And Example Essay
 
ASSESSMENT CHECKLIST FOR DISCUSSION POSTS Content .docx
ASSESSMENT CHECKLIST FOR  DISCUSSION POSTS  Content .docxASSESSMENT CHECKLIST FOR  DISCUSSION POSTS  Content .docx
ASSESSMENT CHECKLIST FOR DISCUSSION POSTS Content .docx
 
A Lesson on Elementary, Worldly Wisdom As It Relates To Invest.docx
A Lesson on Elementary, Worldly Wisdom As It Relates To Invest.docxA Lesson on Elementary, Worldly Wisdom As It Relates To Invest.docx
A Lesson on Elementary, Worldly Wisdom As It Relates To Invest.docx
 
Educator guide
Educator guideEducator guide
Educator guide
 
Educator guide
Educator guideEducator guide
Educator guide
 
Educator Guide
Educator GuideEducator Guide
Educator Guide
 
Educator Guide
Educator GuideEducator Guide
Educator Guide
 
Everything you always wanted to know about psychology and technical communica...
Everything you always wanted to know about psychology and technical communica...Everything you always wanted to know about psychology and technical communica...
Everything you always wanted to know about psychology and technical communica...
 
Night Essays.pdf
Night Essays.pdfNight Essays.pdf
Night Essays.pdf
 
Ten Commandments for Good Teaching expla
Ten Commandments for Good Teaching explaTen Commandments for Good Teaching expla
Ten Commandments for Good Teaching expla
 
A Lesson on Elementary, Worldly Wisdom As It Relates To Invest.docx
A Lesson on Elementary, Worldly Wisdom As It Relates To Invest.docxA Lesson on Elementary, Worldly Wisdom As It Relates To Invest.docx
A Lesson on Elementary, Worldly Wisdom As It Relates To Invest.docx
 
Sawyer mathematicians delight
Sawyer mathematicians delightSawyer mathematicians delight
Sawyer mathematicians delight
 
Introduction About Yourself Essay Examples Sitedoct
Introduction About Yourself Essay Examples SitedoctIntroduction About Yourself Essay Examples Sitedoct
Introduction About Yourself Essay Examples Sitedoct
 

More from luc faucheux

Marcel_Faucheux_Taquin.pdf
Marcel_Faucheux_Taquin.pdfMarcel_Faucheux_Taquin.pdf
Marcel_Faucheux_Taquin.pdfluc faucheux
 
Selection_Brownian_Particles.pdf
Selection_Brownian_Particles.pdfSelection_Brownian_Particles.pdf
Selection_Brownian_Particles.pdfluc faucheux
 
Binary_Potential.pdf
Binary_Potential.pdfBinary_Potential.pdf
Binary_Potential.pdfluc faucheux
 
Periodic_Forcing.pdf
Periodic_Forcing.pdfPeriodic_Forcing.pdf
Periodic_Forcing.pdfluc faucheux
 
ConfinedBrownian MotionPhysRevE.49.5158.pdf
ConfinedBrownian MotionPhysRevE.49.5158.pdfConfinedBrownian MotionPhysRevE.49.5158.pdf
ConfinedBrownian MotionPhysRevE.49.5158.pdfluc faucheux
 
Optical_Thermal_Ratchet.pdf
Optical_Thermal_Ratchet.pdfOptical_Thermal_Ratchet.pdf
Optical_Thermal_Ratchet.pdfluc faucheux
 
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
 
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iiiLf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iiiluc faucheux
 
Lf 2021 rates_viii_a
Lf 2021 rates_viii_aLf 2021 rates_viii_a
Lf 2021 rates_viii_aluc faucheux
 
Lf 2021 rates_iv_a_irma
Lf 2021 rates_iv_a_irmaLf 2021 rates_iv_a_irma
Lf 2021 rates_iv_a_irmaluc faucheux
 
Lf 2021 rates_v_b2
Lf 2021 rates_v_b2Lf 2021 rates_v_b2
Lf 2021 rates_v_b2luc faucheux
 
Lf 2020 stochastic_calculus_ito-ii
Lf 2020 stochastic_calculus_ito-iiLf 2020 stochastic_calculus_ito-ii
Lf 2020 stochastic_calculus_ito-iiluc faucheux
 
Lf 2020 structured
Lf 2020 structuredLf 2020 structured
Lf 2020 structuredluc faucheux
 

More from luc faucheux (20)

Marcel_Faucheux_Taquin.pdf
Marcel_Faucheux_Taquin.pdfMarcel_Faucheux_Taquin.pdf
Marcel_Faucheux_Taquin.pdf
 
Selection_Brownian_Particles.pdf
Selection_Brownian_Particles.pdfSelection_Brownian_Particles.pdf
Selection_Brownian_Particles.pdf
 
Binary_Potential.pdf
Binary_Potential.pdfBinary_Potential.pdf
Binary_Potential.pdf
 
Periodic_Forcing.pdf
Periodic_Forcing.pdfPeriodic_Forcing.pdf
Periodic_Forcing.pdf
 
ConfinedBrownian MotionPhysRevE.49.5158.pdf
ConfinedBrownian MotionPhysRevE.49.5158.pdfConfinedBrownian MotionPhysRevE.49.5158.pdf
ConfinedBrownian MotionPhysRevE.49.5158.pdf
 
Optical_Thermal_Ratchet.pdf
Optical_Thermal_Ratchet.pdfOptical_Thermal_Ratchet.pdf
Optical_Thermal_Ratchet.pdf
 
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
 
Lf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iiiLf 2021 stochastic_calculus_ito-iii
Lf 2021 stochastic_calculus_ito-iii
 
Lf 2021 rates_viii_a
Lf 2021 rates_viii_aLf 2021 rates_viii_a
Lf 2021 rates_viii_a
 
Lf 2021 rates_iv_a_irma
Lf 2021 rates_iv_a_irmaLf 2021 rates_iv_a_irma
Lf 2021 rates_iv_a_irma
 
Lf 2021 rates_vii
Lf 2021 rates_viiLf 2021 rates_vii
Lf 2021 rates_vii
 
Lf 2021 rates_vi
Lf 2021 rates_viLf 2021 rates_vi
Lf 2021 rates_vi
 
Lf 2021 rates_v_b2
Lf 2021 rates_v_b2Lf 2021 rates_v_b2
Lf 2021 rates_v_b2
 
Lf 2021 rates_v_a
Lf 2021 rates_v_aLf 2021 rates_v_a
Lf 2021 rates_v_a
 
Lf 2020 rates_v_a
Lf 2020 rates_v_aLf 2020 rates_v_a
Lf 2020 rates_v_a
 
Lf 2020 rates_iv
Lf 2020 rates_ivLf 2020 rates_iv
Lf 2020 rates_iv
 
Lf 2020 rates_ii
Lf 2020 rates_iiLf 2020 rates_ii
Lf 2020 rates_ii
 
Lf 2020 stochastic_calculus_ito-ii
Lf 2020 stochastic_calculus_ito-iiLf 2020 stochastic_calculus_ito-ii
Lf 2020 stochastic_calculus_ito-ii
 
Lf 2020 structured
Lf 2020 structuredLf 2020 structured
Lf 2020 structured
 
Lf 2020 trees
Lf 2020 treesLf 2020 trees
Lf 2020 trees
 

Recently uploaded

Instant Issue Debit Cards - High School Spirit
Instant Issue Debit Cards - High School SpiritInstant Issue Debit Cards - High School Spirit
Instant Issue Debit Cards - High School Spiritegoetzinger
 
Interimreport1 January–31 March2024 Elo Mutual Pension Insurance Company
Interimreport1 January–31 March2024 Elo Mutual Pension Insurance CompanyInterimreport1 January–31 March2024 Elo Mutual Pension Insurance Company
Interimreport1 January–31 March2024 Elo Mutual Pension Insurance CompanyTyöeläkeyhtiö Elo
 
Mulki Call Girls 7001305949 WhatsApp Number 24x7 Best Services
Mulki Call Girls 7001305949 WhatsApp Number 24x7 Best ServicesMulki Call Girls 7001305949 WhatsApp Number 24x7 Best Services
Mulki Call Girls 7001305949 WhatsApp Number 24x7 Best Servicesnajka9823
 
fca-bsps-decision-letter-redacted (1).pdf
fca-bsps-decision-letter-redacted (1).pdffca-bsps-decision-letter-redacted (1).pdf
fca-bsps-decision-letter-redacted (1).pdfHenry Tapper
 
Vip B Aizawl Call Girls #9907093804 Contact Number Escorts Service Aizawl
Vip B Aizawl Call Girls #9907093804 Contact Number Escorts Service AizawlVip B Aizawl Call Girls #9907093804 Contact Number Escorts Service Aizawl
Vip B Aizawl Call Girls #9907093804 Contact Number Escorts Service Aizawlmakika9823
 
VIP High Class Call Girls Saharanpur Anushka 8250192130 Independent Escort Se...
VIP High Class Call Girls Saharanpur Anushka 8250192130 Independent Escort Se...VIP High Class Call Girls Saharanpur Anushka 8250192130 Independent Escort Se...
VIP High Class Call Girls Saharanpur Anushka 8250192130 Independent Escort Se...Suhani Kapoor
 
Call Girls In Yusuf Sarai Women Seeking Men 9654467111
Call Girls In Yusuf Sarai Women Seeking Men 9654467111Call Girls In Yusuf Sarai Women Seeking Men 9654467111
Call Girls In Yusuf Sarai Women Seeking Men 9654467111Sapana Sha
 
Classical Theory of Macroeconomics by Adam Smith
Classical Theory of Macroeconomics by Adam SmithClassical Theory of Macroeconomics by Adam Smith
Classical Theory of Macroeconomics by Adam SmithAdamYassin2
 
Independent Lucknow Call Girls 8923113531WhatsApp Lucknow Call Girls make you...
Independent Lucknow Call Girls 8923113531WhatsApp Lucknow Call Girls make you...Independent Lucknow Call Girls 8923113531WhatsApp Lucknow Call Girls make you...
Independent Lucknow Call Girls 8923113531WhatsApp Lucknow Call Girls make you...makika9823
 
Attachment Of Assets......................
Attachment Of Assets......................Attachment Of Assets......................
Attachment Of Assets......................AmanBajaj36
 
Andheri Call Girls In 9825968104 Mumbai Hot Models
Andheri Call Girls In 9825968104 Mumbai Hot ModelsAndheri Call Girls In 9825968104 Mumbai Hot Models
Andheri Call Girls In 9825968104 Mumbai Hot Modelshematsharma006
 
OAT_RI_Ep19 WeighingTheRisks_Apr24_TheYellowMetal.pptx
OAT_RI_Ep19 WeighingTheRisks_Apr24_TheYellowMetal.pptxOAT_RI_Ep19 WeighingTheRisks_Apr24_TheYellowMetal.pptx
OAT_RI_Ep19 WeighingTheRisks_Apr24_TheYellowMetal.pptxhiddenlevers
 
Q3 2024 Earnings Conference Call and Webcast Slides
Q3 2024 Earnings Conference Call and Webcast SlidesQ3 2024 Earnings Conference Call and Webcast Slides
Q3 2024 Earnings Conference Call and Webcast SlidesMarketing847413
 
原版1:1复刻堪萨斯大学毕业证KU毕业证留信学历认证
原版1:1复刻堪萨斯大学毕业证KU毕业证留信学历认证原版1:1复刻堪萨斯大学毕业证KU毕业证留信学历认证
原版1:1复刻堪萨斯大学毕业证KU毕业证留信学历认证jdkhjh
 
The Triple Threat | Article on Global Resession | Harsh Kumar
The Triple Threat | Article on Global Resession | Harsh KumarThe Triple Threat | Article on Global Resession | Harsh Kumar
The Triple Threat | Article on Global Resession | Harsh KumarHarsh Kumar
 
Bladex Earnings Call Presentation 1Q2024
Bladex Earnings Call Presentation 1Q2024Bladex Earnings Call Presentation 1Q2024
Bladex Earnings Call Presentation 1Q2024Bladex
 
原版1:1复刻温哥华岛大学毕业证Vancouver毕业证留信学历认证
原版1:1复刻温哥华岛大学毕业证Vancouver毕业证留信学历认证原版1:1复刻温哥华岛大学毕业证Vancouver毕业证留信学历认证
原版1:1复刻温哥华岛大学毕业证Vancouver毕业证留信学历认证rjrjkk
 
government_intervention_in_business_ownership[1].pdf
government_intervention_in_business_ownership[1].pdfgovernment_intervention_in_business_ownership[1].pdf
government_intervention_in_business_ownership[1].pdfshaunmashale756
 

Recently uploaded (20)

Instant Issue Debit Cards - High School Spirit
Instant Issue Debit Cards - High School SpiritInstant Issue Debit Cards - High School Spirit
Instant Issue Debit Cards - High School Spirit
 
Commercial Bank Economic Capsule - April 2024
Commercial Bank Economic Capsule - April 2024Commercial Bank Economic Capsule - April 2024
Commercial Bank Economic Capsule - April 2024
 
Interimreport1 January–31 March2024 Elo Mutual Pension Insurance Company
Interimreport1 January–31 March2024 Elo Mutual Pension Insurance CompanyInterimreport1 January–31 March2024 Elo Mutual Pension Insurance Company
Interimreport1 January–31 March2024 Elo Mutual Pension Insurance Company
 
Mulki Call Girls 7001305949 WhatsApp Number 24x7 Best Services
Mulki Call Girls 7001305949 WhatsApp Number 24x7 Best ServicesMulki Call Girls 7001305949 WhatsApp Number 24x7 Best Services
Mulki Call Girls 7001305949 WhatsApp Number 24x7 Best Services
 
fca-bsps-decision-letter-redacted (1).pdf
fca-bsps-decision-letter-redacted (1).pdffca-bsps-decision-letter-redacted (1).pdf
fca-bsps-decision-letter-redacted (1).pdf
 
Vip B Aizawl Call Girls #9907093804 Contact Number Escorts Service Aizawl
Vip B Aizawl Call Girls #9907093804 Contact Number Escorts Service AizawlVip B Aizawl Call Girls #9907093804 Contact Number Escorts Service Aizawl
Vip B Aizawl Call Girls #9907093804 Contact Number Escorts Service Aizawl
 
Monthly Economic Monitoring of Ukraine No 231, April 2024
Monthly Economic Monitoring of Ukraine No 231, April 2024Monthly Economic Monitoring of Ukraine No 231, April 2024
Monthly Economic Monitoring of Ukraine No 231, April 2024
 
VIP High Class Call Girls Saharanpur Anushka 8250192130 Independent Escort Se...
VIP High Class Call Girls Saharanpur Anushka 8250192130 Independent Escort Se...VIP High Class Call Girls Saharanpur Anushka 8250192130 Independent Escort Se...
VIP High Class Call Girls Saharanpur Anushka 8250192130 Independent Escort Se...
 
Call Girls In Yusuf Sarai Women Seeking Men 9654467111
Call Girls In Yusuf Sarai Women Seeking Men 9654467111Call Girls In Yusuf Sarai Women Seeking Men 9654467111
Call Girls In Yusuf Sarai Women Seeking Men 9654467111
 
Classical Theory of Macroeconomics by Adam Smith
Classical Theory of Macroeconomics by Adam SmithClassical Theory of Macroeconomics by Adam Smith
Classical Theory of Macroeconomics by Adam Smith
 
Independent Lucknow Call Girls 8923113531WhatsApp Lucknow Call Girls make you...
Independent Lucknow Call Girls 8923113531WhatsApp Lucknow Call Girls make you...Independent Lucknow Call Girls 8923113531WhatsApp Lucknow Call Girls make you...
Independent Lucknow Call Girls 8923113531WhatsApp Lucknow Call Girls make you...
 
Attachment Of Assets......................
Attachment Of Assets......................Attachment Of Assets......................
Attachment Of Assets......................
 
Andheri Call Girls In 9825968104 Mumbai Hot Models
Andheri Call Girls In 9825968104 Mumbai Hot ModelsAndheri Call Girls In 9825968104 Mumbai Hot Models
Andheri Call Girls In 9825968104 Mumbai Hot Models
 
OAT_RI_Ep19 WeighingTheRisks_Apr24_TheYellowMetal.pptx
OAT_RI_Ep19 WeighingTheRisks_Apr24_TheYellowMetal.pptxOAT_RI_Ep19 WeighingTheRisks_Apr24_TheYellowMetal.pptx
OAT_RI_Ep19 WeighingTheRisks_Apr24_TheYellowMetal.pptx
 
Q3 2024 Earnings Conference Call and Webcast Slides
Q3 2024 Earnings Conference Call and Webcast SlidesQ3 2024 Earnings Conference Call and Webcast Slides
Q3 2024 Earnings Conference Call and Webcast Slides
 
原版1:1复刻堪萨斯大学毕业证KU毕业证留信学历认证
原版1:1复刻堪萨斯大学毕业证KU毕业证留信学历认证原版1:1复刻堪萨斯大学毕业证KU毕业证留信学历认证
原版1:1复刻堪萨斯大学毕业证KU毕业证留信学历认证
 
The Triple Threat | Article on Global Resession | Harsh Kumar
The Triple Threat | Article on Global Resession | Harsh KumarThe Triple Threat | Article on Global Resession | Harsh Kumar
The Triple Threat | Article on Global Resession | Harsh Kumar
 
Bladex Earnings Call Presentation 1Q2024
Bladex Earnings Call Presentation 1Q2024Bladex Earnings Call Presentation 1Q2024
Bladex Earnings Call Presentation 1Q2024
 
原版1:1复刻温哥华岛大学毕业证Vancouver毕业证留信学历认证
原版1:1复刻温哥华岛大学毕业证Vancouver毕业证留信学历认证原版1:1复刻温哥华岛大学毕业证Vancouver毕业证留信学历认证
原版1:1复刻温哥华岛大学毕业证Vancouver毕业证留信学历认证
 
government_intervention_in_business_ownership[1].pdf
government_intervention_in_business_ownership[1].pdfgovernment_intervention_in_business_ownership[1].pdf
government_intervention_in_business_ownership[1].pdf
 

Lf 2020 stochastic_calculus_ito-i

  • 1. Luc_Faucheux_2020 Stochastic Calculus - Part I Integrals - Ito lemma – SDEs and SIEs 1
  • 2. Luc_Faucheux_2020 What is this class, and what it is not ¨ Not a formal, so please interrupt if any question ¨ More of a pragmatic approach on how to approach stochastic calculus ¨ I have tried as much as possible to alert when there is something to be careful about ¨ I have also tried as much as possible to be as rigorous as possible, without getting lost in the notations, or being too formal just for the sake of being formal ¨ Those notes are more “notes of a practitioner”, and by no means I would dare to hope to substitute a robust course in stochastic calculus ¨ Those slides originated from a class I taught in 2018 at the hedge fund DRW to their first year associate class, 40 students or so from various backgrounds. The class was a general Fixed-Income class over 8 full days. It was intense, exhilarating, and kept me on my toes the whole time ¨ Starting from the usual Ito lemma as in most textbooks (Hull), play with it for a while, then start again from the proper formulation of the stochastic integrals 2
  • 3. Luc_Faucheux_2020 How those slides came about ¨ Those slides originated from a class I taught in 2018 at the hedge fund DRW in the great city of Chicago. I have modified them and added to them over the past couple of years. ¨ The class was composed of 40 students or so with various backgrounds, ranging from computer science students with almost no background in Finance, to recent graduates of Masters in Math, to some graduates of the prestigious MSCF (Master of Science in Computational Finance) at Carnegie Melon University under the guidance of Steven Shreve, to some who had almost no mathematical background in options, bond math, random processes but had advanced degrees in Economy, so it was rather a tricky bunch ¨ The class was a general Fixed-Income class over 8 full days. It was intense, exhilarating, and kept me on my toes the whole time. It pushed me to realize what I had not understood about stochastic processes for 20 years or so, because I never bothered to ask the “what if” questions and took a lot for granted out of sheer intellectual laziness ¨ The textbook we used was Hull so you will see pages reference to this book, as we used it in class as a starting point to further explorations of the derivative pricing theory. 3
  • 4. Luc_Faucheux_2020 How those slides came about - II ¨ It is often said that the only person who learns anything out of a class is the teacher. ¨ I hope that my DRW students got something out of it. ¨ But I know for a fact that without those two weeks in Chicago teaching, I would not have gotten those slides off the ground, and most of the materials would still be in disparate pieces of papers flying around my desk. ¨ Hopefully the end result is not complete garbage, at least I know that I greatly enjoyed putting those together. ¨ So, even if the end result is not up to your standards, I am extremely grateful to DRW for having given me the opportunity to teach the associate class that summer of 2018, and even more so grateful to the students who during those two weeks, took me out of my comfort zone, and forced me to confront what I knew and what I realized I was rather ignorant of. ¨ When you are teaching in front of a bunch of super smart people with different backgrounds for 8 days straight, there is no hiding behind the curtain ¨ So thanks again to DRW and the 2018 associate class ! I miss you guys. 4
  • 5. Luc_Faucheux_2020 How those slides came about - III ¨ Also for those of you who have spent more than 5 minutes with me, you will have noticed that I bring the Ito-Stratonovitch controversy a lot. ¨ First of all that is sort of a hobby of mine, ever since my PhD thesis (Appendix B) ¨ Second, I have found it to be quite enlightening, because everyone takes Ito for granted, and then you ask the question “what if”, and that forces you to make sure that your understanding of Ito was solid. So I am using the example of Stratonovitch to really test the fact that my understanding of ITO is solid ¨ Apologies on the Powerpoint format, there are battles with Microsoft that you cannot win (including the fact that the Solver is a Macro and not a function, as opposed to some of the earlier spreadsheets like Wingz) ¨ At times it might feel like going down the rabbit hole, or going up the river to meet Kurtz, but (at least in my experience), I have found those discursions to be useful. ¨ So this is not really something great about Stratonovitch, it is using Stratonovitch to convince ourselves that we understand ITO 5
  • 6. Luc_Faucheux_2020 How those slides came about - IV ¨ Again quite frankly those notes are more for me (sorry) ¨ I had noticed that I had a number of handwritten notes flying around my desk, and when needed at times I would just rederive from scratch rather then finding the right one. ¨ So this is trying to put in one place, in a format that I can copy/past easily, most of what I had to go through and still use. I have tried (and failed at many places) to keep the notation consistent ¨ Again, this is from a “practitioner” point of view, I have tried to be rigorous when I felt it was needed, and be less so when I felt that this was a detail that was not needed and at times would obscure the intuition. ¨ Notations are hard to keep consistent. I have also find at times that the right notation can be illuminating, whereas the full explicit one can be cumbersome. So I have tried to adapt the notation to what was needed to grasp the concept, and at times be more careful about it to make sure that we do not fall into a trap. ¨ I guess Godel knew that, the right notation can change completely the proof…. 6
  • 7. Luc_Faucheux_2020 If you really want to master stochastic calculus. ¨ The uncontested bible in the field of stochastic calculus for Finance. Quite dense and concise. It sometimes take me 40 pages to understand what Steven Shreve does in 2 lines. 7
  • 8. Luc_Faucheux_2020 If you really want to master stochastic calculus - II ¨ An absolute wonderful short book. The notations at times are infuriating, but an absolute must read 8
  • 9. Luc_Faucheux_2020 If you really want to master stochastic calculus - III ¨ If you are like me coming from a Physics background, this book is still relevant today, a testament to the genius of Van Kampen 9
  • 10. Luc_Faucheux_2020 If you really want to master stochastic calculus - IV ¨ You cannot ignore this book. Every sentence carries meaning, and is worth reading time and time again. 10
  • 11. Luc_Faucheux_2020 If you really want to master stochastic calculus - V ¨ Another wonderful short book. The logic is clear, concise and beautiful. The exercises are worth going through. From one of the most respected options traders in the field. He also has quite a ferocious appetite for chocolate cakes. 11
  • 12. Luc_Faucheux_2020 If you really want to master stochastic calculus - VI ¨ Quite applied. The appendix on SDEs is rather beautiful, and follows Mikosch. Great for practical applications in the field of Fixed-Income 12
  • 13. Luc_Faucheux_2020 If you really want to master stochastic calculus - VII ¨ What not to say about this book? An absolute gem. The 1900 Ph.D. thesis of Louis Bachelier (in French!) with an amazing translation by Davis and Etheridge, and some great chapters about the history of modern finance. Bachelier did it all, 80 years or so before everyone else 13
  • 14. Luc_Faucheux_2020 If you really want to master stochastic calculus - VIII ¨ I could not but not add this one here. It is a movie about the life of Vincent (Wolfgang) Doblin (Doebling) who essentially discovered Ito calculus at least 5 to 10 years before Ito in circumstances so incredible that they made a movie out of his life. Ito lemma and calculus is now usually referred to as Ito-Doblin lemma and calculus in recognition of Vincent’s incredible accomplishments 14
  • 15. Luc_Faucheux_2020 If you really want to master stochastic calculus - IX ¨ Amazing book if you are coming from a Physics background 15
  • 16. Luc_Faucheux_2020 If you really want to master stochastic calculus - X ¨ Found this book as I was almost finished with those slides, and thought about throwing them away, because this book has pretty much anything you want. Pretty heavy on the operators formalism, which is quite elegant once you get used to it, but that it a step that you need to go through 16
  • 17. Luc_Faucheux_2020 What is so hard about Stochastic Calculus? ¨ It is quite recent ¨ It is quite cumbersome ¨ It is not intuitive ¨ It is incomplete ¨ No one really knows how to do it. ¨ This presentation is trying to strike a balance between being practical and being rigorous, so apologies for the many terms in “”, whereas a rigorous math class will define what terms like “stable” or “got to 0” or “go to infinity” or “converge” much more exactly. What we will say is “almost” rigorous and works in practice 99% of the time ¨ This is more of a practitioner’s point of view on how to use (or not) stochastic calculus, as it relates to finance and Physics 17
  • 18. Luc_Faucheux_2020 Stochastic Calculus is quite recent ¨ Geometry ~ -6,000 BC (navigating looking at the stars, buildings, ships,…) ¨ Fractions (music with scale, buildings,,) ¨ Probabilities (~1,600 AD), Pascal triangle, combinatory analysis ¨ Calculus (Newton, Leibniz, finite differences) (~1,600 AD) (we started shooting canons long range, Electricity, magnets, how do they work? The ICP is still asking) ¨ Taylor Expansion (1720) ¨ Brown (1890), Wiener (1940), Bachelier (1900), Einstein (1905),Langevin (1908), Doblin(1940), Ito (1950), Feynman-Kac (1950), Stratonovitch (1966), Black-Sholes (1972) ¨ So let’s go a little easy on ourselves, shall we? ¨ Quantum Stochastic Calculus (1980), random processes (diffusion) in fractal geometries 18
  • 19. Luc_Faucheux_2020 Stochastic calculus is a French thing (except Gauss) ¨ Black-Sholes might have gotten a Nobel prize in 1972, but Louis Bachelier did it all in his Ph.D. thesis in 1900 (almost) ¨ Ito might have been known until recently for the Ito calculus and Ito lemma, but Vincent Doelin wrote it all while on the Ardennes front in World War I. This is now being recognized and some textbooks use the term “Ito-Doeblin” instead of “Ito” ¨ Paul Langevin also essentially wrote the textbook on SDEs ¨ And for the math, all you need is Laplace, Fourier, Cauchy ¨ Taylor expansion is the only non-French, but it is essentially L’Hospital rule, so again…French… ¨ Note: L’Hopital rule is very powerful and often overlooked. ¨ If two functions 𝑓(𝑥) and 𝑔(𝑥) and are differentiable, and lim !→# [ $%(!) (%(!) ] exists, then ¨ lim !→# [ $(!) ((!) ] = lim !→# [ $%(!) (%(!) ] 19
  • 20. Luc_Faucheux_2020 The Ph.D. thesis of Louis Bachelier ¨ Reading the original thesis (both in French if you can and the excellent translation by Mark Davis and Alison Etheridge) is humbling. ¨ Without a strong well-developed theory of stochastic calculus (Ito lemma) that only came about in the 1960s or so ¨ Without a strong theoretical footing of what is a numeraire and how to price a derivative in the risk-neutral probability associated to that numeraire (Pliska 1980 or so) ¨ Without yet the strong connection between PDE (Partial Differential Equations) and SDE (Stochastic Differential Equations) that really came about from the Feynman-Kac formula (1950 roughly) ¨ Louis Bachelier managed to not only built a theory of option pricing that is nowadays coming back in fashion with a vengeance, but perusing through the rather short thesis, one cannot but be amazed at the breadth of his genius, but also at his attention to details. Bachelier at times go through numerical examples with the same precision and clarity of thoughts that he displays in the other more theoretical parts of his thesis. 20
  • 21. Luc_Faucheux_2020 Stochastic Calculus is not intuitive ¨ Regular calculus has usually to do with ”things that are around some other things” ¨ Taylor expansion and derivatives (expansion around a value, local derivatives at or around a point) ¨ Finite difference for integrating functions on an axis or a path (keep following the path in a continuous fashion) ¨ Stochastic Calculus tries to address the problem of dealing with “things around things that are not there”. What do I mean ? ¨ Take the coin flipping problem (Head is +1, Tail is -1). The coin is either head or tail (+1 or - 1), never anything else. And yet we will try to calculate expansions, derivations, integrations of functions around the mean or average (0), which is NOT a possible state of the coin. ¨ Without stating the obvious or oversimplifying, this is the crux of the problem with stochastic calculus, and also that we are so used to usual calculus that we take a lot of things for granted 21
  • 22. Luc_Faucheux_2020 Stochastic processes are “non-differentiable” ¨ A stochastic process essentially “flips a coin” at each point in time. ¨ A “regular” process, meaning it is differentiable, would have a unique tangent for every point in time. If 𝑋 𝑡 is differentiable, there is a unique )*(+) )+ ¨ The stochastic process does NOT have a unique tangent, because, using the coin flip idea, and being somewhat liberal with scaling, at each point in time, 𝑋 𝑡 goes to either {𝑋 𝑡 + 𝛿𝑋} or {𝑋 𝑡 − 𝛿𝑋} with some probability (50% in the simplest case, or driftless case). ¨ The tangent is NOT the average (0) of the two possible tangents. ¨ So in essence for a stochastic process, writing something like )*(+) )+ is meaningless ¨ In stochastic processes, the only real thing that we can do is integrals, almost never differential calculus (not surprising as it is not differentiable) ¨ NOTE that non-differentiable does NOT mean not smooth or not continuous 22
  • 23. Luc_Faucheux_2020 Stochastic calculus is usually self-similar ¨ We will go over this in more details, but essentially the simplest stochastic process is the Wiener process or also called the standard Brownian motion. We will start with it ¨ 𝑊 𝑡 has a normal 𝑁(0, 𝑡) distribution function ¨ 𝑊 𝑡 − 𝑠 has obviously the same normal 𝑁(0, 𝑡 − 𝑠) distribution function ¨ {𝑊 𝑡 − 𝑊(𝑠)} has ALSO the same 𝑁(0, 𝑡 − 𝑠) distribution ¨ Note that they are NOT the same, writing 𝑊 𝑡 − 𝑊 𝑠 = 𝑊(𝑡 − 𝑠) is obviously wrong but you will find sometimes the notation: ¨ 𝑊 𝑡 − 𝑊 𝑠 ≝ 𝑊(𝑡 − 𝑠), meaning distributional identity, NOT pathwise identity ¨ The simple Brownian motion is self-similar with coefficient (1/2) ¨ 𝑇, 𝑊 𝑡- , 𝑇, 𝑊 𝑡. , . . , 𝑇, 𝑊 𝑡/ ≝ 𝑊 𝑇𝑡- , 𝑊 𝑇𝑡. , . . , 𝑊 𝑇𝑡/ , with 𝐻 = 1/2 ¨ This is sometimes used to numerically construct Brownian motion. ¨ If you “scale” up the time scale by a factor 𝑇, the space scale gets only magnified by a factor 𝑇. 23
  • 24. Luc_Faucheux_2020 Stochastic Calculus is cumbersome ¨ Usual knowledge and tricks of calculus do not apply anymore ¨ Chain rule does not work ¨ Functional derivation does not work : 𝑑 𝑙𝑛𝑆 ≠ ( ⁄𝑑𝑆 𝑆) ! ¨ Integration and especially derivations are not well defined ¨ Usual calculus 𝑊(𝑡), when (𝛿𝑡) →0, (𝛿𝑊)~𝛿𝑡, (𝛿𝑊).~(𝛿𝑡). ¨ Stochastic calculus we still have (𝛿𝑊) → 0, BUT WE ALSO HAVE (𝛿𝑊).~𝛿𝑡 so higher orders are mixed together ¨ Coin toss (+1, -1). Average is 0, variance scales linearly with the number of flips ¨ Can you think of processes where variance goes to 0 and we need to go to the next order? ¨ Can you think of a process where (𝛿𝑊). scales maybe as (𝛿𝑡)0, where 𝑈 < 1 ? ¨ Also, if you think about it, when 𝛿𝑊is stochastic, somehow (𝛿𝑊). is now deterministic, or at least has a deterministic component, the inverse is NOT true 24
  • 25. Luc_Faucheux_2020 Always better to integrate than to differentiate ¨ If we have a stochastic process 𝑊(𝑡), we would like to work with functions 𝑓(𝑊) (please note that those functions are usually well behaved, meaning differentiable and such, without going into too much math) ¨ 𝑓(𝑊) is differentiable in 𝑊 ¨ 𝑊(𝑡) is NOT differentiable in 𝑡 ¨ But we are really dealing with 𝑓(𝑊 𝑡 ) so we would like to write things like ¨ 𝛿𝑓 = )$ )1 . )1 )+ . 𝛿𝑡 which is one way to write the traditional “chain rule” ¨ In integral form, the chain rule would read something like: ¨ 𝑓 𝑊 𝑡 − 𝑓 𝑊 0 = ∫234 23+ )$ )1 . )1 )2 . 𝑑𝑠 ¨ The crux of the issue in stochastic calculus is that we do not know what )1 )2 means, and so we should NOT expect to be able to rely on the usual chain rule 25
  • 26. Luc_Faucheux_2020 The whole Ito-Stratanovitch thing ¨ Essentially, ITO breakthrough was to find a way to define: ¨ 𝑓 𝑊 𝑡 − 𝑓 𝑊 0 = ∫234 23+ )$ )1 . )1 )2 . 𝑑𝑠 = ∫234 23+ )$ )1 . 𝑑𝑊 ¨ ITO invented the field of stochastic integrals ¨ The essence of it is that 𝑑𝑊(𝑡) is a “jump” ¨ ITO (1950) defines the ITO integral as a limit of a sum where the value of )$ )1 is taken “before the jump”. A consequence of this convention is that the usual rules of calculus (chain rule, Leibniz rule,..) do not apply. For example 𝑑 𝑙𝑛𝑆 ≠ ( ⁄𝑑𝑆 𝑆) ! On the other hand the ITO integral has the nice property to be a martingale (zero expected value) ¨ Stratanovitch (STRATO) defines the STRATO integral as a limit of a sum where the value of )$ )1 is taken “in the middle of the jump”. A rather intriguing consequence is that in STRATO calculus the usual rules of calculus are (formally!) respected. 𝑑 𝑙𝑛𝑆 = ( ⁄𝑑𝑆 𝑆). HOWEVER the STRATO integral is NOT a martingale. 26
  • 27. Luc_Faucheux_2020 The whole Ito-Stratanovitch thing - II ¨ So we will see how to not get confused between the two equally valid interpretations of the integrals (from a really theoretical point of view mathematicians prefer ITO because it is defined over a wider range of functions than STRATO, but really nothing we should concern ourselves at this point). ¨ This will take us some time, first going over the Riemann integral in regular calculus ¨ Then defining the ITO ¨ Then looking at the relationship between ITO and STRATO integral ¨ Then explicitly proving the ITO lemma, then the equivalent STRATO lemma ¨ From then we look at the SDE (SIE), and we try to map the relations between a specific SDE and its associated PDE for the PDF ¨ SDE: Stochastic Differential Equation ¨ SIE: Stochastic Integral Equation ¨ PDE: Partial Differential Equation (that is usual regular calculus, but not easy by any means) ¨ PDF: Probability Density Function 27
  • 28. Luc_Faucheux_2020 The whole Ito-Stratanovitch thing - III ¨ So apologies in advance if those slides feel pedestrian at time, but unfortunately in order to be somewhat rigorous without losing the intuition, and in order to convince ourselves that we are on somewhat firm ground to justify what we write (without having a 100% rigorous mathematical proof), we have to walk before we run. ¨ Alternatively, you could be a genius like Vincent Doelin and bypass the entire theory of stochastic integrals and express everything as a Brownian time change, 40 years or so before everyone else, while fighting WWII in the Ardennes as a radio operator. ¨ If I have time, part V of those decks will be on that ¨ Also the ITO-STRATO controversy in the 1990s was linked to the concept of thermal ratchets, Brownian motors, biological motors, hence quite a few articles on the subject. ¨ This is also linked to an individual that physicists refer to as the Maxwell’s demon ¨ In deck III we will revisit this unsavory character, who prompted me spending a lot of time on ITO-STRATO over my life and as a PhD student. 28
  • 29. Luc_Faucheux_2020 The whole Ito-Stratanovitch thing - IV ¨ The Maxwell demon putting the moves on Mr. Tompkins fiancée and trying to impress her with his tennis skills 29
  • 30. Luc_Faucheux_2020 The whole Ito-Stratanovitch thing - V ¨ A more common representation of the demon 30
  • 31. Luc_Faucheux_2020 The whole Ito-Stratanovitch thing - VI ¨ There are actually applications in Finance of the Maxell demon, known as the Parrondo paradox. ¨ Two trading strategies (PM at a hedge fund) on average lose money (B and C, blue and green line) ¨ 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 31
  • 32. Luc_Faucheux_2020 Just a taste of how powerful Vincent Doelin was ¨ SDE have usually the form: ¨ 𝑑𝑋 𝑡 = 𝑎 𝑋 𝑡 , 𝑡 . 𝑑𝑡 + 𝑏 𝑋 𝑡 , 𝑡 . 𝑑𝑊, which really should always be written as SIE: ¨ 𝑋 𝑡5 − 𝑋 𝑡6 = ∫+3+6 +3+5 𝑑𝑋 𝑡 = ∫+3+6 +3+5 𝑎 𝑋 𝑡 , 𝑡 . 𝑑𝑡) + ∫+3+6 +3+5 𝑏 𝑋 𝑡 , 𝑡 . 𝑑𝑊(𝑡) ¨ The whole theory of ITO is trying to define what exactly is : ∫+3+6 +3+5 𝑏 𝑋 𝑡 , 𝑡 . 𝑑𝑊(𝑡) ¨ What Doelin did essentially was to say: hey I do not need to define this integral, which is subject to the exact convention of “where to take the value of 𝑏 𝑋 𝑡 , 𝑡 before, during or after the jump 𝑑𝑊(𝑡)”, and run into all sort of Ito-Stratanovitch confusion, because I am a genius, whatever that integral is, it is equal to : ¨ ∫+3+6 +3+5 𝑏 𝑋 𝑡 , 𝑡 . 𝑑𝑊(𝑡) ≝ 𝑊(∫+3+6 +3+5 𝑏 𝑋 𝑡 , 𝑡 .. 𝑑𝑡) which is perfectly well defined. ¨ Boom. Microphone drop. ¨ And by the way he burnt most of his research before killing himself to avoid capture by the Germans, so who knows what else he had discovered…. 32
  • 33. Luc_Faucheux_2020 ITO and DOELIN comparison in 2 pictures 33
  • 34. Luc_Faucheux_2020 Ito-Doblin and stochastic calculus is.. ¨ “First and foremost defined as an integral calculus” ¨ Really the only thing that works is integrating stochastic processes. ¨ There are theories of differentiations for stochastic variables ¨ Malliavin calculus (1980) ¨ One day I will try to understand what that actually means. I still have no idea. But I should try after all Malliavin is also French ¨ So even the Ito lemma as we know it is really better expressed in integral form ¨ However most textbooks do present it in “differential” form like a Taylor expansion, or even deal with SDE quite liberally. This is sometimes for ease of notations and we will fall into that pattern also. Note that this is a FORMAL equivalence to regular calculus equations like Taylor expansion. Taking those literally leads to mistakes, as we will demonstrate 34
  • 35. Luc_Faucheux_2020 Why are PDEs so important, and why SDEs are terrifying ¨ PDEs are Partial Differential Equations )!7(!,+) )!! = )7(!,+) )+ ¨ Usually in the context of stochastic calculus they will appear for the PDF (Probability Density Function) of the stochastic variable 𝑋 (like a Gaussian), 𝑃(𝑥, 𝑡) or for functions of that variable (call payoff or 𝐶(𝑋, 𝜎, 𝐾, 𝑇) for example) ¨ PDEs are well defined and well known (since 1700), tons of knowledge on how to deal with those (Navier-Stokes in fluid mechanic, Maxwell equations in electro-magnetism, Fokker- Planck, Feynman-Kac,..) ¨ Once you know the PDE, you know ALL the moments of the distribution in the case of a PDE ¨ PDEs are complete, SDEs are incomplete ¨ SDEs are Stochastic Differential equations 𝑑𝑆 = 𝜇. 𝑆. 𝑑𝑡 + 𝜎. 𝑆. 𝑑𝑊 ¨ No one really knows how to deal with them, especially if the volatility is a function of the stochastic variable (and it always is, or when you look at functions of X) ¨ But sometimes you can get rid of a SDE and use a related PDE (that was the trick that Black Sholes discovered and got a Nobel prize for), or Dupire equation ¨ Also stay away from SDE, always look for the PDE 35
  • 36. Luc_Faucheux_2020 The structure of those slides ¨ We will first start with the usual textbook (Hull) that presents essentially what are Ito SDEs and Ito lemma being a regular Taylor expansion just making sure that we go high enough to keep all the terms linear in time and linear in the stochastic driver ¨ So in some ways, in the usual calculus 𝑊(𝑡), when (𝛿𝑡) →0, (𝛿𝑊)~𝛿𝑡, (𝛿𝑊).~(𝛿𝑊). ¨ Stochastic calculus we still have (𝛿𝑊) → 0, BUT WE ALSO HAVE (𝛿𝑊).~𝛿𝑡 so higher orders are mixed together, but if we keep the right terms we should be fine ¨ This is usually where most textbooks in finance stops ¨ We will show by using something called Stratonovitch convention, that treating the SDEs and the Taylor expansion the way we would do in regular calculus is wrong. ¨ In fact the Ito lemma and the Ito SDEs are just a formal manner to write Integrals and SIE, which is really the only thing that you can do in stochastic calculus ¨ Remember, you always read that a stochastic process is NOT differentiable, yet somehow we all proceed happy to write Stochastic DIFFERENTIAL Equations, and writing Ito lemma as a Taylor expansion, without really thinking twice about it 36
  • 37. Luc_Faucheux_2020 The structure of those slides - II - Integrals ¨ We then take a couple of steps back and go over a review of the regular integrals (Riemann) in the regular calculus ¨ We extend this to the realm of stochastic calculus ¨ We show that unlike the regular case, the point taken in the partition buckets (the mesh) actually matters. One convention is to take the starting point (ITO). Another one is to take the middle point (Stratonovitch) ¨ We then derive the relationships between the Ito and the Strato integral ¨ At this point there is no finance theory but it will be worth keeping in mind that the Ito integral will have the property of being a martingale (zero expected value) ¨ Note that things like “partition” or ”mesh” will not be super rigorously defined, but we leave that to mathematicians, we use those concepts assuming that they are well defined, and we can check that indeed they are over a certain range of “non-pathological” functions or geometries 37
  • 38. Luc_Faucheux_2020 The structure of those slides - III - Lemma ¨ From what we learned from looking at those integrals, we then look at the issue around making some operations on those integrals. This is where we look at Ito lemma ¨ Ito lemma is crucial in finance ¨ We will see that formally we can write the lemma, and perform operations on functions and integral, in a formal manner that looks like regular calculus, with the exceptions that you have to go “one more up” in any kind of derivations or Taylor expansion, in order to capture ALL the terms linear in time and linear in the stochastic driver ¨ In doing so the “usual” rules of calculus (derivations, integrations, Leibniz,..) are no longer true in stochastic calculus using the Ito interpretation of the integral ¨ We will compare this with the Strato integral ¨ We will show that the ”usual” rules of calculus are still formally present in Strato calculus (remember this is a formal analogy). This can be useful when explicitly solving equations ¨ We then show the relationship between the Ito and the Strato lemma 38
  • 39. Luc_Faucheux_2020 The structure of those slides - IV - SDE ¨ From what we learned looking at the integrals and how to manipulate them, we can try to look at what would be SDE, Stochastic Differential Equations ¨ Remember though, that just the same way we can formally write Ito and Strato lemma in “differential” form for ease of notation, those are ALWAYS simpler way to write down what would really be equations dealing with integrals ¨ In stochastic calculus, I do not know what a differentiation would actually mean ¨ All I can do really is to integrate ¨ Like the integral and the lemma, we will show the relationship between the Ito SDE and the Strato SDE (or really more exactly the Ito SIE and the Strato SIE), and introduce the so-called “drift” between the two representations ¨ SIE: Stochastic Integral Equations 39
  • 40. Luc_Faucheux_2020 The structure of those slides - V - PDE ¨ From the SIE, we then explore the correspondence between the SIE and the PDE (Partial Differential Equation) for the PDF (Probability Distribution Function) of the stochastic variable. ¨ PDE are just part of the regular calculus, there is no issue there ¨ SIE and SDE depends on the interpretation (Ito or Strato). ¨ We will then by extension look at PDE that would correspond to a specific SDE under Ito, and similarly under Strato. ¨ This is where it can get a little confusing (or even more confusing that it already is) ¨ We will revisit our old friend the Maxwell demon and see why it was so confusing in the 1990s when applied to biological concepts of thermal ratchets or Brownian motors ¨ Because a PDE is “exact” (once you know the PDE, and if you can solve it you have the PDF, so you have exactly all the moments of the stochastic variable) whereas an SDE only has the first two moments in most cases, and is also subject to interpretation, it is worth keeping in mind that fact: a PDE is not subject to interpretation. An SDE is subject to interpretation, and needs to be treated with great caution 40
  • 41. Luc_Faucheux_2020 The structure of those slides - VI - PDE ¨ We will look at some cases of PDEs from the world of Physics in order to gain some more intuition on what is a firm ground to stand on (PDE), and a somewhat more recent and still a little shaky one (SDE) ¨ This will be a somewhat indirect introduction to a fundamental theorem that links the world of “regular” well known calculus of PDE (400 years in the making) with the more recent one having to do with probability and stochastic calculus (only 100 years in the making), and still very new. ¨ This year Abel prize was given to pioneers of the ergodic theory, that essentially to crudely oversimplify, try to find the solutions of PDEs by using properties of the associated SDEs ¨ We can then go back to Black-Sholes with a renewed belief in how justified we are in our derivation, in particular we tried to “get away” using simple Taylor expansions and just keeping some higher orders, we showed however how wrong it can get very quickly. 41
  • 42. Luc_Faucheux_2020 ¨ Since we will first fall into the trap of treating an SDE the way we would do usual calculus, a number of slides in this deck are WRONG. It is sometimes more useful to learn from mistakes than to follow a magistral correct demonstration. ¨ So in order to not make the whole deck a complete garbage of my random wrong ramblings on stochastic calculus while going up the river to meet colonel Kurtz, I have put a stamp “WRONG” across those slides. ¨ So do not read those slides at face value, but know that those are WRONG. ¨ It might be an interesting exercise for you to convince yourself why those slides are wrong. ¨ Again, I left them there because I think it is quite instructive to identify the fault in a derivation. ¨ We could have kept all the slides in ITO calculus, but it is quite enlightening to also look at STRATANOVITCH calculus, if only to convince ourselves that we really understand ITO (If we understand the difference between two things, then most likely that increases our knowledge about each of those things) A note of caution 42
  • 43. Luc_Faucheux_2020 Why this whole thing about Ito calculus? Hull textbook ¨ Hull – White chapters 13, 14 and 15 ¨ People got excited about stock prices trading as a percentage (people expect a “return”), p.306, and so what mattered was the return of the stock 𝑆, or ⁄∆𝑆 𝑆 ¨ So then they started writing things like : 𝑑𝑆 = 𝜇. 𝑆. 𝑑𝑡 + 𝜎. 𝑆. 𝑑𝑊, (p.307) ¨ And then they got stuck, because 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊, where 𝑏 𝑥, 𝑡 is a function of the stochastic variable 𝑥, is not something we know how to deal with (p.306, and no, it is NOT a “small approximation” as they claim) ¨ So you need to use a ”guess” on how to deal with 𝑏 𝑥, 𝑡 , which is why it is called a “lemma” ¨ Ito (1951) guessed that you can write 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊 as ∆𝑥 = 𝑎 𝑥, 𝑡 . ∆𝑡 + 𝑏 𝑥, 𝑡 . ∆𝑊 or δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥, 𝑡 . 𝛿𝑊 ¨ That seems like a good guess but then the rules of calculus are no longer applicable, you can barely derive without making a mistake, and forget about trying to integrate (p. 311) ¨ Now you get this weird thing where 𝑑 𝑙𝑛𝑆 = ( ⁄𝑑𝑆 𝑆) − ( ⁄𝜎. 2). 𝑑𝑡, (p.312) 43
  • 44. Luc_Faucheux_2020 Why chose Ito then? 44 ¨ Someone else made a “better” guess, Stratonovitch (1966) ¨ Strato guessed that you can write 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊 as ∆𝑥 = 𝑎 𝑥, 𝑡 . ∆𝑡 + 𝑏 𝑥 + Δ𝑥/2, 𝑡 . ∆𝑊 or δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥 + 𝛿𝑥/2, 𝑡 . 𝛿𝑊 ¨ Within Strato’s convention, the usual rules of calculus FORMALLY apply 𝑑 𝑙𝑛𝑆 = ( ⁄𝑑𝑆 𝑆), and the chain rule is verified in the usual manner ¨ So this is super confusing ¨ We will look at both Ito and Strato, and understand how to go from the SDE (Stochastic Differential Equation) to the PDE (Partial Differential Equation) and the correct PDF (Probability Density Function). ¨ We will go over the binomial trees and binomial distribution, and its limit the Gaussian distribution (HW chapter 13, p.296-299) ¨ We will show why the Gaussian distribution is so common and so central to everything (central limit theorem)
  • 45. Luc_Faucheux_2020 Actually it is not called Ito, but Ito-Doeblin ¨ It was discovered quite recently (2000) that Vincent Doeblin, born Wolfgang Doeblin, Ph.D. at 23, and drafted in the French army in 1938, while posted in the Ardennes as a phone operator, essentially worked out Ito calculus on his own and sent his results to the Academie des Sciences “sous plis scelle”. ¨ He shot himself after burning the rest of his notes when the German army was about to advance and take over his positions ¨ So wo knows what else was in his notes? Malliavin calculus maybe ? 45
  • 46. Luc_Faucheux_2020 Ito and Stratonovitch “Taylor” expansion ¨ Ito (1951) guessed that you can write 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊 as δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥, 𝑡 . 𝛿𝑊 ¨ If 𝐹(𝑥) a function of x, the corresponding SDE as a result of Taylor expansion in ITO is: ¨ 𝛿𝐹 = )9 )! . 𝛿𝑥 + - . . )!9 )!! . 𝑏.. 𝛿𝑡 ¨ That is the celebrated Ito lemma, or how the chain rule gets modified in Ito stochastic calculus ¨ Strato guessed that you can write 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊 as δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥 + 𝛿𝑥/2, 𝑡 . 𝛿𝑊 ¨ So you think that you could write something like this: the SDE for 𝐹(𝑥) in Stratonovitch convention is: ¨ 𝛿𝐹 = )9 )! . 𝛿𝑥 + - . . )!9 )!! . 𝑏.. 𝛿𝑡 + - . . )9 )! . )5 )! . 𝑏. 𝛿𝑡 46
  • 47. Luc_Faucheux_2020 Ito and Stratonovitch “Taylor” expansion - II ¨ We would like to write something like this for Stratonovitch, following the rule of expanding to the second order and keeping the terms linear in time and linear in the stochastic driver (so essentially in a way use Ito calculus in the Stratonovitch convention) ¨ Bear in mind that this is absolutely wrong ¨ It is quite insightful to go through it though and see where it is wrong ¨ So we have a function 𝐹 of the stochastic variable 𝑥, the function 𝐹 in itself is nothing weird, it is a regular function that we assume to be differentiable 𝐹(𝑥) ¨ Ito calculus tells us that if we want to look at the variations of that function in terms of the stochastic variable 𝑥, assuming a process 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊 ¨ You can formally write within Ito calculus: δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥, 𝑡 . 𝛿𝑊 ¨ 𝛿𝐹 = )9 )! . 𝛿𝑥 + - . . )!9 )!! . 𝑏.. 𝛿𝑡 47
  • 48. Luc_Faucheux_2020 Ito and Stratonovitch “Taylor” expansion - III ¨ More exactly ¨ 𝛿𝐹 = )9 )! . 𝛿𝑥 + - . . )!9 )!! . 𝛿𝑥. ¨ And : δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥, 𝑡 . 𝛿𝑊 ¨ So keeping the terms linear in time and in the stochastic driver ¨ 𝛿𝑥. = 𝑏.. 𝛿𝑡 ¨ And we get the usual Ito formula: 𝛿𝐹 = )9 )! . 𝛿𝑥 + - . . )!9 )!! . 𝑏.. 𝛿𝑡 ¨ The “canonical” example is usually 𝐹 𝑥 = 𝐿𝑛(𝑥), with )9 )! = - ! , and )!9 )!! = :- !!, and with the stochastic process: δ𝑥 = 𝑥. 𝛿𝑊, so 𝑏 = 𝑥, and 𝑏. = 𝑥. ¨ And so we get: 𝛿 𝐿𝑁 𝑥 = - ! . 𝛿𝑥 + - . . :- !! . 𝑏.. 𝛿𝑡 = - ! . 𝑥. 𝛿𝑊 + - . . :- !! . 𝑥.. 𝛿𝑡 48
  • 49. Luc_Faucheux_2020 Ito and Stratonovitch “Taylor” expansion - IV ¨ Or again: ¨ 𝛿 𝐿𝑁 𝑥 = 𝛿𝑧 − - . . 𝛿𝑡 = ;! ! − - . . 𝛿𝑡 ¨ Whereas the “usual rule of calculus would read : 𝛿 𝐿𝑁 𝑥 = ;! ! ¨ That is the usual example in most textbooks, especially in Finance ¨ NOW comes the weird little Stratanovitch trick, where we assume that we can write something like : ¨ Strato guessed that you can write 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊 as δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥 + 𝛿𝑥/2, 𝑡 . 𝛿𝑊 ¨ So again we follow the “Ito” rule of calculus by expanding in linear terms in time and in the stochastic driver 49
  • 50. Luc_Faucheux_2020 Ito and Stratonovitch “Taylor” expansion - V ¨ 𝛿𝐹 = )9 )! . 𝛿𝑥 + - . . )!9 )!! . 𝛿𝑥. ¨ And : δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥 + 𝛿𝑥/2, 𝑡 . 𝛿𝑊 ¨ Or: δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥, 𝑡 . 𝛿𝑊 + ;! . . )5 )! . 𝛿𝑊, which keeping only the terms linear in time and in the stochastic driver 𝛿𝑊, leads to: ¨ δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥, 𝑡 . 𝛿𝑊 + - . . )5 )! . 𝑏. 𝛿𝑡 ¨ And so: 𝛿𝐹 = )9 )! . 𝛿𝑥 + - . . )!9 )!! . 𝛿𝑥. ¨ 𝛿𝐹 = )9 )! . 𝛿𝑥 + - . . )!9 )!! . 𝑏.. 𝛿𝑡 + - . . )9 )! . )5 )! . 𝑏. 𝛿𝑡 50
  • 51. Luc_Faucheux_2020 Ito and Stratonovitch “Taylor” expansion - VI ¨ So we think that within the Ito rules of calculus (doing a Taylor expansion and keeping only the terms linear in time and the stochastic driver) but following the Stratonovitch convention we can write something like: ¨ 𝛿𝐹 = )9 )! . 𝛿𝑥 + - . . )!9 )!! . 𝑏.. 𝛿𝑡 + - . . )9 )! . )5 )! . 𝑏. 𝛿𝑡 ¨ The “canonical” example again is 𝐹 𝑥 = 𝐿𝑛(𝑥), with )9 )! = - ! , and )!9 )!! = :- !!, and with the stochastic process: δ𝑥 = 𝑥. 𝛿𝑊, so 𝑏 = 𝑥, and 𝑏. = 𝑥., and )5 )! = 1 ¨ So we get: ¨ 𝛿 𝐿𝑁 𝑥 = - ! . 𝛿𝑥 − - . . - !! . 𝛿𝑡 + - . . - ! . 1. 𝑥. 𝛿𝑡 ¨ The last two terms cancel out and we get: 𝛿 𝐿𝑁 𝑥 = - ! . 𝛿𝑥 ¨ Which is the usual result in ”regular” (non stochastic) calculus, and we think that we now understood stochastic calculus because textbooks are telling us that the usual rules of calculus are preserved in Strato 51
  • 52. Luc_Faucheux_2020 Ito and Stratonovitch “Taylor” expansion - VII ¨ So far we seem to be pretty happy because we found that for the canonical example, the “usual rules of calculus” are preserved when using the Stratonovitch convention within the Ito calculus (meaning that we are using formally a Taylor expansion, making sure to keep all the terms linear in time and in the stochastic driver, i.e. the Ito rule of calculus, but assuming that we are taking the “mid-point” of the jump for functions multiplying this jump, i.e. what we think to be what the Stratonovitch convention is) ¨ We will show that nothing could be more wrong ¨ Both Ito and Stratonovitch are calculus on their own on the same footing ¨ We just cannot do the Taylor expansion within the Stratonovitch framework ¨ Not super obvious, took me a long time to be confused about it, the point is to always go back to the fact that in stochastic calculus the only thing that you can write is an integral, and sometimes for ease of notation we write something that looks like a Taylor expansion or and SDE. But this is only a formal way of writing, and the fact that it looks like regular calculus does NOT allow us to use those equations “as is” 52
  • 53. Luc_Faucheux_2020 Ito and Stratonovitch “Taylor” expansion – VII - a ¨ So for example, when in Ito calculus we write the chain rule as: ¨ 𝛿𝐹 = )9 )1 . 𝛿𝑊 + - . . )!9 )1! . 𝛿𝑊. ¨ What we are really writing, (and what we should always write from time to remind us, since we do not have enough paper and ink to write it all the time at every step) is: ¨ 𝑓 𝑊 𝑡5 − 𝑓 𝑊 𝑡6 = ∫+3+6 +3+5 )$ )* . ([). 𝑑𝑊(𝑡) + - . ∫+3+6 +3+5 )!9 )1! (𝑊 𝑡 ). 𝑑𝑡 ¨ In the ”limit” of small time increments, this can be written formally as the Ito lemma: ¨ 𝛿𝑓 = )$ )* . 𝛿𝑊 + - . . )!9 )1! . 𝛿𝑡 ¨ We will go over it once we rebuild our knowledge of stochastic calculus around the integral ¨ Here we are following mostly Thomas Mikosch “Elementary Stochastic Calculus with Finance in View”, a wonderful little book that is at times frustrating for some of the weird notations that he uses. 53
  • 54. Luc_Faucheux_2020 Ito and Stratonovitch “Taylor” expansion - VIII ¨ A very quick manner to realize how wrong we are is to use another example: ¨ 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊 ¨ with 𝑏 𝑥, 𝑡 = σ. 𝑥 , )5 )! = 𝜎 , 𝑏. = 𝜎.. 𝑥. and 𝑎 𝑥, 𝑡 = 0, ¨ 𝐹 𝑥 = 𝑥., )9 )! = 2𝑥 , )!9 )!! = 2 ¨ Using ITO, 𝛿𝐹 = )9 )! . 𝛿𝑥 + - . . )!9 )!! . 𝑏.. 𝛿𝑡 = 2𝑥. 𝛿𝑥 + 𝜎.. 𝛿𝑡 ¨ Using Strato, 𝛿𝐹 = )9 )! . 𝛿𝑥 + - . . )!9 )!! . 𝑏.. 𝛿𝑡 + - . . )9 )! . )5 )! . 𝑏. 𝛿𝑡 ¨ We then get: 𝛿𝐹 = 2𝑥. 𝛿𝑥 + 𝜎.. 𝛿𝑡 + - . . 2𝑥. σ. σ. 𝑥. 𝛿𝑡 = 2𝑥. 𝛿𝑥 + 2.𝜎.. 𝛿𝑡 ¨ So neither what we call Ito and what we call Strato do follow the usual rule of calculus. That should be a sign that we did something very wrong when we loosely used the Taylor expansion and applied it to the Stratonovitch case, because textbooks are telling us that the usual rules of calculus (chain rule) are ”preserved” (similar in form) in Strato 54
  • 55. Luc_Faucheux_2020 Ito and Stratonovitch “Taylor” expansion - IX ¨ The answer is not completely obvious in identifying what we did wrong ¨ Conversely, someone could say it is absolutely obvious, because stochastic process are NOT differentiable, and so any kind of Taylor expansion is wrong ¨ As a note, even though in most textbooks in Finance the Ito lemma is expressed as a Taylor expansion using the formal rules of calculus, the rule is that with stochastic processes you can NEVER differentiate (at least in a manner that makes sense and is safe), you can ONLY integrate ¨ And so a more formally correct formulation of the Ito Lemma is: ¨ 𝐹 𝑊 𝑡 = 𝑇 − 𝐹 𝑊 𝑡 = 𝑆 = ∫+3< +3= )9 )1 . 𝑑𝑊 + - . ∫+3< +3= )!9 )1! . 𝑑𝑡 ¨ Or also: ¨ ∫+3< +3= 𝑑𝐹(𝑊 𝑡 ) = ∫+3< +3= )9 )1 . 𝑑𝑊(𝑡) + - . ∫+3< +3= )!9 )1! . 𝑑𝑡 55
  • 56. Luc_Faucheux_2020 Ito and Stratonovitch “Taylor” expansion - X ¨ What is then the correct formulation of Stratonovitch lemma? ¨ Is that? ¨ 𝐹 𝑥 𝑡 = 𝑇 − 𝐹 𝑥 𝑡 = 𝑆 = ∫+3< +3= )9 )! . 𝑑𝑥 + ∫+3< +3= [ - . )!9 )!! . 𝑏. + - . . )9 )! . )5 )! . 𝑏]. 𝑑𝑡 ¨ Or: ¨ ∫+3< +3= 𝑑𝐹(𝑥 𝑡 ) = ∫+3< +3= )9 )! . 𝑑𝑥 + ∫+3< +3= [ - . )!9 )!! . 𝑏. + - . . )9 )! . )5 )! . 𝑏]. 𝑑𝑡 ¨ Or to identify the distinctions between the two: ¨ STRATO(∫+3< +3= 𝑑𝐹(𝑥 𝑡 ))=ITO(∫+3< +3= 𝑑𝐹(𝑥 𝑡 ))+ - . ∫+3< +3= )9 )! . )5 )! . 𝑏. 𝑑𝑡 ¨ This is still wrong, as we mixed Taylor expansion using the usual rules of calculus (Ito rules) with something completely different. We will now show what is the correct way to look at it using integrals. 56
  • 57. Luc_Faucheux_2020 Simple rules of stochastic calculus (cheat sheet) ¨ NEVER EVER EVER work with processes where the volatility is a function of the stochastic variable ¨ If you do or have no choice, transform it into a constant volatility equation (Hull p. 320), or find a way to set the volatility term to 0 (p.330), or give up and come up with something different (SABR model) ¨ If you are working in Finance (and ”discrete processes) -> use ITO ¨ If you are working with a DIGITAL computer -> use ITO ¨ If you are working with an ANALOG computer -> use STRATO ¨ If you are working in physics, and you do not like to break the time invariance, and you are also not that smart, so you want the usual rules of calculus -> use STRATO ¨ In ALL cases, especially when working with the SDE, and discrete computer simulations, ALWAYS check that your drift has not been “polluted” by the variable diffusion (“spurious” drift, Ryter 1980) 57
  • 58. Luc_Faucheux_2020 58 Building our knowledge of stochastic calculus around the integral
  • 59. Luc_Faucheux_2020 We have to start from the basics ¨ Because clearly using ITO lemma as a Taylor expansion where we keep certain terms and using still the usual rules of calculus is wrong. ¨ As we hinted at it, we should never write something that looks like a differential but always as an integral. ¨ Let’s now go through that derivation, and start with the usual Riemann integral in the usual calculus ¨ We then show that extending that concept to a stochastic variable is not well defined, and needs a convention, or interpretation of the integral. ITO is one interpretation, STRATO is another one. ¨ We show the correspondence between the two interpretations. ¨ Extending the concept of the integral being a limit of sums subject to an interpretation, we then derive the ITO lemma as well as the STRATO lemma 59
  • 60. Luc_Faucheux_2020 “Regular” Riemann integrals (definite integrals) ¨ Riemann integrals are the regular integrals ¨ Interval [a,b] on regular ”continuous” variable t ¨ N sections of width (𝑏 − 𝑎)/𝑁, left side 𝐿>, right side 𝑅>, and middle 𝑀> ¨ The Riemann integral ∫+36 +35 𝑓 𝑡 . 𝑑𝑡 is the limit when (𝑁 → ∞) of the Riemann sums ¨ LEFT Riemann Sum: 𝐿𝑅𝑆 = 5:6 / ∑>3- >3/ 𝑓(𝐿>) ¨ RIGHT Riemann Sum: 𝑅𝑅𝑆 = 5:6 / ∑>3- >3/ 𝑓(𝑅>) ¨ MIDDLE Riemann Sum: 𝑀𝑅𝑆 = 5:6 / ∑>3- >3/ 𝑓(𝑀>) ¨ SOMETHING Riemann Sum: 𝑆𝑅𝑆 = 5:6 / ∑>3- >3/ 𝑓(𝑆>) where 𝑆> is somewhere in the section indexed by k, we could also define irregular partitions or “mesh” if we like ¨ All those different sums converge to the same integral 60
  • 61. Luc_Faucheux_2020 61 ITO and STRATO integrals in the simple case W(t)
  • 62. Luc_Faucheux_2020 Stochastic Integrals ¨ When not integrating over a “regular” continuous variable t but over a stochastic variable X, those sums do NOT converge to the same value. What does that even look like? ¨ So first of all we would like to write something like this ∫*3*6 *3*5 𝑓 𝑋 . 𝑑𝑋 ¨ The problem is that this is not well defined, what is the path over 𝑋 that we will integrate over? Remember 𝑋 is really 𝑋(𝑡) and is stochastic (jumps all over the place) ¨ So really the only integration we can do is over 𝑡 ¨ So we are looking for something like this: ∫+3+6 +3+5 𝑓 𝑋 𝑡 . 𝑑𝑋(𝑡) ¨ When 𝑡 increases by a small amount 𝛿𝑡, 𝑋(𝑡) jumps by a small amount δ𝑋(𝑡) ¨ So, breaking the time interval into 𝑁 sections of small time increment 𝛿𝑡 = (+5:+6) / ¨ We looking at something like : SOMETHING = ∑>3- >3/ 𝑓(𝑋(𝑡>)). [𝑋(𝑡>?-) − 𝑋(𝑡>)] 62
  • 63. Luc_Faucheux_2020 Ito Integral ¨ That something is the ITO sum, which converges to the ITO integral. Note that at this point it is just how we define it ¨ ∫+3+6 +3+5 𝑓 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) = lim /→@ {∑>3- >3/ 𝑓(𝑋(𝑡>)). [𝑋(𝑡>?-) − 𝑋(𝑡>)]} ¨ Note in the integral the usual (.) is replaced by ([) to explicitly indicate ITO convention ¨ This is to indicate that we take for 𝑓(𝑋(𝑡>) the value of 𝑓 𝑋 BEFORE the jump ¨ This is to be compared to the ITO lemma where 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊 gets written in discrete manner as ∆𝑥 = 𝑎 𝑥, 𝑡 . ∆𝑡 + 𝑏 𝑥, 𝑡 . ∆𝑊 or δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥, 𝑡 . 𝛿𝑊 ¨ Note also that to be fairly rigorous there are a lot of conditions that need to be verified for that “SOMETHING” to converge in a well defined manner to a well defined function 63
  • 64. Luc_Faucheux_2020 Stratonovitch Integral ¨ Similar to the Stratonovitch lemma, where 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊 gets written in the discrete manner as ∆𝑥 = 𝑎 𝑥, 𝑡 . ∆𝑡 + 𝑏 𝑥 + Δ𝑥/2, 𝑡 . ∆𝑊 or δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥 + 𝛿𝑥/2, 𝑡 . 𝛿𝑊 ¨ The Stratonovitch integral is defined as: ¨ ∫+3+6 +3+5 𝑓 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = lim /→@ {∑>3- >3/ 𝑓 [𝑋(𝑡> + 𝑋(𝑡>?-)]/2). [𝑋(𝑡>?-) − 𝑋(𝑡>)]} ¨ If ITO was the “left side Riemann”, Stratonovitch is the “middle Riemann” ¨ Surprise surprise, they do NOT converge to the same value ¨ Notice in the integral the usual (.) is replaced by (∘) to indicate that we take the middle point 64
  • 65. Luc_Faucheux_2020 Reverse ITO integral ¨ Similarly we can also define a Reverse ITO (OTI?) integral as ¨ ∫+3+6 +3+5 𝑓 𝑋 𝑡 . (]). 𝑑𝑋(𝑡) = lim /→@ {∑>3- >3/ 𝑓(𝑋(𝑡>?-)). [𝑋(𝑡>?-) − 𝑋(𝑡>)]} ¨ Notice in the integral the usual (.) is replaced by (]) to indicate that we take the right side ¨ The Reverse Ito lemma would then expand 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊 in the discrete manner as ∆𝑥 = 𝑎 𝑥, 𝑡 . ∆𝑡 + 𝑏 𝑥 + Δ𝑥, 𝑡 . ∆𝑊 or δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥 + 𝛿𝑥, 𝑡 . 𝛿𝑊 ¨ This would be equivalent as “reading into the future” because we would take the value after the jump, but we do not know yet what the jump will be ¨ Stratonovitch is also sometimes said to be “similar”, as in you need to know the jump ahead of time to evaluate the function, even though you do not know the jump ahead of time ¨ ITO is well adapted to finance and to “Martingales” (you do not know the future, expected value is the current value). The technical term is that the ITO integral is “non-anticipating” 65
  • 66. Luc_Faucheux_2020 Couple of notes here ¨ It would seem that ITO would be well suited for processes that are “discontinuous” or “discrete” in nature (a very poor choice of words on my part), like: ¨ Discrete computer simulation (on a digital computer) ¨ Finance, gambling, games of chance ¨ Radioactive decay (number of particles is discrete and the rate only depends on the previous state) ¨ On the other hand STRATO seems to be better suited for “continuous” processes like most processes in Physics and Biology (diffusion, advection, chemotaxis, Brownian motors,..) ¨ Part of the confusion will arise when trying to model in a discrete fashion a “continuous” process 66
  • 67. Luc_Faucheux_2020 Conversion between ITO and Stratonovitch ¨ ∫+3+6 +3+5 𝑓 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = lim /→@ {∑>3- >3/ 𝑓 [𝑋(𝑡> + 𝑋(𝑡>?-)]/2). [𝑋(𝑡>?-) − 𝑋(𝑡>)]} ¨ ∫+3+6 +3+5 𝑓 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = lim /→@ {∑>3- >3/ 𝑓 𝑋(𝑡> + [*(+"#$):*(+")] . ). [𝑋(𝑡>?-) − 𝑋(𝑡>)]} ¨ ∫+3+6 +3+5 𝑓 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = lim /→@ {∑>3- >3/ 𝑓 𝑋(𝑡> ). [𝑋(𝑡>?-) − 𝑋(𝑡>)]} + lim /→@ {∑>3- >3/ 𝑓′ 𝑋(𝑡> . [*(+"#$):*(+")] . ). [𝑋(𝑡>?-) − 𝑋(𝑡>)]} ¨ ∫+3+6 +3+5 𝑓 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = ∫+3+6 +3+5 𝑓 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) + lim /→@ {∑>3- >3/ 𝑓′ 𝑋(𝑡> . ( [*(+"#$):*(+")] . ). [𝑋(𝑡>?-) − 𝑋(𝑡>)]} ¨ 𝑆𝑇𝑅𝐴𝑇𝑂(𝑓) = 𝐼𝑇𝑂(𝑓) + lim /→@ {∑>3- >3/ 𝑓′ 𝑋(𝑡> . ( [*(+"#$):*(+")] . ). [𝑋(𝑡>?-) − 𝑋(𝑡>)]} 67
  • 68. Luc_Faucheux_2020 Conversion between ITO and Stratonovitch 2 ¨ 𝑆𝑇𝑅𝐴𝑇𝑂(𝑓) = 𝐼𝑇𝑂(𝑓) + lim /→@ {∑>3- >3/ 𝑓′ 𝑋(𝑡> . ( [*(+"#$):*(+")] . ). [𝑋(𝑡>?-) − 𝑋(𝑡>)]} ¨ NOW this is a limit of a Riemann sum, because (𝛿𝑋).~𝛿𝑡 and is now deterministic (in the case of a simple Wiener process for 𝑋 = 𝑊) ¨ In the more general case of 𝑑𝑋 𝑡 = 𝑎 𝑋 𝑡 , 𝑡 . 𝑑𝑡 + 𝑏 𝑋 𝑡 , 𝑡 . 𝑑𝑊(𝑡) we will show that ¨ (𝛿𝑋).~𝑏 𝑋 𝑡 , 𝑡 . 𝛿𝑡 but for now let’s keep 𝑋 𝑡 = 𝑊(𝑡) the simple Brownian motion ¨ That Riemann sum converges to the definite Riemann integral - . ∫+3+6 +3+5 𝑓′ 𝑊 𝑡 . 𝑑𝑡 ¨ ∫+3+6 +3+5 𝑓 𝑊 𝑡 . (∘). 𝑑𝑊(𝑡) = ∫+3+6 +3+5 𝑓 𝑊 𝑡 . ([). 𝑑𝑊(𝑡) + - . ∫+3+6 +3+5 𝑓′ 𝑊 𝑡 . 𝑑𝑡 ¨ 𝑆𝑇𝑅𝐴𝑇𝑂(𝑓) = 𝐼𝑇𝑂 𝑓 + - . ∫+3+6 +3+5 𝑓′ 𝑊 𝑡 . 𝑑𝑡 ¨ 𝑆𝑇𝑅𝐴𝑇𝑂(𝑓) = 𝐼𝑇𝑂 𝑓 + - . . 𝑅𝐼𝐸𝑀𝐴𝑁𝑁(𝑓%) 68
  • 70. Luc_Faucheux_2020 Chain Rule (Ito lemma)- I ¨ ITO integral: ¨ ∫+3+6 +3+5 𝑓 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) = lim /→@ {∑>3- >3/ 𝑓(𝑋(𝑡>)). [𝑋(𝑡>?-) − 𝑋(𝑡>)]} ¨ Another way to express it is the following: ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = ∑>3- >3/ {𝑓(𝑋(𝑡>)) − 𝑓(𝑋(𝑡>:-))} ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim /→@ ∑>3- >3/ {𝑓(𝑋(𝑡>)) − 𝑓(𝑋(𝑡>:-))} ¨ As long as we have defined a reasonable “mesh” for the sequence of 𝑋(𝑡>) ¨ 𝑓(𝑋(𝑡>)) = 𝑓(𝑋(𝑡>:-)) + )$ )* . ([). 𝛿𝑋 + - . . )!$ )!! . ([). (𝛿𝑋). ¨ Where we use the ([) notation to reflect the “left-hand” Riemann. ¨ )$ )* . ([). 𝛿𝑋 = )$ )* 𝑋 = 𝑋(𝑡>:- ]. [𝑋(𝑡>) − 𝑋(𝑡>:-)] 70
  • 71. Luc_Faucheux_2020 Chain Rule (Ito lemma)- II ¨ - . . )!9 )!! . ([). 𝛿𝑋 . = - . . )!$ )!! [𝑋 = 𝑋(𝑡>:-)]. [𝑋(𝑡>) − 𝑋(𝑡>:-)]. ¨ )$ )* . ([). 𝛿𝑋 = )$ )* 𝑋 = 𝑋(𝑡>:- ]. [𝑋(𝑡>) − 𝑋(𝑡>:-)] ¨ 𝑓(𝑋(𝑡>)) = 𝑓(𝑋(𝑡>:-)) + )$ )* . ([). 𝛿𝑋 + - . . )!$ )!! . ([). (𝛿𝑋). ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim /→@ ∑>3- >3/ {𝑓(𝑋(𝑡>)) − 𝑓(𝑋(𝑡>:-))} ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim /→@ ∑>3- >3/ { )$ )* . ([). 𝛿𝑋 + - . . )!$ )!! . ([). (𝛿𝑋).} ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = ∫+3+6 +3+5 )$ )* . ([). 𝑑𝑋(𝑡) + ∫+3+6 +3+5 - . . )!$ )!! . ([). (𝛿𝑋). ¨ In the ”limit” of small time increments, this can be written formally as the Ito lemma: ¨ 𝛿𝑓 = )$ )* . 𝛿𝑋 + - . . )!$ )!! . (𝛿𝑋). 71
  • 72. Luc_Faucheux_2020 Chain Rule (Strato lemma)- I ¨ We will now expand around the middle point ¨ We still have: ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = ∑>3- >3/ {𝑓(𝑋(𝑡>)) − 𝑓(𝑋(𝑡>:-))} ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim /→@ ∑>3- >3/ {𝑓(𝑋(𝑡>)) − 𝑓(𝑋(𝑡>:-))} ¨ We now write it a little differently by looking at expanding around the middle point: ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim /→@ ∑>3- >3/ {𝑓(𝑋(𝑡>)) − 𝑓(𝑋(𝑡>:-))} ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim /→@ ∑>3- >3/ 𝑓 𝑋(𝑡> − 𝑓 ∘ + 𝑓 ∘ − 𝑓(𝑋(𝑡>:-))} ¨ Where for sake of clarity we noted : 𝑓 ∘ = 𝑓 *(+" ? *(+"%$ . 72
  • 73. Luc_Faucheux_2020 Chain Rule (Strato lemma)- II ¨ We then have the two following expansions of the function around the middle point: ¨ 𝑓(𝑋(𝑡>)) = 𝑓(∘) + )$ )* . (∘). ;* . + - . . )!$ )!! . (∘). ( ;* . ). ¨ 𝑓(𝑋(𝑡>:-)) = 𝑓(∘) − )$ )* . (∘). ;* . + - . . )!$ )!! . (∘). ( ;* . ). ¨ 𝛿𝑋 = 𝑋(𝑡>) − 𝑋(𝑡>:-) ¨ 𝑋(𝑡>) − *(+")?*(+"%$) . = ;* . ¨ *(+")?*(+"%$) . − 𝑋(𝑡>:-) = ;* . ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim /→@ ∑>3- >3/ 𝑓 𝑋(𝑡> − 𝑓 ∘ + 𝑓 ∘ − 𝑓(𝑋(𝑡>:-))} ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim /→@ {∑>3- >3/ )$ )* . ∘ . ;* . + - . . )!$ )!! . ∘ . ;* . . − (− )$ )* . (∘). ;* . + - . . )!$ )!! . (∘). ( ;* . ).)} 73
  • 74. Luc_Faucheux_2020 Chain Rule (Strato lemma)- III ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim /→@ {∑>3- >3/ )$ )* . ∘ . ;* . + - . . )!$ )!! . ∘ . ;* . . − (− )$ )* . (∘). ;* . + - . . )!$ )!! . (∘). ( ;* . ).)} ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim /→@ {∑>3- >3/ )$ )* . ∘ . ;* . + −(− )$ )* . (∘). ;* . )} ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim /→@ {∑>3- >3/ )$ )* . ∘ . 𝛿𝑋} ¨ And so defining the integral with the Stratonovitch convention: ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = ∫+3+6 +3+5 )$ )* . (∘). 𝑑𝑋(𝑡) ¨ In the ”limit” of small time increments, this can be written formally as the Strato lemma: ¨ 𝛿𝑓 = )$ )* . ∘ . 𝛿𝑋 ¨ Please note that we are keeping the notation : ∘ 74
  • 75. Luc_Faucheux_2020 Chain Rule (Strato lemma)- IV ¨ Using the Stratonovitch definition of the stochastic integral, we can write : ¨ 𝛿𝑓 = )$ )* . ∘ . 𝛿𝑋 ¨ This is the usual chain rule ¨ So formally in some textbooks, you will see the following statement: ¨ “We do not mean that the Stratonovitch stochastic integral is a classical (Riemann) integral. We only claim that the corresponding chain rules have a similar structure”. Mikosh p127 ¨ It is important to note that BOTH the Ito and the Stratonovitch integrals are defined in a mathematically correct manner. ¨ Ito rules of calculus are not the usual ones but the Ito integral is a martingale ¨ Strato rules of calculus are the usual ones (FORMALLY) but the Strato integral is NOT a martingale ¨ We still have to review how we treat SDE and PDE in those frameworks. 75
  • 76. Luc_Faucheux_2020 Chain Rule (Ito and Strato lemma)- I ¨ The crux of the matter is which interpretation of an integral do you want to use: ¨ Because lim /→@ {∑>3- >3/ )$ )* . ∘ . 𝛿𝑋} ¨ And ¨ lim /→@ ∑>3- >3/ { )$ )* . ([). 𝛿𝑋} ¨ Do NOT converge to the same value, unlike a regular Riemann integral ¨ In fact we just showed that ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim /→@ {∑>3- >3/ )$ )* . ∘ . 𝛿𝑋} ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim /→@ ∑>3- >3/ { )$ )* . ([). 𝛿𝑋 + - . . )!9 )!! . ([). (𝛿𝑋).} 76
  • 77. Luc_Faucheux_2020 Chain Rule (Ito and Strato lemma)- II ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim /→@ {∑>3- >3/ )$ )* . ∘ . 𝛿𝑋} ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim /→@ ∑>3- >3/ { )$ )* . ([). 𝛿𝑋 + - . . )!9 )!! . ([). (𝛿𝑋).} ¨ We formally define in integral form: ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = ∫+3+& +3+' 𝑑𝐹(𝑥 𝑡 ) ¨ And ITO: ¨ ∫+3+6 +3+5 𝐹 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) = lim /→@ {∑>3- >3/ 𝐹(𝑋(𝑡>)). [𝑋(𝑡>?-) − 𝑋(𝑡>)]} ¨ And STRATO: ¨ ∫+3+6 +3+5 𝐹 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = lim /→@ {∑>3- >3/ 𝐹 [𝑋(𝑡> + 𝑋(𝑡>?-)]/2). [𝑋(𝑡>?-) − 𝑋(𝑡>)]} 77
  • 78. Luc_Faucheux_2020 Chain Rule (Ito and Strato lemma)- III ¨ We then have for the chain rule using the ITO interpretation of the integral: ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim /→@ ∑>3- >3/ { )$ )* . ([). 𝛿𝑋 + - . . )!9 )!! . ([). (𝛿𝑋).} ¨ ∫+3+& +3+' 𝑑𝐹(𝑋 𝑡 ) = ∫+3+6 +3+5 )$ )* . ([). 𝑑𝑋(𝑡) + - . ∫+3+6 +3+5 )!9 )!! 𝑋 𝑡 . (𝑑𝑋). ¨ Note that all integrals are on : ∫+3+& +3+' () ¨ HOWEVER, the increment of integrands are different, 𝑑𝐹, 𝑑𝑋 and 𝑑𝑡 ¨ BOTH 𝐹 and 𝑋 are functions of 𝑡 ultimately, 𝐹(𝑋 𝑡 ) and 𝑋(𝑡) ¨ So only in a formal manner we write: ¨ 𝛿𝑓 = )$ )* . 𝛿𝑋 + - . . )!9 )!! . (𝛿𝑋)., which is the celebrated Ito lemma (chain rule) 78
  • 79. Luc_Faucheux_2020 Chain Rule (Ito and Strato lemma)- IV ¨ 𝛿𝑓 = )$ )* . 𝛿𝑋 + - . . )!9 )!! . 𝑏.. 𝛿𝑡, which is the celebrated Ito lemma (chain rule) ¨ We somehow convinced ourselves at first that this was just using regular Taylor expansion but just keeping higher order terms, in particular the second order in 𝑋, because: ¨ 𝛿𝑋. 𝛿𝑋~𝑏.. 𝛿𝑡 ¨ But even though it looks formally the same, the only rigorous manner in which to write it is: ¨ ∫+3+& +3+' 𝑑𝐹(𝑋 𝑡 ) = ∫+3+6 +3+5 )$ )* . ([). 𝑑𝑋(𝑡) + - . ∫+3+6 +3+5 )!9 )!! 𝑋 𝑡 . 𝑏 𝑋 𝑡 , 𝑡 .. 𝑑𝑡 ¨ With the ITO interpretation of the integral: ¨ ∫+3+6 +3+5 𝐹 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) = lim /→@ {∑>3- >3/ 𝐹(𝑋(𝑡>)). [𝑋(𝑡>?-) − 𝑋(𝑡>)]} 79
  • 80. Luc_Faucheux_2020 Chain Rule (Ito and Strato lemma)- V ¨ We then have for the chain rule using the STRATO interpretation of the integral: ¨ 𝑓 𝑋 𝑡5 − 𝑓 𝑋 𝑡6 = lim /→@ {∑>3- >3/ )$ )* . ∘ . 𝛿𝑋} ¨ ∫+3+& +3+' 𝑑𝐹(𝑋 𝑡 ) = ∫+3+6 +3+5 )$ )* . ∘ . 𝑑𝑋(𝑡) ¨ Note that all integrals are on : ∫+3+& +3+' () ¨ HOWEVER, the increment of integrands are different, 𝑑𝐹, 𝑑𝑋 and 𝑑𝑡 ¨ BOTH 𝐹 and 𝑋 are functions of 𝑡 ultimately, 𝐹(𝑋 𝑡 ) and 𝑋(𝑡) ¨ So only in a formal manner we write: ¨ 𝛿𝑓 = )$ )* . 𝛿𝑋, which is the celebrated STRATO lemma (chain rule) 80
  • 81. Luc_Faucheux_2020 Chain Rule (Ito and Strato lemma)- VI ¨ 𝛿𝑓 = )$ )* . 𝛿𝑋, which is the celebrated STRATO lemma (chain rule) ¨ HOWEVER, the only rigorous manner in which to write it is: ¨ ∫+3+& +3+' 𝑑𝐹(𝑋 𝑡 ) = ∫+3+6 +3+5 )$ )* . ∘ . 𝑑𝑋(𝑡) ¨ With the STRATO interpretation of the integral: ¨ ∫+3+6 +3+5 𝐹 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = lim /→@ {∑>3- >3/ 𝐹 [𝑋(𝑡> + 𝑋(𝑡>?-)]/2). [𝑋(𝑡>?-) − 𝑋(𝑡>)]} 81
  • 82. Luc_Faucheux_2020 Chain Rule (Ito and Strato lemma)- VII ¨ So sometimes we will just stick to the notation: ¨ Ito lemma: ¨ 𝛿𝑓 = )$ )* . ([). 𝛿𝑋 + - . . )!9 )!! . ([). 𝛿𝑋. ¨ Just to remind us that we are using stochastic calculus and that it is NOT a usual product ¨ Similarly when using STRATO we will sometimes use: ¨ 𝛿𝑓 = )$ )* . ∘ . 𝛿𝑋 ¨ To remind us that even if it looks like the regular chain rule, we are in the stochastic world where things are a little weird. ¨ Again the only rigorous manner to deal with those is to always go back to the integrals, and the interpretation of is as a limit of a sum, which is rigorous. 82
  • 83. Luc_Faucheux_2020 Chain Rule (Ito and Strato lemma)- VIII ¨ Note that if the function 𝐹(𝑋 𝑡 ) has an explicit dependency on time 𝐹(𝑋 𝑡 , 𝑡) ¨ Then for the time part the regular chain rule applies and we will obtain terms of the expression: )9 )+ . 𝑑𝑡 in BOTH ITO and STRATO ¨ Because integrating over time is a normal Riemann integral, and both sums converge to the same limit ¨ ∫+3+6 +3+5 )9 )+ . ∘ . 𝑑𝑡 = ∫+3+6 +3+5 )9 )+ . [ . 𝑑𝑡 = ∫+3+6 +3+5 )9 )+ . 𝑑𝑡 83
  • 85. Luc_Faucheux_2020 Leibniz rule - I ¨ In most textbooks, it is usually also presented as ¨ Hey do a Taylor expansion ¨ Just make sure to keep the higher order terms ¨ And you good ¨ Surprisingly it is formally the same expression ¨ Even more surprisingly, if you were to be working in a Strato world, textbooks would say: ¨ Hey just use regular calculus ¨ Since we have seen now that stochastic calculus can be tricky, let’s spend some time on convincing ourselves that we can adapt the Leibniz rule in stochastic calculus (it will also be super useful when changing numeraires in the Numeraire deck) 85
  • 86. Luc_Faucheux_2020 Leibniz rule - II ¨ It will be easier to do it once we have a more general expression for the ITO and STRATO integrals, so for now we will just state them (Leibniz rule for first order) ¨ Leibniz rule in the REGULAR calculus: ¨ 𝛿 𝑓𝑔 = 𝑔 )$ )* . 𝛿𝑋 + 𝑓 )( )* . 𝛿𝑋 ¨ Leibniz rule in the ITO calculus: ¨ 𝛿(𝑓𝑔) = 𝑔 )$ )* . ([). 𝛿𝑋 + 𝑔 - . . )!$ )*! . ([). 𝛿𝑋. + 𝑓 )( )* . ([). 𝛿𝑋 + 𝑓 - . . )!( )*! . ([). 𝛿𝑋. + )( )* . )$ )* . ([). 𝛿𝑋. ¨ Leibniz rule in the STRATO calculus: ¨ 𝛿 𝑓𝑔 = 𝑔 )$ )* . ∘ . 𝛿𝑋 + 𝑓 )( )* . ∘ . 𝛿𝑋 86
  • 88. Luc_Faucheux_2020 A worked out example to gain some intuition ¨ ∫+3+6 +3+5 𝑓 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) = lim /→@ {∑>3- >3/ 𝑓(𝑋(𝑡>)). [𝑋(𝑡>?-) − 𝑋(𝑡>)]} ¨ This is the definition of the Ito integral ¨ Let’s try with the simple case of the function 𝑓(𝑋(𝑡>)) = 𝑋(𝑡>) ¨ The Riemann sum is: 𝑆/ = ∑>3- >3/ 𝑓(𝑋(𝑡>)). [𝑋(𝑡>?-) − 𝑋(𝑡>)] ¨ For this specific case: 𝑆/ = ∑>3- >3/ 𝑋(𝑡>). [𝑋(𝑡>?-) − 𝑋(𝑡>)] 88
  • 89. Luc_Faucheux_2020 In the “regular” case – Riemann integral ¨ ∫!3!6 !3!5 𝑓 𝑥 . 𝑑𝑥 = lim /→@ {∑>3- >3/ 𝑓(𝑥>). [(𝑥>?-) − (𝑥>)]} ¨ This is the usual “triangle” representation 89
  • 90. Luc_Faucheux_2020 In the “regular” case – Riemann integral - II ¨ In the regular case, ¨ (𝑥>?-) − (𝑥>) = 𝛿𝑥 = !':!& / = ;* / ¨ Another way to look at it, is 𝑥 = 𝑡, so at every point in time, δ𝑥 = 𝛿𝑡 ¨ In particular, δ𝑥. = δ𝑡. 90 Xi i+1i-1 F(X) to integrate
  • 91. Luc_Faucheux_2020 In the stochastic case, we cannot draw that picture ¨ We cannot write (𝑥>?-) − (𝑥>) = 𝛿𝑥 = !':!& / = ;* / ¨ If anything, what we can write is (𝑥>?-) − (𝑥>) = 𝛿𝑥> = ±𝛿𝑥 ¨ And so ∑(𝑥>?-) − (𝑥>) = ∑ 𝛿𝑥> = 0 ¨ And ∑[(𝑥>?-) − (𝑥>)].= ∑ 𝛿𝑥> . = ∑ 𝛿𝑥. = 𝑁. 𝛿𝑥. 91 Xi i+1i-1
  • 92. Luc_Faucheux_2020 Another way to think about the difference ¨ The regular case, the horizontal axis is the regular non-stochastic variable ¨ Any integration of function is the regular Riemann integral ¨ In the stochastic case, the horizontal axis becomes time, and the vertical is the stochastic variable X, and we will need to integrate a function of it over the vertical axis, but whereas time is regular, a stochastic process cannot be such that each step is just the interval divided by the number of steps 92 X(t) is stochastic Time t is regular F(X(t))tointegrate
  • 93. Luc_Faucheux_2020 The regular case revisited ¨ ∫!3!6 !3!5 𝑓 𝑥 ([)𝑑𝑥 = lim /→@ {∑>3- >3/ 𝑓(𝑥>). [(𝑥>?-) − (𝑥>)]} ¨ ∫!3!6 !3!5 𝑓 𝑥 ([)𝑑𝑥 = lim /→@ ∑>3- >3/ 𝑓(𝑥>). 𝛿𝑥> = lim /→@ !':!& / . ∑>3- >3/ 𝑓(𝑥>) ¨ In the case where 𝑓(𝑥>) = 𝑥> = 𝑘. 𝛿𝑥 = 𝑘. !':!& / ¨ ∫!3!6 !3!5 𝑥([)𝑑𝑥 = lim /→@ !':!& / . !':!& / . ∑>3- >3/ 𝑘 = lim /→@ !':!& / . !':!& / . /(/?-) . ¨ ∫!3!6 !3!5 𝑥([)𝑑𝑥 = (𝑥5 − 𝑥6). lim /→@ - / . - / . /(/?-) . ¨ Using 𝑥6 = 0 without losing any generality, ¨ ∫!34 !3* 𝑥([)𝑑𝑥 = - . 𝑋., which is the usual result 93
  • 94. Luc_Faucheux_2020 The regular case revisited – II – Strato convention ¨ ∫!3!6 !3!5 𝑓 𝑥 (∘)𝑑𝑥 = lim /→@ {∑>3- >3/ 𝑓( !"#$?!" . ). [(𝑥>?-) − (𝑥>)]} ¨ ∫!3!6 !3!5 𝑓 𝑥 (∘)𝑑𝑥 = lim /→@ ∑>3- >3/ 𝑓( !"#$?!" . ). 𝛿𝑥> = lim /→@ !':!& / . ∑>3- >3/ 𝑓( !"#$?!" . ) ¨ In the case where𝑓( !"#$?!" . ) = !"#$?!" . = (𝑘 + - . ). 𝛿𝑥 = (𝑘 + - . ). !':!& / ¨ ∫!3!6 !3!5 𝑥(∘)𝑑𝑥 = lim /→@ !':!& / . !':!& / . ∑>3- >3/ (𝑘 + - . ) = lim /→@ !':!& / . !':!& / . [ / /?- . + / . ] ¨ ∫!3!6 !3!5 𝑥(∘)𝑑𝑥 = (𝑥5 − 𝑥6). lim /→@ - / . - / . [ / /?- . + / . ] ¨ Using 𝑥6 = 0 without losing any generality, ¨ ∫!34 !3* 𝑥(∘)𝑑𝑥 = - . 𝑋., which is the usual result 94
  • 95. Luc_Faucheux_2020 The regular case revisited – III ¨ So in the regular case, the Riemann integral gives the same result, irrespective of where inside the small interval we pick the value of the function ¨ ∫!34 !3* 𝑥([)𝑑𝑥 = - . 𝑋. = ∫!34 !3* 𝑥(∘)𝑑𝑥 ¨ In essence, this is because the terms ∑>3- >3/ . 𝛿𝑥> scale like 1, and so any terms in ∑>3- >3/ . 𝛿𝑥> . will go to 0 in the limit 𝑁 → ∞ ¨ The “quadratic variation” ∑>3- >3/ . 𝛿𝑥> . does not add up to any finite number in the limit. ¨ In the case of a stochastic variable, the terms ∑>3- >3/ . 𝛿𝑥> scale like 0, because 𝛿𝑥> = ±𝛿𝑥, and the terms like ∑>3- >3/ . 𝛿𝑥> . will actually scale like time. ¨ The “quadratic variation” ∑>3- >3/ . 𝛿𝑥> . is said to scale linearly with time (some textbooks use the expression additive with time) and will not “disappear” to 0 in the limit. 95
  • 96. Luc_Faucheux_2020 Back to the worked-out example ¨ The Riemann sum is: 𝑆/ = ∑>3- >3/ 𝑋(𝑡>). [𝑋(𝑡>?-) − 𝑋(𝑡>)] ¨ We can expand: [𝑋(𝑡>?-) − 𝑋(𝑡>)].= 𝑋(𝑡>?-). + 𝑋(𝑡>). − 2. 𝑋(𝑡>). 𝑋(𝑡>?-) ¨ And: 𝑋(𝑡>). [𝑋(𝑡>?-) − 𝑋(𝑡>)] = 𝑋(𝑡>). 𝑋(𝑡>?-) − 𝑋(𝑡>). ¨ And : 𝑋(𝑡>). [𝑋(𝑡>?-) − 𝑋(𝑡>)] = - . . [𝑋(𝑡>?-). + 𝑋(𝑡>). − [𝑋(𝑡>?-) − 𝑋(𝑡>)].−2. 𝑋(𝑡>).] ¨ Or: 𝑋(𝑡>). [𝑋(𝑡>?-) − 𝑋(𝑡>)] = - . . [𝑋(𝑡>?-). − 𝑋(𝑡>).] − - . . [𝑋(𝑡>?-) − 𝑋(𝑡>)]. ¨ So: 𝑆/ = ∑>3- >3/ 𝑋(𝑡>). [𝑋(𝑡>?-) − 𝑋(𝑡>)] = - . . 𝑋(𝑡/?-). − - . ∑>3- >3/ [𝑋(𝑡>?-) − 𝑋(𝑡>)]. ¨ The first term is the expected *! . ¨ The second term would usually disappear when the variable X is regular, because the sum would scale as 𝑁. ( - / ).~ - / which will tend to 0 when 𝑁 → ∞ 96
  • 97. Luc_Faucheux_2020 Back to the worked-out example - II ¨ 𝑆/ = ∑>3- >3/ 𝑋(𝑡>). [𝑋(𝑡>?-) − 𝑋(𝑡>)] = - . . 𝑋(𝑡/?-). − - . ∑>3- >3/ [𝑋(𝑡>?-) − 𝑋(𝑡>)]. ¨ We define 𝑄/ = ∑>3- >3/ [𝑋(𝑡>?-) − 𝑋(𝑡>)]. ¨ We can use here the simple “binary” assumption that [𝑋(𝑡>?-) − 𝑋(𝑡>)].= (𝑡>?-−𝑡>) ¨ Note that this is capturing the essence of the matter ¨ A more accurate and general way to go about this would be to assume the following: ¨ {𝑋(𝑡>?-) − 𝑋(𝑡>)} follows the 𝑁(0, (𝑡>?-−𝑡>)) distribution ¨ We would then estimate the expected value of 𝑄/ ¨ 𝐸 𝑄/ = ∑>3- >3/ 𝐸[𝑋(𝑡>?-) − 𝑋(𝑡>)]. = ∑>3- >3/ (𝑡>?-−𝑡>) = 𝑡/?- ¨ We would have to show then that: lim /→@ 𝑄/ = 𝐸[𝑄/] ¨ This opens up the can of worms of how you define convergence (convergence in distribution, in probability, almost sure convergence, Lp-convergence). Will try to incorporate later, but trying to keep it simple to not hide the intuition behind notations 97
  • 98. Luc_Faucheux_2020 What did we get? ¨ ∫+34 +3= 𝑓 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) = lim /→@ {∑>3- >3/ 𝑓(𝑋(𝑡>)). [𝑋(𝑡>?-) − 𝑋(𝑡>)]} ¨ In the simple case 𝑓 𝑋(𝑡> = 𝑋(𝑡>) ¨ ∫+34 +3= 𝑋(𝑡). ([). 𝑑𝑋(𝑡) = - . 𝑋. − - . 𝑇 ¨ A couple of notes: ¨ Ito integral does NOT recover the usual rules of calculus ¨ Ito integral is called a martingale, meaning if X is a martingale (E[X]=0), then the Ito integral of a function f(X) is also a martingale ¨ 𝐸 𝑋 = 0 and 𝐸[∫+34 +3= 𝑋 𝑡 . [). 𝑑𝑋 𝑡 = 𝐸 - . 𝑋. − - . 𝑇 = 𝐸 - . 𝑋. − - . 𝑇 = 0 ¨ So 𝐸 𝑋. = 𝑇 ¨ We recover the usual variance of the Brownian motion 98
  • 99. Luc_Faucheux_2020 A couple more notes ¨ Let’s recall the relationship between the Ito and Stratonovitch integral (we can also work it out from the Riemann sum) ¨ ∫+3+6 +3+5 𝑓 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = ∫+3+6 +3+5 𝑓 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) + - . ∫+3+6 +3+5 𝑓′ 𝑋 𝑡 . 𝑑𝑡 ¨ Here, 𝑓 𝑋 = 𝑋 so 𝑓% 𝑋 = 1 ¨ ∫+34 +3= 𝑓 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = ∫+34 +3= 𝑓 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) + - . ∫+34 +3= 1. 𝑑𝑡 ¨ And we have: ∫+34 +3= 𝑋(𝑡). ([). 𝑑𝑋(𝑡) = - . 𝑋. − - . 𝑇 ¨ So ∫+34 +3= 𝑓 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = - . 𝑋. ¨ This is the usual rule of calculus ¨ The Stratonovitch integral preserves the usual rule of calculus 99
  • 100. Luc_Faucheux_2020 100 What does the usual exponential even mean in stochastic calculus? The ITO exponential
  • 101. Luc_Faucheux_2020 A couple more notes - II ¨ The Stratonovitch integral is NOT a martingale (this makes sense since taking the mid point introduces correlation, and thus the terms in the product are NOT independent) ¨ Because the Ito lemma does NOT recover the usual rules of calculus, what a function actually means has to be at times redefined. ¨ For example, we all know the exponential function 𝑓 𝑡 = exp(𝑡) ¨ It is such that : 𝑓% 𝑡 = 𝑓(𝑡) ¨ However we would not expect this function to preserve the same property when dealing with a stochastic variable (another way to say it is that this function is convex, and so the Jensen inequality will introduce a convexity correction, see the lecture on options) ¨ We know that Stratonovitch will preserve the rules of calculus ¨ So ∫+34 +3= 𝑒𝑥𝑝 𝑋 𝑡 . ∘ . 𝑑𝑋 𝑡 = exp(𝑋(𝑇)) ¨ And ∫+34 +3= 𝑓 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = ∫+34 +3= 𝑓 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) + - . ∫+34 +3= 𝑓′ 𝑋 𝑡 . 𝑑𝑡 101
  • 102. Luc_Faucheux_2020 The Ito exponential function ¨ ∫+34 +3= 𝑓 𝑋 𝑡 . (∘). 𝑑𝑋(𝑡) = 𝑒𝑥𝑝 𝑋 𝑇 = ∫+34 +3= 𝑓 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) + - . ∫+34 +3= 𝑓′ 𝑋 𝑡 . 𝑑𝑡 ¨ 𝑒𝑥𝑝 𝑋 𝑇 = ∫+34 +3= 𝑓 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) + - . ∫+34 +3= 𝑒𝑥𝑝 𝑋 𝑡 . 𝑑𝑡 ¨ Since the second term is always positive, we have ¨ ∫+34 +3= 𝑒𝑥𝑝 𝑋 𝑡 . ([). 𝑑𝑋(𝑡) <> 𝑒𝑥𝑝 𝑋 𝑇 ¨ So clearly under Ito, the usual exponential function does not verify 𝐹% 𝑋 = 𝐹(𝑋) ¨ For that reason, the convention is to rename some of the functions by specifying “Ito” in front of them ¨ This can be confusing at times 102
  • 103. Luc_Faucheux_2020 The Ito exponential function -II ¨ Ito lemma has : 𝛿𝐹 = )9 )* . 𝛿𝑋 + - . . )!9 )*! . 𝛿𝑋. + )9 )+ . 𝛿𝑡 ¨ With 𝐹 𝑋 = exp 𝑋 , 𝐹% 𝑋 = exp 𝑋 , 𝐹%% 𝑋 = exp(𝑋) ¨ We get 𝛿𝐹 = exp 𝑋 . 𝛿𝑋 + - . . exp 𝑋 . 𝛿𝑋. = exp(𝑋). 𝛿𝑋 + - . . exp(𝑋). 𝛿𝑡 ¨ With 𝐹 𝑋, 𝑡 = exp(𝑋 − + . ) , )9 )* = exp 𝑋 − + . , )!9 )*! = exp 𝑋 − + . , )9 )+ = :- . . exp(𝑋 − + . ) ¨ We get 𝛿𝐹 = exp 𝑋 − + . . 𝛿𝑋 + - . . exp 𝑋 − + . . 𝛿𝑋. − - . . exp 𝑋 − + . . 𝛿𝑡 ¨ Or: 𝛿𝐹 = exp 𝑋 − + . . 𝛿𝑋 = 𝐹 𝑋, 𝑡 . 𝛿𝑋 ¨ So you will see sometimes the function 𝐹 𝑋, 𝑡 = exp 𝑋 − + . being referred to as the Ito- exponential, whereas the regular exponential is of course 𝐹 𝑋 = exp 𝑋 103
  • 104. Luc_Faucheux_2020 The Ito exponential function -III ¨ So this is another indication that we should be careful with using regular functions and some of their properties. ¨ In regular calculus, the exponential function 𝑓 𝑥 = exp(𝑥) is such that : 𝑓% 𝑥 = 𝑓(𝑥) ¨ In ITO calculus, the “ITO exponential function” that still verifies: 𝐹% 𝑋(𝑡) = 𝐹(𝑋(𝑡)) is given by: 𝐹 𝑋(𝑡), 𝑡 = exp 𝑋(𝑡) − + . ¨ In STRATO calculus, the “STRATO exponential function” that still verifies: 𝐹% 𝑋(𝑡) = 𝐹(𝑋(𝑡)) is given by: 𝐹 𝑋(𝑡), 𝑡 = exp 𝑋(𝑡) ¨ Note that we are somehow lucky that those functions are still local (i.e. depends only on the value of 𝑋(𝑡) and 𝑡. It is not clear that we could not have ended up with a more complicated function that depends on the history or the path ∫+34 +3= 𝑋 𝑡 . 𝑑𝑡 for example. 104
  • 105. Luc_Faucheux_2020 A nice little summary Name of Integral RIEMANN ITO STRATONOVITCH Type Deterministic Stochastic Stochastic Points of integration Doesn’t matter LEFT MIDDLE Martingale NO YES NO Non-Anticipating NO YES NO Usual Calculus Rule YES NO YES(*) Main Usage Everywhere Finance Physics 105 ¨ (*) ONLY from a formal point of view. This is still a stochastic integral and a nasty beast not to be messed with “à la légère”
  • 106. Luc_Faucheux_2020 106 Back to trying to “integrate” X
  • 107. Luc_Faucheux_2020 Another example: can you integrate X ? ¨ Somewhat cultural side note: In French “to integrate” (or “integrer”) is the same word to perform a mathematical integration or to be accepted in a school. ¨ One of the most prestigious schools is called “Polytechnique” or “X” because of the logo on their hats of two crossed swords (it was created and is still technically a military school, a little like West Point here, but starts around the junior college level) ¨ So anyways, after high school, there are special schools called “classes preparatoires” where you just study for the entrance exam to those advanced schools (“hautes ecoles”), with names like X, Ecole Normale Superieure (Normal Superieure school) or others with names that are usually tied to their original concentration: Ponts et Chaussees (bridges and sidewalks), Mines (mining), Supelec (Superior Electricity), and so on ¨ Usually students spend around 2 years studying in “classes preparatoires” before taking the entrance exam (everything is an exam, there is no application, you take the exam, you get ranked and then the school offers you a spot in the order of ranking). ¨ So you get the idea, the question is how many years does it take you to integrate X ¨ 50 years later, people will still remember what “halfs” they were, or they will lie about it 107
  • 108. Luc_Faucheux_2020 Another example: can you integrate X ? - II ¨ If you are a genius and if it takes you only one year to integrate X, the student is called a ”one half” because: ∫*34 *3- 𝑋. 𝑑𝑋 = [ *! . ]*34 *3- = - . ¨ I have heard of “one-half” I have never met one. ¨ This is not the same as “half-bloods” from Harry Potter, I have never met one of those either ¨ If you are a decent student and if it takes you two years to integrate X, the student is called a ”three half” because: ∫*3- *3. 𝑋. 𝑑𝑋 = [ *! . ]*3- *3. = C . − - . = D . ¨ (I was a 3/2, but I did not integrate X, my father did, my brother did, I am the black sheep of the family) ¨ If you are a decent student and if it takes you three years to integrate X, the student is called a ”five half” because: ∫*3. *3D 𝑋. 𝑑𝑋 = [ *! . ]*3. *3D = E . − C . = F . ¨ If it takes you four years to integrate X, the student is called a ”seven half” because: ∫*3D *3C 𝑋. 𝑑𝑋 = [ *! . ]*3D *3C = -G . − E . = H . ¨ Being called a “seven half” is an insult 108
  • 109. Luc_Faucheux_2020 Another example: can you integrate X ? - III ¨ Oh also if you integrate 𝑋 not in 𝑑𝑋 but in 𝑑𝑡, you are called “endette”, or in debt ¨ OK, so that was a little digression ¨ Let’s get back to the matter at hands. ¨ If 𝑋 is not stochastic (also called deterministic), this is the usual calculus that we are used to, and so: ¨ ∫ 𝑋. 𝑑𝑋 = *! . + 𝐶 ¨ What happens if 𝑋 is now a stochastic variable? ¨ The truth is that I have never met anyone who can integrate a stochastic variable (I have also never met a “one half”), and so usually people always treat the problem from the other way, you start with a guess of what the integral is, you use ITO lemma and then you check in an iterative manner 109
  • 110. Luc_Faucheux_2020 Another example: can you integrate X ? - IV ¨ If we have a function 𝐹(𝑋), Ito lemma is the following: ¨ Ito (1951) guessed that you can write 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑧 as δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥, 𝑡 . 𝛿𝑧 ¨ If 𝐹(𝑥) a funcyon of x, the corresponding SDE as a result of “Taylor” expansion in ITO is: ¨ Bear in mind that this is really not a Taylor expansion, it is just ITO lemma that looks like a Taylor expansion where you kept some terms and not others ¨ 𝛿𝐹 = )9 )! . 𝛿𝑥 + - . . )!9 )!! . 𝑏.. 𝛿𝑡 ¨ 𝛿𝐹 = )9 )* . 𝛿𝑋 + - . . )!9 )*! . 𝛿𝑋. ¨ Let’s use 𝐹 𝑋 = 𝑋. as a good guess ¨ 𝐹 𝑋 = 𝑋., )9 )* = 2𝑋, )!9 )*! = 2, we then get: 𝛿 𝑋. = 2𝑋. 𝛿𝑋 + - . . 2. 𝛿𝑋. 110
  • 111. Luc_Faucheux_2020 Another example: can you integrate X ? - V ¨ 𝛿 𝑋. = 2𝑋. 𝛿𝑋 + - . . 2. 𝛿𝑋. ¨ In the case of the Geometric Brownian motion (Hull): 𝑑𝑋 = 𝜎𝑋𝑑𝑊 ¨ 𝛿 𝑋. = 2𝑋. 𝛿𝑋 + - . . 2. (𝜎𝑋). 𝛿𝑡 ¨ 𝛿 𝑋. = 2𝑋. 𝛿𝑋 + (𝜎𝑋). 𝛿𝑡 ¨ And so: ∫ 𝑋. 𝑑𝑋 = ∫{ ; *! . − - . . (𝜎𝑋). 𝛿𝑡} ¨ ∫ 𝑋. 𝑑𝑋 = [ *! . ] − - . ∫(𝜎𝑋). 𝛿𝑡 ¨ With the usual convention that 𝑋 𝑡 = 0 = 0, in the ITO convention: ¨ ∫+34 +3= 𝑋. 𝑑𝑋 = *(=)! . − - . ∫+34 +3= (𝜎𝑋). 𝛿𝑡 111
  • 112. Luc_Faucheux_2020 Another example: can you integrate X ? - VI ¨ Things are a little different in the Stratonovitch convention: ¨ Strato guessed that you can write 𝑑𝑥 = 𝑎 𝑥, 𝑡 . 𝑑𝑡 + 𝑏 𝑥, 𝑡 . 𝑑𝑊 as δ𝑥 = 𝑎 𝑥, 𝑡 . 𝛿𝑡 + 𝑏 𝑥 + 𝛿𝑥/2, 𝑡 . 𝛿𝑊 ¨ The SDE for 𝐹(𝑥) in Stratonovitch convention is (using what is known as ITO calculus): ¨ 𝛿𝐹 = )9 )! . 𝛿𝑥 + - . . )!9 )!! . 𝑏.. 𝛿𝑡 + - . . )9 )! . )5 )! . 𝑏. 𝛿𝑡 ¨ We know this is wrong, but let’s just illustrate how wrong it is: ¨ Rewriting it without the drift term we get: ¨ 𝛿𝑋 = 𝑏 𝑋 + ;* . , 𝑡 . 𝛿𝑊 = 𝑏 𝑋 . 𝛿𝑊 + - . . )5 )! . 𝛿𝑋. 𝛿𝑊 ¨ 𝛿𝑋 = 𝑏 𝑋 + ;* . , 𝑡 . 𝛿𝑊 = 𝑏 𝑋 . 𝛿𝑊 + - . . )5 )! . 𝑏 𝑋 . 𝛿𝑡 ¨ 𝛿𝑋 = 𝑏 𝑋 . 𝛿𝑊 + - . . )5 )! . 𝑏 𝑋 . 𝛿𝑡 in Stratonovitch while Ito had: 𝛿𝑋 = 𝑏 𝑋 . 𝛿𝑊 112
  • 113. Luc_Faucheux_2020 Another example: can you integrate X ? - VII ¨ Redoing it wrong just to convince ourselves of how wrong it is: ¨ 𝛿𝑋 = 𝑏 𝑋 ∘ 𝛿𝑊 = 𝑏 𝑋 + ;* . , 𝑡 . 𝛿𝑊 = 𝑏 𝑋 . 𝛿𝑊 + - . . )5 )! . 𝑏 𝑋 . 𝛿𝑡 ¨ 𝛿𝐹 = )9 )* . 𝛿𝑋 + - . . )!9 )*! . 𝑏.. 𝛿𝑡 + - . . )9 )* . )5 )* . 𝑏. 𝛿𝑡 ¨ 𝛿𝐹 = )9 )* . 𝛿𝑋 + - . . )!9 )*! . 𝛿𝑋. ¨ 𝛿𝐹 = )9 )* . 𝑏 𝑋 + ;* . . 𝛿𝑊 + - . . )!9 )*! . 𝑏 𝑋 + ;* . , 𝑡 . . 𝛿𝑡 ¨ 𝛿𝐹 = )9 )* . 𝑏 𝑋 . 𝛿𝑊 + )9 )* . )5 )* . ;* . . 𝛿𝑊 + - . . )!9 )*! . 𝑏 𝑋 + ;* . , 𝑡 . . 𝛿𝑡 113
  • 114. Luc_Faucheux_2020 Another example: can you integrate X ? - IX ¨ We get: ¨ 𝛿𝐹 = )9 )* . 𝑏 𝑋 . 𝛿𝑊 + )9 )* . )5 )* . ;* . . 𝛿𝑊 + - . . )!9 )*! . 𝑏 𝑋 + ;* . . . 𝛿𝑡 ¨ And : ¨ 𝑏 𝑋 + ;* . . = 𝑏 𝑋 . + 2. 𝑏 𝑋 . )5 )* . ;* . ¨ 𝛿𝐹 = )9 )* . 𝑏 𝑋 . 𝛿𝑊 + )9 )* . )5 )* . ;* . . 𝛿𝑊 + - . . )!9 )*! . 𝑏 𝑋 .. 𝛿𝑡 + - . . )!9 )*! . 2𝑏 𝑋 . )5 )* . ;* . 𝛿𝑡 ¨ Keeping only the terms in order 1 in 𝛿𝑍 and 𝛿𝑡 ¨ 𝛿𝐹 = )9 )* . 𝑏 𝑋 . 𝛿𝑊 + - . . )!9 )*! . 𝑏 𝑋 .. 𝛿𝑡 + - . . )9 )* . )5 )* . 𝑏. 𝛿𝑡 114
  • 115. Luc_Faucheux_2020 Another example: can you integrate X ? - X ¨ 𝛿𝐹 = )9 )* . 𝑏 𝑋 . 𝛿𝑊 + - . . )!9 )*! . 𝑏 𝑋 .. 𝛿𝑡 + - . . )9 )* . )5 )* . 𝑏. 𝛿𝑡 ¨ In the case of the simple Geometric Brownian motion (Hull): 𝑑𝑋 = 𝜎𝑋𝑑𝑊 ¨ Using the guess 𝐹 𝑋 = 𝑋. ¨ 𝛿(𝑋.) = 2𝑋. 𝜎𝑋. 𝛿𝑊 + - . . 2. (𝜎𝑋). 𝛿𝑡𝛿𝑡 + - . . 2𝑋. 𝜎. 𝜎𝑋. 𝛿𝑡 ¨ 𝛿(𝑋.) = 2𝑋. 𝛿𝑋 + 2. (𝜎𝑋).. 𝛿𝑡 ¨ In Ito we had : 𝛿 𝑋. = 2𝑋. 𝛿𝑋 + - . . 2. (𝜎𝑋).. 𝛿𝑡 ¨ So using ITO: ∫+34 +3= 𝑋. 𝑑𝑋 = *(=)! . − - . ∫+34 +3= (𝜎𝑋).. 𝛿𝑡 ¨ Using Stratonovitch: ∫+34 +3= 𝑋. 𝑑𝑋 = *(=)! . − ∫+34 +3= (𝜎𝑋).. 𝛿𝑡 115
  • 116. Luc_Faucheux_2020 Another example: can you integrate X ? - Xa ¨ If we were to look at the simple Brownian motion case: ¨ 𝑑𝑋 = 𝑑𝑊 ¨ ITO: 𝛿 𝑋. = 2𝑋. 𝛿𝑋 + - . . 2. 𝛿𝑋. ¨ 𝛿 𝑋. = 2𝑋. 𝛿𝑋 + 𝛿𝑡 ¨ And so: ∫ 𝑋. 𝑑𝑋 = ∫{ ; *! . − - . . 𝛿𝑡} ¨ ∫ 𝑋. 𝑑𝑋 = [ *! . ] − - . ∫ 𝛿𝑡 ¨ With the usual convention that 𝑋 𝑡 = 0 = 0, in the ITO convention: ¨ ∫+34 +3= 𝑋. 𝑑𝑋 = *(=)! . − - . 𝑇 116
  • 117. Luc_Faucheux_2020 Another example: can you integrate X ? - Xb ¨ ∫+34 +3= 𝑋. 𝑑𝑋 = *(=)! . − - . 𝑇 ¨ Note that we know this to be consistent since the ITO integral is a martingale (or a trading strategy) ¨ We will go over this again at more length but we have: ¨ 𝔼 ∫+34 +3= 𝑋. 𝑑𝑋 = 0 = 𝔼 * = ! . − 𝔼 - . 𝑇 = 𝔼 * = ! . − - . 𝑇 ¨ And we recover the usual dispersion expectation for the simple Brownian motion: ¨ 𝔼 𝑋 𝑇 . = 𝑇 117
  • 118. Luc_Faucheux_2020 Another example: can you integrate X ? - Xc ¨ If we were to look at the simple Brownian motion case: ¨ 𝑑𝑋 = 𝑑𝑊 𝑏 𝑋 = 1 so )5 )* = 0 ¨ STRATO: 𝛿𝐹 = )9 )* . 𝑏 𝑋 . 𝛿𝑊 + - . . )!9 )*! . 𝑏 𝑋 .. 𝛿𝑡 + - . . )9 )* . )5 )* . 𝑏. 𝛿𝑡 ¨ 𝛿 𝑋. = 2𝑋. 𝛿𝑋 + 𝛿𝑡 This is exactly the same as ITO ¨ So we get the strange result that in BOTH ITO and STRATO we will recover for the simple Brownian motion: ¨ ∫+34 +3= 𝑋. 𝑑𝑋 = *(=)! . − - . 𝑇 ¨ This is clearly wrong, and once again the error was in “using a STRATO convention in the ITO calculus”, and not even that, as we took “ITO calculus” as “doing regular Taylor expansions like in regular calculus and just keeping some terms and neglecting some other terms” 118
  • 119. Luc_Faucheux_2020 Another example: can you integrate X ? - XI ¨ HOWEVER, remember the formula in the simple Wiener case : 𝑋 = 𝑊 ¨ 𝑆𝑇𝑅𝐴𝑇𝑂(𝑓) = 𝐼𝑇𝑂 𝑓 + - . . 𝑅𝐼𝐸𝑀𝐴𝑁𝑁(𝑓%) ¨ In the case of 𝑓 = 𝑋, 𝑓% = 1 ¨ 𝑆𝑇𝑅𝐴𝑇𝑂 𝑋 = 𝐼𝑇𝑂 𝑋 + - . . 𝑅𝐼𝐸𝑀𝐴𝑁𝑁(1) with ¨ 𝑅𝐼𝐸𝑀𝐴𝑁𝑁 1 = ∫+34 +3= 1. 𝑑𝑡 = 𝑇 ¨ So Ito : ∫+34 +3= 𝑋. 𝑑𝑋 = *(=)! . − - . ∫+34 +3= 1. 𝛿𝑡 or ∫+34 +3= 𝑋. 𝑑𝑋 = *(=)! . − - . 𝑇 ¨ Stratonovitch: ∫+34 +3= 𝑋 ∘ 𝑑𝑋 = ∫+34 +3= 𝑋. 𝑑𝑋 + - . 𝑇 = *(=)! . − - . 𝑇 + - . 𝑇 = *(=)! . ¨ Startonovitch as expected follows in a formal manner the usual rules of calculus 119
  • 120. Luc_Faucheux_2020 Another example: can you integrate X ? - XII ¨ So clearly we failed to integrate X, because we found the following wrong results: ¨ In the Geometric Brownian motion case, Strato does not match formally the usual rules of calculus ¨ In the simple Brownian motion case, BOTH ITO and STRATO returns the same result ¨ This is not possible as we know that the ITO and STRATO integrals are different and we know what the difference is ( - . . 𝑅𝐼𝐸𝑀𝐴𝑁𝑁) ¨ But again, apologies if so far the mistake was obvious, but NEVER rely on usual rules of calculus, or think that Stratanovitch is just “taking the middle point” and that you can still use ITO lemma or Taylor expansion in a nested manner. If you choose STRATO you CANNOT use ITO calculus, you have to stay in STRATO calculus, and vice versa ¨ Also note that ITO lemma is very generic in (𝛿𝑋). but when choosing a specific expression for𝑋(𝑡) as a function of the Brownian motion (Wiener process 𝑊(𝑡)), we end up with very different expressions, some tractable, some not so much 120
  • 121. Luc_Faucheux_2020 Another example: can you integrate X ? - XIII ¨ The right way to do it is of course: ¨ ITO: 𝛿𝑓 = )$ )* . ([). 𝛿𝑋 + - . . )!9 )!! . ([). 𝛿𝑋. ¨ The results we obtained in the preceding slides for ITO are still correct ¨ STRATO: 𝛿𝑓 = )$ )* . ∘ . 𝛿𝑋 ¨ And then we do get in the case of the simple Brownian motion: 𝑑𝑋 = 𝑑𝑊 ¨ 𝐹 𝑋 = 𝑋. ¨ 𝛿𝑓 = 2𝑋. ∘ . 𝛿𝑊 ¨ ∫+34 +3= 𝑋. ∘ . 𝑑𝑋 = - . ∫+34 +3= 2𝑋. ∘ . 𝛿𝑊 = - . ∫+34 +3= 𝑑 𝑋. = [ *! . ]+34 +3= ¨ As expected if STRATO follows formally the usual rules of calculus 121
  • 122. Luc_Faucheux_2020 Quick notes ¨ Usually when dealing with an SDE, people will try ¨ 1) get back to an SDE where the stochastic scaling does not depend on the stochastic variable. For example, when dealing with 𝑑𝑆 = 𝜇. 𝑆. 𝑑𝑡 + 𝜎. 𝑆. 𝑑𝑊, the natural road to explore is writing something like I< < = 𝜇. 𝑑𝑡 + 𝜎. 𝑑𝑊, and then try to ”define” what I< < is. ¨ It seems intuitive that I< < should have something to do with 𝑑(ln 𝑆 ), so we would love to write something like 𝑑(𝑙𝑛 𝑆 ) = 𝜇. 𝑑𝑡 + 𝜎. 𝑑𝑊, but it turns out that is not quite right, because 𝑙𝑛 𝑆 is convex as a function of 𝑆, and 𝑆 is stochastic, not deterministic ¨ So always re-derive ITO lemma to make sure you are not dropping terms in the “Taylor expansion” ¨ 2) try to get rid of the term in from of the 𝑑𝑧, because then you are not dealing with a SDE anymore, but with a PDE. This is what Black Sholes did by adding a “delta hedge”, and off to a Nobel prize they went 122
  • 123. Luc_Faucheux_2020 Quick notes - II ¨ We also see the famous “convexity adjustment” that we looked at in the Options deck. ¨ 𝑙𝑛 𝑆 is negatively convex as a function of the stochastic variable ¨ So from an intuition point of view as we saw ¨ < 𝑙𝑛 𝑆 > ≠ 𝑙𝑛 < 𝑆 > ¨ Actually we know that : < 𝑙𝑛 𝑆 > = 𝔼(𝑙𝑛 𝑆 ) ≤ 𝑙𝑛 𝔼(𝑆) = 𝑙𝑛 < 𝑆 > ¨ So it would make sense that around the point 𝔼(𝑆) the function 𝑙𝑛 𝑆 would not behave as the “regular” function 𝑙𝑛 𝔼(𝑆) ¨ Note that the distinction though, the convexity adjustment was coming from integrating over the possible outcomes at a given time of the stochastic variable. Here we are integrating over the time (over the path over time of the stochastic variable). They are related nonetheless. ¨ Obviously if you have the expression that resulted from integrating over the path, you can now use it to integrate over the distribution 123
  • 124. Luc_Faucheux_2020 Quick notes - III ¨ Just to make it explicit. ¨ ∫+3+& +3+' 𝑑𝐹(𝑋 𝑡 ) = ∫+3+6 +3+5 )9 )* . ([). 𝑑𝑋(𝑡) + - . ∫+3+6 +3+5 )!9 )!! 𝑋 𝑡 . (𝑑𝑋). ¨ 𝛿𝐹 = )9 )* . 𝛿𝑋 + - . . )!9 )!! . (𝛿𝑋)., which is the celebrated Ito lemma (chain rule) ¨ 𝐹 𝑋 = ln(𝑋), )9 )* = 1/𝑋, )!9 )*! = −1/𝑋. ¨ In the case of the simple Brownian motion: 𝑑𝑋 = 𝑑𝑊 ¨ 𝛿(𝑙𝑛𝑊) = - 1 . 𝛿𝑊 − - . . - 1! . 𝛿𝑡 this is quite ugly to deal with ¨ In the case of the Geometric Brownian motion (Hull) 𝑑𝑋 = 𝑋. 𝑑𝑊 ¨ 𝛿 𝑙𝑛𝑋 = - * . 𝑋. 𝛿𝑊 − - . . - *! . 𝑋.. 𝛿𝑡 = 𝛿𝑊 − - . . 𝛿𝑡 this is quite nice and easy to deal with 124
  • 125. Luc_Faucheux_2020 Quick notes - IV ¨ So for the log function and in the case of a geometric Brownian motion, we have ¨ 𝛿 𝑙𝑛𝑋 = 𝛿𝑊 − - . . 𝛿𝑡 ¨ Or more rigorously: ¨ ∫+3+& +3+' 𝑑𝐿𝑁(𝑋 𝑡 ) = 𝐿𝑛 𝑋 𝑡5 − 𝐿𝑛(𝑋 𝑡6 ) = ¨ 𝐿𝑛 𝑋 𝑡5 − 𝐿𝑛 𝑋 𝑡6 = ∫+3+6 +3+5 𝑑𝑊 𝑡 − - . ∫+3+6 +3+5 𝑑𝑡 = 𝑊 𝑡5 − 𝑊 𝑡6 − - . (𝑡5 − 𝑡6) ¨ One could be tempted to identify - . . 𝛿𝑡 as a convexity adjustment (which it is in some way, since if the function was not convex, i.e. )!9 )*! = 0, this term would not appear), but it is not exactly the one we are dealing with in the option deck, as that one also depends on the specific distribution being used. 125
  • 126. Luc_Faucheux_2020 Quick notes - V ¨ Still rather crudely (but rigorous, as we will show in section V on the GBM) ¨ 𝑑𝑋 = 𝑋. 𝑑𝑊 careful, this is not the same as writing 𝑋 𝑡 = exp(𝑊 𝑡 ) ¨ 𝔼 𝑋 𝑡5 = 𝑋 𝑡6 ¨ 𝐿𝑛 𝑋 𝑡5 − 𝐿𝑛 𝑋 𝑡6 = ∫+3+6 +3+5 𝑑𝑊 𝑡 − - . ∫+3+6 +3+5 𝑑𝑡 = 𝑊 𝑡5 − 𝑊 𝑡6 − - . (𝑡5 − 𝑡6) ¨ 𝔼 𝐿𝑛 𝑋 𝑡5 − 𝐿𝑛 𝑋 𝑡6 = 𝔼{𝑊 𝑡5 − 𝑊 𝑡6 − - . (𝑡5 − 𝑡6)} ¨ 𝔼 𝐿𝑛 𝑋 𝑡5 − 𝐿𝑛 𝑋 𝑡6 = 𝔼 − - . 𝑡5 − 𝑡6 = − - . (𝑡5 − 𝑡6) ¨ 𝔼 𝐿𝑛 𝑋 𝑡5 = 𝔼 𝐿𝑛 𝑋 𝑡6 − - . 𝑡5 − 𝑡6 = 𝐿𝑛 𝑋 𝑡6 − - . (𝑡5 − 𝑡6) ¨ 𝔼 𝐿𝑛 𝑋 𝑡5 = 𝐿𝑛 𝔼 𝑋 𝑡5 − - . (𝑡5 − 𝑡6) ¨ This is indeed the convexity adjustment (again makes sense, it is the second order) 126
  • 127. Luc_Faucheux_2020 Quick notes - VI ¨ Just to make it explicit. ¨ ∫+3+& +3+' 𝑑𝐹(𝑋 𝑡 ) = ∫+3+6 +3+5 )9 )* . ([). 𝑑𝑋(𝑡) + - . ∫+3+6 +3+5 )!9 )!! 𝑋 𝑡 . (𝑑𝑋). ¨ 𝛿𝐹 = )9 )* . 𝛿𝑋 + - . . )!9 )!! . (𝛿𝑋)., which is the celebrated Ito lemma (chain rule) ¨ 𝐹 𝑋 = 𝑋., )9 )* = 2𝑋, )!9 )*! = 2 ¨ In the case of the simple Brownian motion: 𝑑𝑋 = 𝑑𝑊 ¨ 𝛿(𝑊.) = 2𝑊. 𝛿𝑊 + - . . 2. 𝛿𝑡 ¨ In the case of the Geometric Brownian motion (Hull) 𝑑𝑋 = 𝑋. 𝑑𝑊 ¨ 𝛿 𝑋. = 2𝑋. 𝑋. 𝛿𝑊 + - . . 2. 𝑋.. 𝛿𝑡 = 2 𝑋. 𝛿𝑊 + (𝑋.). 𝛿𝑡 127
  • 128. Luc_Faucheux_2020 Quick notes - VII ¨ So for the square function and in the case of a simple Brownian motion, we have ¨ 𝛿(𝑊.) = 2𝑊. 𝛿𝑊 + - . . 2. 𝛿𝑡 ¨ Or more rigorously: ¨ ∫+3+& +3+' 𝑑𝑊. = 𝑊.(𝑡5) − 𝑊.(𝑡6) = ¨ 𝑊. 𝑡5 − 𝑊. 𝑡6 = ∫+3+6 +3+5 2𝑊 𝑡 . [ . 𝑑𝑊 𝑡 + ∫+3+6 +3+5 𝑑𝑡 ¨ 𝑊. 𝑡5 − 𝑊. 𝑡6 = ∫+3+6 +3+5 2𝑊 𝑡 . ([). 𝑑𝑊 𝑡 + (𝑡5 − 𝑡6) 128
  • 129. Luc_Faucheux_2020 Quick notes - VIII ¨ Still rather crudely (but rigorous, as we will show in section V on the GBM) ¨ 𝑑𝑋 = 𝑑𝑊 ¨ 𝔼 𝑊 𝑡5 = 𝑊 𝑡6 ¨ 𝑊. 𝑡5 − 𝑊. 𝑡6 = ∫+3+6 +3+5 2𝑊 𝑡 . ([). 𝑑𝑊 𝑡 + 𝑡5 − 𝑡6 ¨ 𝔼 𝑊. 𝑡5 − 𝑊. 𝑡6 = 𝔼{∫+3+6 +3+5 2𝑊 𝑡 . ([). 𝑑𝑊 𝑡 + 𝑡5 − 𝑡6 } ¨ 𝔼 𝑊. 𝑡5 − 𝑊. 𝑡6 = 𝔼 𝑡5 − 𝑡6 = (𝑡5 − 𝑡6) ¨ 𝔼 𝑊. 𝑡5 = 𝔼 𝑊. 𝑡6 + 𝑡5 − 𝑡6 = 𝑊. 𝑡6 + (𝑡5 − 𝑡6) ¨ 𝔼 𝑊. 𝑡5 = 𝔼 𝑊 𝑡5 . + (𝑡5 − 𝑡6) ¨ This is indeed the convexity adjustment (again makes sense, it is the second order) ¨ Since the square function is positively convex we would expect the convexity adjustment to be positive 129