Scribd
Upload
42 views4 pages

Stat 150 Class Notes: Onur Kaya 16292609

1. The document discusses martingales, which are sequences of random variables where the expected future value is equal to the present value, and sub/supermartingales where the expected future value is less than or greater than the present value. 2. To solve Gambler's ruin problems, the document creates a martingale based on weighting the random walk increments, allowing the calculation of the probability of hitting ruin or escape. 3. For a biased coin with probability p of heads, the weighting factor is q/p, and the probability a gambler starting with $a goes broke is (q/p)a if q ≤ p and 1 if q > p.

Uploaded by

spitzersglare
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
42 views4 pages

Stat 150 Class Notes: Onur Kaya 16292609

1. The document discusses martingales, which are sequences of random variables where the expected future value is equal to the present value, and sub/supermartingales where the expected future value is less than or greater than the present value. 2. To solve Gambler's ruin problems, the document creates a martingale based on weighting the random walk increments, allowing the calculation of the probability of hitting ruin or escape. 3. For a biased coin with probability p of heads, the weighting factor is q/p, and the probability a gambler starting with $a goes broke is (q/p)a if q ≤ p and 1 if q > p.

Uploaded by

spitzersglare
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

STAT 150 CLASS NOTES

Onur Kaya 16292609 May 18, 2006

Martingales: A sequence of random variables (Mn ) is a martingale relative to the sequence (Xn ) if: 1. Mn is some measurable function of X1 , X2 ,....., Xn 2. E [Mn+1 |X1 , X2 , ....., Xn ] = Mn Notice that (1) + (2) E [Mn+1 |M1 , M2 , ....., Mn ] = Mn . Variations: If we replace = in (2) by then it is a submartingale a favorable game then it is a supermartingale unfavorable game Fundamental: I. If Mn is a MG then E (Mn ) = E (M0 ) is constant SubMG then E (Mn ) is increasing SuperMG then E (Mn ) is decreasing II. Take a MG Mn and T a stopping time relative to the sequence Xn . Then, T = n is a function of X1 , X2 ....Xn . Look at the process Mn MT n := Mn V T M n is a MG relative to Xn . Then, M Application: Back to Gamblers Ruin Setup. Xi = 1 w.p. 1/2. 1 if if nT n>T

n := M

2 M0 = a, Mn = a + X1 + X2 + .... + Xn T=rst n s.t.Mn 0, b E [Mn V T ] = E [M0 ] = a E [Mn V T ] = bP (T n, MT = b) + 0.P (T n, MT = 0) + E [Mn 1(T > n)] Let n , then the last term is 0 and P (MT = b) = a/b Pa (Tb < T0 ) = a/b Pa (T0 < Tb ) = 1 (a/b) = (b a)/b Claim: Pa (T0 < ) = 1 Proof: If T0 < Tb , then certainly T0 <
ba b

= Pa (T0 < Tb ) Pa (T0 < ) 1


ba b

Fix a and let b ,

Observe that Sn is a martingale relative to X1 , X2 , ....Xn . Ea (Sn+1 |X1 , ..., Xn ) = Sn and T0 is a stopping time. Look at: ST0 V n = Sn 1(T0 > n) Sn if T0 > n = 0 otherwise

Ea (ST0 V n ) = a as n , eventually n T0 ST0 V n = 0 We see that for w in a set of probability 1, ST0 V n (w) 0 as n . ST0 V n ST0 = 0 E (limYn ) = limE (Yn ) where Yn := ST0 V n Gamblers ruin: Biased Coin Sn = a + X1 + ... + Xn as before but now distribution of Xi is 1 w.p. q and +1 w.p. p where p + q = 1 and p = q . Pa (Tb < T0 ) =?

3 Idea: Create a martingale to solve the problem. Notice that (Sn n(p q ), n = 1, 2, ..) is a MG. This is true but not immediately useful. We want to deal with stopping at T = T0 (r
Sn

Tb . Consider the process

, n = 1, 2, ...) for some r. We can pick r so this is a MG.

Ea (rSn+1 |X1 ....Xn )

= Ea (rSn +Xn+1 |X1 ....Xn ) = rSn Ea (rXn+1 |X1 ....Xn ) = rSn Ea (rXn+1 ) by independence = rSn [r+1 p + r1 q ]

We want to choose r so this reduces to rSn (i.e. We want rp + r1 q = 1 r = q/p T = T0 Tb

Observe rSn is between r0 and rb for n T which means that 0 Sn b. Look at Ea [rSn Also, rb if Tb T0 and Tb n = r0 if T n and ST = 0 between r0 and rb otherwise
V T

] = ra

r Sn

By same argument as before, P (Ta < ) = 1 Pa (T > n) 0 as n . Finally, ra = rb Pa (Tb < T0 ) + r0 Pa (T0 < Tb ) + 0 1 = Pa (Tb < T0 ) + Pa (T0 < Tb ) Pa (hit b before 0)= Pa (hit 0 before b)=
a

r a 1 r b 1

r b r a r b 1

where r=q/p. =
a b

1 Check: limr1 r = r b 1

ar a1 br b1

Example: Gambler starts with $a and plays a favorable game (pq). What is the probability that the gambler goes broke?
r Pa (T0 < ) = limb Pa (T0 < Tb ) = limb rrb = 1
b a

r a 1

= ra =

qa p

Notice that a = 1:

q p

= P1 (ever reach 0)

q 2 (p ) = P2 (ever reach 0)

4
q P2 (ever reach 1)= p

Summary ( q )a if q p p 1 if q > p (by similar limit)

Pa (T0 < ) =

You might also like