Copyright

c 2009 by Karl Sigman

1 Gambler’s Ruin Problem

Let N ≥ 2 be an integer and let 1 ≤ i ≤ N − 1. Consider a gambler who starts with an
initial fortune of $i and then on each successive gamble either wins $1 or loses $1 independent
of the past with probabilities p and q = 1 − p respectively. Let Xn denote the total fortune
after the nth gamble. The gambler’s objective is to reach a total fortune of $N , without first
getting ruined (running out of money). If the gambler succeeds, then the gambler is said to win
the game. In any case, the gambler stops playing after winning or getting ruined, whichever
happens first.
{Xn } yields a Markov chain (MC) on the state space S = {0, 1, . . . , N }. The transition
probabilities are given by Pi,i+1 = p, Pi,i−1 = q, 0 < i < N , and both 0 and N are absorbing
states, P00 = PN N = 1.1
For example, when N = 4 the transition matrix is given by
 
1 0 0 0 0

 q 0 p 0 0 

P = 0 q 0 p 0 .
 
 
 0 0 q 0 p 
0 0 0 0 1
While the game proceeds, this MC forms a simple random walk
Xn = i + ∆1 + · · · + ∆n , n ≥ 1, X0 = i,
where {∆n } forms an i.i.d. sequence of r.v.s. distributed as P (∆ = 1) = p, P (∆ = −1) = q =
1 − p, and represents the earnings on the successive gambles.
Since the game stops when either Xn = 0 or Xn = N , let
τi = min{n ≥ 0 : Xn ∈ {0, N }|X0 = i},
denote the time at which the game stops when X0 = i. If Xτi = N , then the gambler wins, if
Xτi = 0, then the gambler is ruined.
Let Pi (N ) = P (Xτi = N ) denote the probability that the gambler wins when X0 = i.

Pi (N ) denotes the probability that the gambler, starting initially with $i, reaches a
total fortune of N before ruin; 1 − Pi (N ) is thus the corresponding probably of ruin

Clearly P0 (N ) = 0 and PN (N ) = 1 by definition, and we next proceed to compute Pi (N ), 1 ≤

i ≤ N − 1.

Proposition 1.1 (Gambler’s Ruin Problem)

 q i
 1−( p ) , if p 6= q;
q N
Pi (N ) = 1−( p ) (1)
 i
N, if p = q = 0.5.
1
There are three communication classes: C1 = {0}, C2 = {1, . . . , N − 1}, C3 = {N }. C1 and C3 are recurrent
whereas C2 is transient.

1
Proof : For our derivation, we let Pi = Pi (N ), that is, we suppress the dependence on N for
ease of notation. The key idea is to condition on the outcome of the first gamble, ∆1 = 1 or
∆1 = −1, yielding
Pi = pPi+1 + qPi−1 . (2)
The derivation of this recursion is as follows: If ∆1 = 1, then the gambler’s total fortune
increases to X1 = i+1 and so by the Markov property the gambler will now win with probability
Pi+1 . Similarly, if ∆1 = −1, then the gambler’s fortune decreases to X1 = i − 1 and so
by the Markov property the gambler will now win with probability Pi−1 . The probabilities
corresponding to the two outcomes are p and q yielding (2). Since p + q = 1, (2) can be
re-written as pPi + qPi = pPi+1 + qPi−1 , yielding
q
Pi+1 − Pi = (Pi − Pi−1 ).
p

In particular, P2 − P1 = (q/p)(P1 − P0 ) = (q/p)P1 (since P0 = 0), so that

P3 − P2 = (q/p)(P2 − P1 ) = (q/p)2 P1 , and more generally
q i
Pi+1 − Pi = ( ) P1 , 0 < i < N.
p
Thus
i
X
Pi+1 − P1 = (Pk+1 − Pk )
k=1
i
X q k
= ( ) P1 ,
k=1
p

yielding
i i
X q k X q k
Pi+1 = P1 + P1 ( ) = P1 ( )
k=1
p k=0
p

1−( pq )i+1
P1 , if p 6= q;

= 1−( pq ) (3)
P1 (i + 1), if p = q = 0.5.


1−ai+1
(Here we are using the “geometric series” equation in=0 ai =
P
1−a , for any number a and
any integer i ≥ 1.)
Choosing i = N − 1 and using the fact that PN = 1 yields

1−( pq )N
P1 , if p 6= q;

1 = PN = 1−( pq )
P1 N, if p = q = 0.5,


from which we conclude that

 q
 1− p , if p 6= q;
q N
P1 = 1−( p
)
 1
N, if p = q = 0.5,

2
thus obtaining from (3) (after algebra) the solution
 q i
 1−( p ) , if p 6= q;
q N
Pi = 1−( p ) (4)
 i
N, if p = q = 0.5.

1.1 Becoming infinitely rich or getting ruined

In the formula (1), it is of interest to see what happens as N → ∞; denote this by Pi (∞) =
limN →∞ Pi (N ). This limiting quantity denotes the probability that the gambler , if allowed
to play forever unless ruined, will in fact never get ruined and instead will obtain an infinitely
large fortune.

Proposition 1.2 Define Pi (∞) = limN →∞ Pi (N ). If p > 0.5, then

q
Pi (∞) = 1 − ( )i > 0. (5)
p
If p ≤ 0.50, then
Pi (∞) = 0. (6)
Thus, unless the gambles are strictly better than fair (p > 0.5), ruin is certain.

Proof : If p > 0.5, then pq < 1; hence in the denominator of (1), ( pq )N → 0 yielding the result.
If p < 0.50, then pq > 1; hence in the the denominator of (1), ( pq )N → ∞ yielding the result.
Finally, if p = 0.5, then pi (N ) = i/N → 0.

Examples
1. John starts with $2, and p = 0.6: What is the probability that John obtains a fortune of
N = 4 without going broke?
SOLUTION i = 2, N = 4 and q = 1 − p = 0.4, so q/p = 2/3, and we want

1 − (2/3)2
P2 (4) = = 0.91
1 − (2/3)4

2. What is the probability that John will become infinitely rich?

SOLUTION
P2 (∞) = 1 − (2/3)2 = 5/9 = 0.56

3. If John instead started with i = $1, what is the probability that he would go broke?
SOLUTION
The probability he becomes infinitely rich is P1 (∞) = 1 − (q/p) = 1/3, so the probability
of ruin is 1 − P1 (∞) = 2/3.

3
1.2 Applications
Risk insurance business
Consider an insurance company that earns $1 per day (from interest), but on each day, indepen-
dent of the past, might suffer a claim against it for the amount $2 with probability q = 1 − p.
Whenever such a claim is suffered, $2 is removed from the reserve of money. Thus on the
nth day, the net income for that day is exactly ∆n as in the gamblers’ ruin problem: 1 with
probability p, −1 with probability q.
If the insurance company starts off initially with a reserve of $i ≥ 1, then what is the
probability it will eventually get ruined (run out of money)?
The answer is given by (5) and (??): If p > 0.5 then the probability is given by ( pq )i > 0,
whereas if p ≤ 0.5 ruin will always ocurr. This makes intuitive sense because if p > 0.5, then
the average net income per day is E(∆) = p − q > 0, whereas if p ≤ 0.5, then the average net
income per day is E(∆) = p − q ≤ 0. So the company can not expect to stay in business unless
earning (on average) more than is taken away by claims.

1.3 Random walk hitting probabilities

Let a > 0 and b > 0 be integers, and let Rn = ∆1 + · · · + ∆n , n ≥ 1, R0 = 0 denote a simple
random walk initially at the origin. Let

p(a) = P ({Rn } hits level a before hitting level −b).

By letting i = b, and N = a + b, we can equivalently imagine a gambler who starts with

i = b and wishes to reach N = a + b before going broke. So we can compute p(a) by casting
the problem into the framework of the gambler’s ruin problem: p(a) = Pi (N ) where N = a + b,
i = b. Thus
 q b
 1−(q p ) , if p 6= q;
p(a) = 1−( p )a+b (7)
 b
a+b , if p = q = 0.5.

Examples
1. Ellen bought a share of stock for $10, and it is believed that the stock price moves (day
by day) as a simple random walk with p = 0.55. What is the probability that Ellen’s
stock reaches the high value of $15 before the low value of $5?
SOLUTION
We want “the probability that the stock goes up by 5 before going down by 5.” This is
equivalent to starting the random walk at 0 with a = 5 and b = 5, and computing p(a).

1 − ( pq )b 1 − (0.82)5
p(a) = = = 0.73
1− ( pq )a+b 1 − (0.82)10

2. What is the probability that Ellen will become infinitely rich?

SOLUTION

4
 Here we equivalently want to know the probability that a gambler starting with i = 10
becomes infinitely rich before going broke. Just like Example 2 on Page 3:
1 − (q/p)i = 1 − (0.82)10 ≈ 1 − 0.14 = 0.86.

1.4 Maximums and minimums of the simple random walk

Formula (7) can immediately be used for computing the probability that the simple random
walk {Rn }, starting initially at R0 = 0, will ever hit level a, for any given positive integer
a ≥ 1: Keep a fixed while taking the limit as b → ∞ in (7). The result depends on wether
p < 0.50 or p ≥ 0.50. A little thought reveals that we can state this problem as computing the
def
tail P (M ≥ a), a ≥ 0, where M = max{Rn : n ≥ 0} is the all-time maximum of the random
walk; a non-negative random variable, because {M ≥ a} = {Rn = a, for some n ≥ 1}.
def
Proposition 1.3 Let M = max{Rn : n ≥ 0} for the simple random walk starting initially at
the origin (R0 = 0).
1. When p < 0.50,
P (M ≥ a) = (p/q)a , a ≥ 0;
M has a geometric distribution with “success” probability 1 − (p/q):

P (M = k) = (p/q)k (1 − (p/q)), k ≥ 0.

In this case, the random walk drifts down to −∞, wp1, but before doing so reaches the
finite maximum M .

2. If p ≥ 0.50, then P (M ≥ a) = 1, a ≥ 0: P (M = ∞) = 1; the random walk will, with

probability 1, reach any positive integer a no matter how large.
Proof : Taking the limit in (7) as b → ∞ yields the result by considering the two cases p < 0.5
or p ≥ 0.5: If p < 0.5, then (q/p) > 1 and so both (q/p)b and (q/p)a+b tend to ∞ as b → ∞.
But before taking the limit, multiply both numerator and denominator by (q/p)−b = (p/q)b ,
yielding
(p/q)b − 1
p(a) = .
(p/q)b − (q/p)a
Since (p/q)b → 0 as b → ∞, the result follows.
If p > 0.5, then (q/p) < 1 and so both (q/p)b and (q/p)a+b tend to 0 as b → ∞ yielding the
limit in (7) as 1. If p = 0.5, then p(a) = b/(b + a) → 1 as b → ∞.

If p < 0.5, then E(∆) < 0, and if p > 0.5, then E(∆) > 0; so Proposition 1.3 is consistent
with the fact that any random walk with E(∆) < 0 (called the negative drift case) satisfies
limn→∞ Rn = −∞, wp1, and any random walk with E(∆) > 0 ( called the positive drift case)
satisfies limn→∞ Rn = +∞, wp1. 2
But furthermore we learn that when p < 0.5, although wp1 the chain drifts off to −∞, it
first reaches a finite maximum M before doing so, and this rv M has a geometric distribution.
2 Rn
From the strong law of large numbers, limn→∞ n
= E(∆), wp1, so Rn ≈ nE(∆) → −∞ if E(∆) < 0 and
→ +∞ if E(∆) > 0.

5
 Finally Proposition 1.3 also offers us a proof that when p = 0.5, the symmetric case, the
random walk will wp1 hit any positive value, P (M ≥ a) = 1.
By symmetry, we also obtain analogous results for the minimum, :
def
Corollary 1.1 Let m = min{Rn : n ≥ 0} for the simple random walk starting initially at the
origin (R0 = 0).

1. If p > 0.5, then

P (m ≤ −b) = (q/p)b , b ≥ 0.
In this case, the random walk drifts up to +∞, but before doing so drops down to a finite
minimum m ≤ 0. Taking absolute values of m makes it non-negative and so we can
express this result as P (|m| ≥ b) = (q/p)b , b ≥ 0; |m| has a geometric distribution with
“success” probability 1 − (q/p): P (|m| = k) = (q/p)k (1 − (q/p)), k ≥ 0.

2. If p ≤ 0.50, then P (m ≤ −b) = 1, b ≥ 0: P (m = −∞) = 1; the random walk will, with

probability 1, reach any negative integer a no matter how small.

Note that when p < 0.5, P (M = 0) = 1 − (p/q) > 0. This is because it is possible that
the random walk will never enter the positive axis before drifting off to −∞; with positive
probability Rn ≤ 0, n ≥ 0. Similarly, if p > 0.5, then P (m = 0) = 1 − (q/p) > 0; with positive
probability Rn ≥ 0, n ≥ 0.

Recurrence of the simple symmetric random walk

Combining the results for both M and m in the previous section when p = 0.5, we have

Proposition 1.4 The simple symmetric (p = 0.50) random walk, starting at the origin, will
wp1 eventually hit any integer a, positive or negative. In fact it will hit any given integer a
infinitely often, always returning yet again after leaving; it is a recurrent Markov chain.

Proof : The first statement follows directly from Proposition 1.3 and Corollary 1.1; P (M =
∞) = 1 = P (m = −∞) = 1. For the second statement we argue as follows: Using the first
statement, we know that the simple symmetric random walk starting at 0 will hit 1 eventually.
But when it does, we can use the same result to conclude that it will go back and hit 0 eventually
after that, because that is stochastically equivalent to starting at R0 = 0 and waiting for the
chain to hit −1, which also will happen eventually. But then yet again it must hit 1 again
and so on (back and forth infinitely often), all by the same logic. We conclude that the chain
will, over and over again, return to state 0 wp1; it will do so infinitely often; 0 is a recurrent
state for the simple symmetric random walk. Thus (since the chain is irreducible) all states are
recurrent.
Let τ = min{n ≥ 1 : Rn = 0 | R0 = 0}, the so-called return time to state 0. We just argued
that τ is a proper random variable, that is, P (τ < ∞) = 1. This means that if the chain starts
in state 0, then, if we wait long enough, we will (wp1) see it return to state 0. What we will
prove later is that E(τ ) = ∞; meaning that on average our wait is infinite. This implies that
the simple symmetric random walk forms a null recurrent Markov chain.