Stochastic Calculus Cheatsheet Standard Brownian Motion / Wiener process E[dX 2 ] = dt E[dX] = 0 limdt→0 dX 2 = dt √ Discrete approx: dX = φ dt where φ ∼ N (0, 1) dX is O(dt1/2 ) dtdX is O(dt3/2 ) Itô Product Rule Characterization: 1. 2. 3. 4. X(0) = 0 Continuous everywhere, differentiable nowhere X(t) − X(s) ∼ N (0, |t − s|) X(t + s) − X(t) is independent of X(t) Levy’s characterization: 3. Xt is a martingale w.r.t. the filtration Ft 4. |X|2 − t is a martingale w.r.t. the filtration Ft If dXt = αdt + βdWt and dYt = γdt + λdWt , d(Xt Yt ) = Xt dYt + Yt dXt + dXdY 1 = Xt dYt + Yt dXt + βλdt 2 Stochastic Differential Equations (General Form) dS = f (t, S) dt + g(t, S) dXi dSi = fi (t, S0 , . . . , Sn ) dt + gi (t, S0 , . . . , Sn ) dXi where f is the drift, g is the diffusion Itô’s Lemma and Basic Stochastic Integration For F (Xt ) dF 1 d2 F dF = dXt + dt dX 2 dX 2 Z F (Xt ) = F (X0 ) + 0 t dF 1 dXτ + dX 2 Z 0 t d2 F dτ dX 2 For F (Xt , t) dF = ∂F dXt + ∂X ∂F 1 ∂2F + ∂t 2 ∂X 2 t Z dt F (Xt , t) = F (X0 , 0) + 0 ∂F dXτ + ∂X Z t 0 1 ∂2F ∂F + ∂t 2 ∂X 2 Functions of Stochastic Functions 1-dimensional: V (t, S) dV = = 1. Apply Taylor expansion on V 2. Apply Itô’s Lemma: ∂V ∂V 1 ∂2V dt + dS + g 2 2 dt ∂t ∂S 2 ∂S ∂V ∂V 1 ∂2V ∂V +f + g 2 2 dt + g dX ∂t ∂S 2 ∂S ∂S • dXi2 → dt • dXi dXj → ρij dt 3. Regroup the terms in dt and dXi 4. Sto.integ.: integrate the resulting DE 2-dimensional: V (t, S1 , S2 ) dV = ∂V ∂V ∂V 1 ∂2V ∂2V 1 ∂2V + f1 + f2 + g12 2 + ρg1 g2 + g22 2 ∂t ∂S1 ∂S2 2 ∂S1 ∂S1 ∂S2 2 ∂S2 dt + g1 ∂V ∂V dX1 + g2 dX2 ∂S1 ∂S2 n-dimensional: V (t, S1 , . . . , Sn ) dV = n X n 1X ∂V ∂V + fi + ∂t ∂S 2 i i=1 i=1 gi2 n X n X ∂ V ∂ V ∂V + ρ g g dt + gi dXi ij i j ∂Si2 i=1,j>1 ∂Si ∂Sj ∂S i i=1 2 2 dτ Transition Density Functions Solution Forward Kolmogorov 1 ∂2 ∂p ∂ = B(y 0 , t0 )2 p − 0 (A(y 0 , t0 )p) 0 02 ∂t 2 ∂y ∂y log p(S, t; S 0 , t0 ) = 1 σS 0 p 2π(t0 − t) − S S0 e 2 + µ − 21 σ 2 (t0 − t) 2σ 2 (t0 − t) Common Processes/Dynamics Geometric Brownian Motion (Lognormal) Brownian Motion with Drift dS = µS dt + σS dX dS = µ dt + σ dX dS = µ dt + σ dX S Cox, Ingersoll, Ross Vasiček (1977) dS = γ(r̄ − r) dt + σ dX FIXME TODO add others, Ho Lee and company... 1 dS = (υ − σS) dt + σS 2 dX All you need to know about Sto.Calc (FIXME integrate these words of wisdom from Antoine.) • If Xt → N (µ, σ) then E(xXt ) = eµ+ σ2 2 . • Itô: d(f (Xt )) • Itô: d(Xt Yt ) = Xt dYt + Yt dXt + 12 βλdt where dXt = αdt + βdWt and dYt = γdt + λdWt R • E[ Xt dWt ] = 0 R R • V ar[ Xt dWt ] = Xt2 dt • Girsanov’s theorem. • Generating correlated X and Y . Martingales Probability Spaces Unconditional Expectation “Let (Ω, F, P) be a probability space. . . ” Expected value under a prob. measure (Lebesgue integral): Z Z Z E[h(X)] = h(x)p(x)dx = h(x)d(P(x)) = h(x)d P Ω ZΩ ZΩ E[1{X∈A} ] = 1{X∈A} d P = d P = P(A) • Ω: sample space • F: filtration (information set), (Note that Ft1 ⊆ Ft2 ⊆ FT ≡ F) • P: probability measure Ω A Conditional Expectation Martingales (Definition) (Use these to prove that a process is a Martingale; use the definition.) E[Mt ] < ∞ E[Mt+1 |Ft ] = Mt ∀0 ≤ s ≤ t E[Mt+1 |Ft ] ≤ Mt (supermartingale) E[Mt+1 |Ft ] ≥ Mt (submartingale) 1. Linearity: E[aX + bY |F] = aE[X|F] + bE[Y |F] 2. Tower Property: if F ⊂ G, E[E[X|G] |F] = E[X|F] Wiener ∈ Martingale (driftless) ⊂ Markov (memoryless) ⊂ nonMarkov E[E[X|F]] = E[X] 3. Taking out what is known: Equivalent Measures E[X|F] = X Absolute continuity: if P (A) = 0 → Q(A) = 0 Q is “absolutely continuous” w.r.t. P, and Q << P. ∀A. “It is allright to tinker with the probabilities as long as we do not tinker with the (im)possibilities.” Equivalent measures: if Q << P and P << Q. Radon-Nikodým Theorem Q(A) = Z A ΛdP E[XY |F] = XE[Y |F] 4. Independence: if X is independent from F, E[X|F] = E[X] 5. Positivity: if X ≥ 0 then E[X|F] ≥ 0 6. Jensen’ Inequality: if f is a convex function, then f (E[X|F]) ≤ E[f (X)|F] Exponential Martingale dQ is the R.N. derivative. dP where Λ = (if X is F-measurable but not Y :) M (t) = exp(St + f (t)) where f (t) = −(µ + 1 2 σ )t 2 Itô Integrals & Martingales Properties of Itô Integrals Itô integrals are Martingales: Z 1. Linearity: T E[ g(t, Xt )dXt ] = 0 0 Z T Z (αf (t)+βg(t))dXt = Martingale Representation Theorem If M is a Martingale, there exists g(t, X) such that Z T MT = M0 + g(t, X)dXt 0 0 T Z αf (t)dXt + 0 βg(t)dXt 0 2. Isometry: 2 "Z # Z T T E f (t)dXt = E |f (t)|2 dt 0 0 The rightmost term is an Itô integral (and thus also a Martingale). 3. Martingale: "Z Fubini’s Theorem "Z E T # Z f (Xt )dt = 0 E T E [f (Xt )] dt 0 T 0 T # Z s f (t)dXt Fs = f (t)dXt 0 Application of Martingales to Asset Pricing Warning: I still need to complete and arrange this page of notes. Fundamental Asset Pricing Formula Value = E Meas. [P V (expected cash flows)] Novikov Condition h 1 RT 2 i E e 2 0 θs ds < ∞ Risk-free Asset dBt = rBt dt, Mtθ = e(− B(0) = B0 B(t) = B0 ert 0 θs dXs− 21 Rt θs2 ds) 0 is a Martingale Girsanov’s Theorem Rt Rt 2 1 dQ = e(− 0 θs dXs− 2 0 θs ds) dP Z t XtQ = XtP + θ(s)ds Underlying S dSt = µSt dt + σSt dX, S(t) = S0 e Rt S(0) = S0 µt− 12 σ 2 +σXt 0 • Provides an expression for the Radon-Nikodým derivative. Removing the TVM • Gives an explicit correspondence btw P and Q in terms of their Brownian motion. S(T ) S (T ) = rt e 1 2 ∗ S (t) = S0∗ e(µ−r− 2 σ )t+σXt ∗ dS ∗ = (µ − r)S ∗ dt + σS ∗ dX Self-financing Portfolios . . . but does not tell you what θ is. We assume θ and check that it satisfies the Novikov condition. Then we have the RN derivative, and we can change measures! Doléans/Stochastic Exponential Trading Strategy: t Z φt = (φSt , φB t ) processes E θs dXs θs dXs − 0 Value : Vt (φ) = φSt St + Z t Vt (φ) = V0 (φ) + Z ∀t ∈ [0, T ] Z t φSu dSu + φB u dBu Z t θs2 ds 0 t θ(s)ds Feynman-Kač Equivalence 0 Arbitrage opportunity: PDE: V0 (φ) = 0 and 1 2 0 φB t Bt 0 with P (VT (φ) > 0) > 0 0 XtQ = XtP − Self-financing portfolio: no in/out flows. t Z = exp P (VT (φ) < 0) = 0 ∂V ∂V 1 ∂2V +µ + σ 2 2 − rV = 0, ∂t ∂S 2 ∂S dSt = µ(t, St )dt + σ(t, St )dXt V (T, S) = G(S) m Expectation: V (t, St ) = e−r(T −t) E [G(ST )|Ft ]