TEACHING NOTE 96-05: ITÔ’S LEMMA AND STOCHASTIC INTEGRATION

Version date: July 18, 2008 C:\Classes\Teaching Notes\TN96-05.doc

One of the most important mathematical results used in finance is Itô’s Lemma. Though this result was found around 1950, it did not make its way into the financial models until 1973 when Black and Scholes discovered that it could be used to help find the price of an option. Although there is a great deal of formal mathematic rigor to Itô’s Lemma, the essential elements are quite simple. Let us begin, however, with a few basic statements from ordinary calculus. Recall that any differential in ordinary calculus is considered to have a limit of zero if raised to a power greater than 1.0. In other words, dtk → 0 if k > 1. Consider any well-behaved mathematical function like F(X,t). 1 Using a Taylor series expansion, the change in value of the function can be expressed as dF = ∂F ∂F 1 ∂2F 1 ∂2F 2 ∂2F 2 dX + dt + dX + dt + dXdt + ∂X ∂t 2 ∂X 2 2 ∂t 2 ∂X∂t

Because dX2 → 0, dt2 → 0 and dXdt → 0, we write this as

dF =

∂F ∂F dX + dt , ∂X ∂t

It means that the change in F is a function of the changes in X and t. The change in X is multiplied by the partial derivative of F with respect to X and the change in t is multiplied by the partial derivative of F with respect to t. This is a formal way of stating that as X and t change, they induce a change in F. The changes in X and t are so small, however, that squared changes in X and t are zero in the limit and that their product is also zero in the limit. In ordinary calculus, all variables are non-stochastic. This simply means that when we talk about a particular value of X, that value is known for certain. When X is stochastic, we leave the world of ordinary calculus and enter the world of stochastic calculus. There we cannot talk about a particular value of X; we must talk about a domain of possible values of X. We often summarize information like that in terms of expected values and variances. In the stochastic calculus, results are proven by demonstrating what happens when squared values of a
1

By well-behaved, we mean that its first and second derivatives exist. D. M. Chance, TN96-05 1

Itô Lemma and Stochastic Integration

because dWtdt goes to zero in the limit. Recall that dWt = ε t dt . 4 2 See the previous footnote. which can be reduced to dX2 = σ(X. It describes the stochastic process of a function F(X. A result is said to hold in “mean square limit. as we noted in previous notes. t )dWt . With the expressions μ(X. such as dWt or a more generalized process such as dX = μ(X. 2 ∂X ∂t 2 ∂X 2 2 Furthermore.t)2dt. 3 From the definition of dWt. t )dtdWt . 5 Note that the uncertainty in F comes from the uncertainty in Wt. which is the mean squared deviation around the expected value. the expression dWtdt will be εtdt3/2. A Let us now propose that X is stochastic and follows an Itô process. dX2 is not zero because. 2 ∂X ∂t 2 ∂X This is known as Itô’s Lemma. which is driven by an Itô process X and time t. t ) dt + μ(X. While dt2 is still zero because time is not stochastic. t ) ⎟ dF = ⎜ X . 3 This gives us dF = ∂F ∂F 1 ∂2F dX + dt + dX 2 . Chance. being named for the Japanese mathematician who discovered it. The 3/2 power on dt drives it to zero in the limit. 2 ⎜ ∂t ∂X ⎟ 2 ∂X ∂X ⎝ ⎠ In this manner we see that the term in parentheses is the expected change in F and the variance is given as (∂F / ∂X )σ(X.t). t dt + σ(X. t ) dt 2 + σ(X. t )dt + σ(X. Note that by substituting for dX and dX2 we could write the stochastic process for F as ⎛ ∂F ∂F 1 ∂2F ∂F 2⎞ ( ) + μ + σ(X. t ) .t). M. t )dt + σ(X. It is approximately correct to say that a result in stochastic calculus holds when the variance converges to a finite value. t ) dt .variable are multiplied by probabilities. t )dt 2 + σ(X. t )dWt . t )dWt dt = 0 . The term “mean square limit” can be thought of somewhat like the concept of variance. we are allowing the expectation and variance of X to be functions of the level of X and time t. Note that dXdt is zero in the limit because dXdt = (μ(X.t) and look at the Taylor series expansion when X is stochastic. D. t )dWt )dt = μ(X.” 2 more formal statement of this concept is presented in later pages of this note. dWt2 = dt. Now suppose that we go back to the unspecified function F(X.t) and σ(X. 4 This gives us dF = ∂F ∂F 1 ∂2F 2 dX + dt + σ(X. we can state that dX 2 = μ(X. TN96-05 2 Itô Lemma and Stochastic Integration . t )σ(X.

This process is called Riemann integration. which is a requirement for a derivative to exist. to integrate a stochastic function by defining the process somewhat differently. The latter is done by dividing the area under a curve into rectangles. t )dt + σ(X (t ). The variance is simply the square of whatever term is multiplied by dWt times the variance of dWt. ∂X t ∂t 2 ∂X 2 t Now suppose we integrate over the period time 0 to time t.Because we often need to price derivative contracts. It is possible. we obtain dF(X t . t )dWt . In addition Itô’s Lemma can be expressed in integral form. the derivative must be defined for all values of t. the zig-zaggedness of Xt renders it impossible for the slope of a tangent line at any point to have a finite limit. which is dt. where we have subscripted t on X to reinforce that X takes on different values at different times t. In a stochastic function. which follows the stochastic process: dX t = μ(X (t ). hence the expected value of dF comes from the first term on the right-hand side. Alternatively. Suppose F is the price of an option or other derivative contract on an asset whose value is X. t ) 1 ∂ 2 F(X t . TN96-05 Itô Lemma and Stochastic Integration 3 5 . however. The price of a derivative is said to be “derived” from the price of the underlying asset and time. Instead of being the limit of the area under each rectangle under the curve. The variance in the stochastic process comes from the second term on the right-hand side because the first term is the expected value. If we let that asset price evolve according to the Itô process. M. Thus. As the number of rectangles goes to infinity. a derivative such as dXt/dt is not defined. Consequently. meaning somewhat loosely that the integral is the expected value of the Remember that the expected value of dWt is zero. Let us restate the problem. depending on the evolution of Wt through time. a stochastic integral is defined as the mean square limit. or its variation Stieljes integration. and is not defined the same way as standard or non-stochastic integration. then we can be assured that the change in the option price is described by Itô’s Lemma. D. F(Xt. t ) = ∂F(X t . Chance. however. To put it somewhat crudely.t). the sum of the areas of the rectangles precisely approaches the area under the curve.t) is a convenient specification of a derivative price. t ) ∂F(X t . This process is called stochastic integration. standard integration can be viewed as adding up all of the infinitesimally small values dX/dt across values of t. We are given a random variable Xt. Itô’s Lemma is widely used in finance. F(X. The area under the curve is approximately the sum of the areas of each of the rectangles. multiplying by dt scales it by the length of the time interval. t ) 2 dX t + dt + dX t . Applying Itô’s Lemma to the function. as stated in the above equation. which is a constant. The term in parentheses is the expected value.

k )[X k − X k −1 ]⎟ . 2 2 ∂X μ 0 t This looks like it is going to be a problem because of the term dXu2. μ )dX μ ⎟ ⎟ = 0. n →∞ 0 ⎠ ⎝ k =1 2 Conveniently. by definition t ∫ dX 0 μ = Xt − X0 . Using that result and combining terms gives t t ⎡ ∂F(X μ . μ ) 2⎤ ( ) F(X t . t ) − F(X 0 . some of the properties of ordinary integration hold in stochastic integration. In other words. Now let us write Itô’s Lemma in its integral form. d σ μ μ + dX μ .t) = σ for all t. k )[X k − X k −1 ] − ∫ σ(X μ . the sum of the changes in X from X0 to Xt is. but we have previously noted that dXt2 = σ(Xt. For example. μ ) ∂μ 2 1 ∂ F(X μ . Chance. σ(Xt.t)2dt. Such a limit will exist for the processes we typically encounter in finance. M. μ ) 2 dμ + ∫ dX μ . Using stochastic integration.e.. Xt . by definition. μ ) = ∫ μ t t ∂F(X μ . In the special case where the volatility is constant. μ ) ∂F(X μ . More formally the stochastic integral known as the Itô integral. whether X is stochastic or not. μ )dX μ 0 μ .sum of the squared product of the volatility times the change in the stochastic variable. n→∞ ⎝ k =1 ⎠ where t ⎞ ⎛ n ⎜ lim E⎜ ∑ σ(X k −1 . t ∫ σ(X . we have ∫ dF(X . μ ) ∂X μ 0 0 dX μ + ∫ 0 t ∂F(X μ . In this sense a stochastic integral is much more like a volatility measure. we can pull the constant out and obtain t ∫ σdX 0 μ = σ[X t − X 0 ] . TN96-05 4 Itô Lemma and Stochastic Integration .X0. ⎥ μ 2 ∫ ∂ μ 2 X ∂ X ∂ ⎥ μ μ 0 ⎢ 0 ⎣ ⎦ D.0) = ∫ ⎢ + X . is ⎛ n ⎞ lim E⎜ ∑ σ(X k −1 . i. μ ) 1 ∂ 2 F(X μ .

Now consider another process.which is Itô’s Lemma in integral formula. = X. Remember that either the differential or integral version of Itô’s Lemma automatically implies that the other exists. we obtain the partial derivatives easily: ∂Z ∂Z = Y. 6 There is no term related to dt since Z is not directly determined by t. we do not have to specify those processes precisely. We want to identify the stochastic process for Z. so either can be used. + + 2 ∂X 2 2 ∂Y 2 ∂X ∂Y ∂t ∂X∂Y Depending on the specifications of dX and dY. Z. Let us look at two more general cases of Itô’s Lemma. For right now. we obtain: dZ = YdX + XdY + σ X σ Y dt . or sometimes called Itô’s stochastic integral. M. both with different but constant parameters. we can usually proceed to simplify this further. Chance. D.t). ∂X ∂Y ∂2Z ∂2Z = = 0 . one is preferred over the other. Now consider two specific processes. dY = μ Y dt + σ Y dW. In other words let us say that X and Y follow stochastic differential equations driven by an Itô process. Then applying Itô’s Lemma to F(X. ∂Y 2 ∂X∂Y ∂ 2Z = 0. Note that there is no interaction term with dX and dt or with dY and dt as these will go to zero by the product of dX and dW or dY and dW.Y. X and Y. we obtain dF = 1 ∂2F 1 ∂2F ∂F ∂F ∂F ∂2F 2 2 dX + dY + dt + dX dY dXdY . ∂X ⎝ ∂Y ⎠ ∂Y ⎝ ∂X ⎠ Substituting these results into the above. TN96-05 5 Itô Lemma and Stochastic Integration . Here both X and Y are driven by the same Brownian motion W. Y and t. Consider a function F driven by X. dX = μ x dt + σ x dW. Applying Itô’s Lemma to Z we obtain 6 : 1 ∂2Z 1 ∂2Z ∂Z ∂Z ∂2Z 2 2 dZ = dX + dY + dX + dY + dXdY . 2 ∂X 2 2 ∂Y 2 ∂X ∂Y ∂X∂Y Since Z is a very simple function of X and Y. First we look at the case where our random process is a function of two random processes. and in some cases. ∂X 2 ∂ ⎛ ∂Z ⎞ ∂ ⎛ ∂Z ⎞ ⎜ ⎟= ⎜ ⎟ = 1. defined as Z = XY.

Now let us assume the two processes are driven by different Brownian motions: dX = μ x dt + σ x dWX . for which a closed-form solution sometimes exists and otherwise. Usually a differential equation is restricted by certain conditions that define its value at certain combinations of the variables. Of course if these are independent Brownian motions. D. this leads to an ordinary. specifically defined as: dWX = ε X dt dWY = ε Y dt . We have Var(dWxdWy) = 0. These are called boundary conditions. So dZ = YdX + XdY + σxσyρxydt. such as A numerical solution is a solution to a differential equation that is obtained by solving the differential equation over a range of possible values of the variables. We have dXdY = σxσydWxdWy. Let us examine the expression dWxdWy. dY = μ Y dt + σ Y dWY . non-stochastic partial differential equation. 7 References We know that the two Weiner increments are more The references for TN96-04. E(dWxdWy) = dWxdWy = ρxydt. Now let us take the variance of their product. Modeling Asset Prices as Stochastic Processes I. Good mathematical treatments are found in many places. This was result is presented in TN00-05. When that is accomplished. TN96-05 Itô Lemma and Stochastic Integration 6 7 . The trick is to remove the uncertainty by forming a riskless hedge consisting of units of the derivative and units of the asset. Going back to what we originally wanted. the covariance term disappears. dXdY = σxσyρxydt. M. These results provide only the basic tools that we shall use for pricing derivatives.This result reflects the fact that dXdY = σxσydt. Let ρxy be the correlation between the Wx and Wy. all generally apply here as well. This is obviously the covariance between X and Y times dt. a numerical solution can be obtained. Chance.

S. Chs. Brock. Malliaris. 2. Cambridge: Cambridge University Press (1996). 637-654. I. Ch. H. Options. Scholes. M. For a fairly readable look from the finance perspective. 3. Black. C. A. Eckardt. TN96-05 7 Itô Lemma and Stochastic Integration . and W. Brownian Motion and Stochastic Calculus. and M. E. Chs. San Diego: Academic Press (2000).” Financial Review 16 (Spring.: John Wiley and Sons (1998). G. Shreve. see Aucamp.K. Futures and Other Derivatives. Wilmott. J. Rennie. Stochastic Methods in Economics and Finance Amsterdam: North-Holland (1982). 5th ed. Ch. and J. de Varenne. D. F. 1981). 2nd. Ch. D. Nielsen. 5. The Mathematics of Financial Derivatives. Bellalah. Cambridge: Cambridge University Press (1995). 41-50. F. 2nd ed. L.. E. “The Pricing of Options and Corporate Liabilities. C.Karatzas. Upper Saddle River. Pricing and Hedging of Derivative Securities. Chs. Jr. M. DeWynne. L. Futures and Exotic Derivatives. Ch.: Oxford University Press (1999). Chance. M. Options.. 3.K. Oxford. and S. A.” The Journal of Political Economy 81 (May-June. 9. and W. Hull. T. 2. Howison. New York: Springer-Verlag (1991). Ch. An Introduction to the Mathematics of Financial Derivatives. Baxter. P. Financial Calculus. M. New Jersey: Prentice-Hall (2003). and A. 2. Neftci. U. Chichester. ed. 10. The first finance application of Itô’s Lemma was the classic Black-Scholes model. Salih. Briys. “An Intuitive Look at Itô’s Lemma: A Pedagogical Note. 2. 1973). U. 11. Mai.