Brownian Motion Chuan-Hsiang Han November 24, 2010 Symmetric Random Walk Given Ω∞ , โฑ, ๐ ; let ๐ = ๐1 , ๐2 , ๐3 โฏ ∈ Ω∞ ๐ and ๐ ๐ป = ๐ ๐ป = , and ๐๐ denotes the ๐ outcome of ๐th toss. Define the r.v.'s ∞ ๐๐ that for each ๐ ๐=1 +1, ๐๐ = −1, ๐๐ ๐๐ = ๐ป ๐๐๐๐ = ๐ A S.R.W. is a process ๐๐ ∞ ๐=0 such that ๐0 = 0 ๐ and ๐๐ = ๐=1 ๐๐ , ๐ = 1,2, โฏ . Independent Increments of S.R.W. Choose 0 = ๐0 < ๐1 < โฏ < ๐๐ , the r.v.s ๐๐1 = ๐๐1 − ๐๐0 , ๐๐2 − ๐๐1 , โฏ ๐๐๐ − ๐๐๐−1 are independent, where the increment is defined by ๐๐+1 ๐๐๐+1 − ๐๐๐ = ๐=๐๐ +1 ๐๐ . Note: (1) Increments are independent. (2) The increment ๐๐๐+1 − ๐๐๐ has mean 0 and variance ๐๐+1 − ๐๐ .(Stationarity) Martingale Property of S.R.W. For any nonnegative integers ๐ > ๐, ๐ธ ๐๐ โฑ๐ = ๐ธ ๐๐ − ๐๐ + ๐๐ โฑ๐ = ๐๐ โฑ๐ contains all the information of the first ๐ coin tosses. If R.W. is not symmetric, it is not a martingale. Markov Property of S.R.W. For any nonnegative integers ๐ > ๐ and any integrable function ๐ ๐ธ ๐ ๐๐ โฑ๐ = ๐ธ ๐ ๐๐ − ๐๐ + ๐๐ โฑ๐ ๐−๐ = ๐ถ ๐ − ๐, ๐ ๐=0 1 2 ๐ 1 2 ๐−๐−๐ ๐ 2๐ − ๐ + ๐ Quadratic Variation of S.R.W. The quadratic variation up to time ๐ is defined to be ๐ ๐, ๐ ๐ = ๐๐ − ๐๐−1 2 =๐ ๐=1 Note the difference between ๐๐๐ ๐๐ = ๐ (an average over all paths), and ๐, ๐ ๐ = ๐. (pathwise property) Scaled S.R.W. Goal: to approximate Brownian Motion 1 ๐ ๐ ๐ก๐ = ๐๐๐ก๐ ๐๐ก๐ ∈๐+ ๐ 1 ๐ 1. new time interval is "very small" of instead of 1 2. magnitude is "small" of 1 ๐ instead of 1. For any ๐ก ∈ 0, ∞ , ๐ ๐ ๐ก can be defined as a linear interpolation between the nearest ๐ก๐ such that ๐ก๐ ≤ ๐ก < ๐ก๐+1 . Properties of Scaled S.R.W. (i) independent increments: for any 0 = ๐ก0 < ๐ก1 < โฏ < ๐ก๐ , ๐ ๐ ๐ ๐ ๐ก1 − ๐ ๐ก2 − ๐ ๐ ๐ ๐ก1 ๐ก0 , ,โฏ ๐ ๐ ๐ก๐ − (iii) Martingale property ๐ธ ๐ ๐ ๐ก โฑ๐ = ๐ ๐ ๐ . (iv) Markov Property: for any function ๐, these exists a function ๐ so that ๐ธ ๐ ๐ ๐ ๐ก โฑ๐ = ๐ ๐ ๐ ๐ . (v) Quadratic variation: for any ๐ก ≥ 0, ๐๐ก ๐ ๐ ๐๐ก = ๐=1 ,๐ ๐ ๐ก = ๐ ๐=1 1 = ๐ก. ๐ ๐ ๐ −๐ ๐ ๐ ๐−1 ๐ ๐๐ก 2 = ๐=1 1 ๐๐ ๐ 2 Limiting (Marginal) Distribution of S.R.W. Theorem 3.2.1. (Central Limit Theorem) For any fixed ๐ก ≥ 0, ๐ ๐ ๐ก ๐↑∞ ๐ โ ๐๐ก ~๐ฉ 0, ๐ก in dist. or ๐ ๐ ๐ ๐ก ≤๐ฅ ๐↑∞ ๐ฅ −∞ Proof: shown in class. 1 2๐๐ก ๐ −๐ง 2 2๐ก ๐๐ง . A Numerical Example ๐ ๐: 0 ≤ ๐ 100 0.25 ≤ 0.2 = ๐ ๐: 0 ≤ ๐25 ≤ 2 = 0.1555 0.2 2 −2๐ง 2 ๐ ๐: 0 ≤ ๐ 0.25 ≤ 0.2 = ๐ ๐๐ง 2๐ 0 ≈ 0.1554 Log-Normality as the Limit of the Binomial Model Theorem 3.2.2. (Central Limit Theorem) For any fixed ๐ก ≥ 0, ๐๐ ๐ก = ๐ 0 =๐ 0 ๐ป๐๐ก ๐๐๐ก ๐↑∞ ๐ข๐ ๐๐ ๐ ๐ก ๐2๐ก − −๐๐๐ก 2 ๐ in the distribution sense, where ๐ข๐ = 1 + ๐๐ = 1 − ๐ ,and ๐ ๐ท ๐=๐ป = 1+๐−๐๐ ๐ข๐ −๐๐ ๐ , ๐ What is Brownian Motion? "If ๐ is a continuous process with independent increments that are normally distributed, then ๐ is a Brownian motion." Standard Brownian Motions check Definition 3.3.1 in the text. Definition of SBM: Let the stochastic process ๐๐ก , ๐ก ≥ 0 under a probability space Ω, โฑ, P be continuous and satisfy: 1. ๐0 = 0 2. ๐๐ก+๐ − ๐๐ก ~๐ฉ 0, ๐ 3. ๐๐ก+๐ − ๐๐ก is independent of ๐๐ก๐ − ๐๐ก๐+1 for ๐ก0 < โฏ ๐ก๐ = ๐ก. Covariance Matrix Check ๐ถ๐๐ฃ ๐๐ , ๐๐ก = ๐๐๐ ๐ , ๐ก for any nonnegative ๐ and ๐ก ๐ For any vector ๐ = ๐๐ก1 , ๐๐ก2 , โฏ , ๐๐ก๐ with 0 ≤ ๐ก1 ≤ โฏ ≤ ๐ก๐ , ๐ก1 ๐ก1 โฏ ๐ก1 ๐ก1 ๐ก2 โฏ ๐ก2 ๐ ๐ถ๐๐ฃ ๐๐ = โ๐ถ โฎ โฎ โฎ ๐ก1 ๐ก2 โฏ ๐ก๐ In fact, ๐~๐ฉ 0, ๐ถ . Joint Moment-Generating Function of BM ๐ ๐ข1 , ๐ข2 , โฏ , ๐ข๐ = ๐ธ ๐๐ฅ๐ ๐ข โ ๐ = ๐ธ ๐๐ฅ๐ ๐ข1 ๐๐ก1 + ๐ข2 ๐๐ก2 + โฏ + ๐ข๐ ๐๐ก๐ 1 = ๐๐ฅ๐ ๐ข1 + ๐ข2 + โฏ + ๐ข๐ 2 ๐ก1 2 1 + ๐ข2 + โฏ + ๐ข๐ 2 ๐ก2 − ๐ก1 + โฏ 2 Alternative Characteristics of Brownian Motion (Theorem 3.3.2) For any continuous process ๐๐ก , ๐ก ≥ 0 with ๐0 = 0, the following three properties are equivalent. (i) increments are independent and normally distributed. (ii) For any 0 ≤ ๐ก0 ≤ ๐ก1 ≤ โฏ ≤ ๐ก๐ , ๐๐ก1 , ๐๐ก2 , โฏ , ๐๐ก๐ are jointly normally distributed. (ii) ๐๐ก1 , ๐๐ก2 , โฏ , ๐๐ก๐ has the joint momentgenerating function as before. If any of the three holds, then ๐๐ก , ๐ก ≥ 0, is a SBM. Filtration for B.M. Definition 3.3.3 Let Ω, โฑ, ๐ท be a probability space on which the B.M. ๐๐ก ๐ก≥0 is defined. A filtration for the B.M. is a collection of ๐-algebras โฑ๐ก ๐ก≥0 , satisfying (i) (Information accumulates) For 0 ≤ ๐ ≤ ๐ก, โฑ๐ ⊆ โฑ๐ก . (ii) (Adaptivity) each ๐๐ก is โฑ๐ก -measurable. (iii) (Independence of future increments) 0 ≤ ๐ก < ๐ข , the increment ๐๐ข − ๐๐ก is independent of โฑ๐ก . [Note, this property leads to Efficient Market Hypothesis.] Martingale property Theorem 3.3.4 B.M. is a martingale. Proof: ๐ธ ๐๐ก |โฑ๐ = โฏ = ๐๐ Levy's Characteristics of Brownian Motion The process ๐๐ก is SBM iff the conditional characterization function is ๐ธ ๐ ๐๐ข ๐๐ก −๐๐ฅ |โฑ๐ = ๐ข2 ๐ก−๐ − 2 ๐ Variations: First-Order (Total) Variation Given a function ๐ defined on 0, ๐ , the total variation ๐๐๐ ๐ is defined by ๐−1 ๐๐๐ ๐ = lim Π →0 ๐ ๐ก๐+1 − ๐ ๐ก๐ ๐=0 where the partition Π = ๐ก0 = 0, ๐ก1 , โฏ , ๐ก๐ = ๐ and Π = ๐๐๐ฅ๐=0,โฏ,๐−1 ๐ก๐+1 − ๐ก๐ If ๐ is differentiable, ๐ ๐ก๐+1 − ๐ ๐ก๐ = ๐′ ๐ก๐โ ๐ก๐+1 − ๐ก๐ for some ๐ก๐โ ๐ ๐ก๐+1 , ๐ก๐ . Then ๐๐๐ ๐ = lim lim Π →0 Π →0 ๐−1 โ ๐′ ๐ก ๐ ๐=0 ๐−1 ๐=0 ๐ ๐ก๐+1 − ๐ ๐ก๐ ๐ก๐+1 − ๐ก๐ = ๐ 0 = ๐′ ๐ก ๐๐ก. Quadratic Variation Def. 3.4.1 The quadratic variation of ๐ up to time ๐ is defined by ๐−1 ๐, ๐ ๐ = lim Π →0 ๐ ๐ก๐+1 − ๐ ๐ก๐ ๐=0 2 If ๐ is continuous differentiable, ๐ ๐ก๐+1 − ๐ ๐ก๐ = ๐′ ๐ก๐โ ๐ก๐+1 − ๐ก๐ for some ๐ก๐โ ๐ ๐ก๐+1 , ๐ก๐ . Then ๐, ๐ ๐ = lim ๐−1 ๐ ๐ก๐+1 ๐=0 Π →0 ๐−1 โ 2 lim ๐=0 ๐′ ๐ก๐ ๐ก๐+1 − ๐ก๐ Π →0 ๐ 2 ๐′ ๐ก ๐๐ก 0 − ๐ ๐ก๐ = lim 2 Π →0 ≤ Π × Quadratic Variation of B.M. Thm. 3.4.3 Let ๐๐ก≥0 be a Brownian Motion. Then ๐, ๐ ๐ = ๐ for all ๐ ≥ 0 a.s.. B.M. accumulates quadratic variation at rate one per unit time. Informal notion: ๐๐๐ก โ ๐๐๐ก = ๐๐ก, ๐๐๐ก โ ๐๐ก = 0, ๐๐ก โ ๐๐ก = 0 Geometric Brownian Motion The geometric Brownian motion is a process of the following form ๐๐ก = ๐0 ๐๐ฅ๐ ๐๐๐ก + ๐ − ๐ 2 2 ๐ก . where ๐0 is the current value, ๐๐ก≥0 is a B.M., ๐ is the drift and ๐ > 0 is the volatility. For each partition ๐ก0 = 0, ๐ก1 , โฏ , ๐ก๐ = ๐ , define the log returns ๐๐ก๐+1 ๐๐๐ = ๐ ๐๐ก๐+1 − ๐๐ก๐ + ๐ − ๐ 2 2 ๐ก๐+1 − ๐ก๐ ๐๐ก๐ Volatility Estimation of GBM The realized variance is defined by 2 ๐−1 ๐๐ก๐+1 ๐๐๐ ๐๐ก๐ ๐=0 which converges to ๐ 2 ๐ as Π → 0 BM is a Markov process Thm. 3.5.1 Let ๐๐ก≥0 be a B.M. and โฑ๐ก≥0 be a filtration for this B.M.. Then (1)Wt0 is a Markov process. Thm. 3.6.1. (2)๐๐ก = ๐๐ฅ๐ ๐๐๐ก − 1 2 ๐ ๐ก 2 is martingale. (We call ๐๐ก exponential martingale.) Note: ๐ธ ๐๐ก = 1