Markov Chains

tags: Stochastic Processes

Introduction

Consider a process that has a value in ecah time period. Let $X_{n}$ denote its value in time period $n$ , and suppose we want to make a probability model for the sequence of successive values $X_0, X_1, …, $. The simplest model would probably assume that the $X_{n}$ are independent random variables, but often such an assumption is clearly unjustified. However, it may be reasonable to assume that the conditional distribution of $X_{n + 1}$ given all the past observations, only depends on $X_{n}$ . Such an assumption defines a Markov chain, a type of stochastic process.

Let ${X_{n}, n = 0, 1, 2, \dots}$ be a stochastic process that takes on a finite or countable number of possible values. Unless otherwise mentioned, this set of possible values of the process will be denoted by the set of nonnegative integers ${0, 1, 2, \dots}$ . If $X_{n} = i$ , then the process is said to be in state $i$ at time $n$ . We suppose that whenever the process is in state $i$ , there is a fixed probabibility $P_{i j}$ that it will be next be in state $j$ . That is, we suppose that:

$P {X_{n + 1} = j | X_{n} = i, X_{n - 1} = i_{n - 1}, \dots} = P_{i j}$

for all states $i_{0}, i_{1}, \dots, i_{n - 1}, i, j$ and all $n \geq 0$ . The value $P_{i j}$ represents the probability that the proces will, when in state $i$ , next make a transition into state $j$ .

Chapman-Kolmogorov Equations

Having defined the one-step transition probabilities $P_{i j}$ , we can now define the $n$ -step transition probaiblities $P_{i j}^{n}$ to be the probability that the process in state $i$ will be in state $j$ after $n$ additional transitions. That is:

$P_{i j}^{n} = P {X_{n + k} = j | X_{k} = i}$

The Chapman-Kolmogorov equations provide a method for computing these n-step transition probabilities. These equations are:

$P_{i j}^{n + m} = \sum_{k = 0}^{\infty} P_{i j}^{n} P_{k j}^{m} for all n, m \geq 0, all i, j$

and are most easily understood by noting that $P_{i k}^{n} P_{k j}^{m}$ represents the probability that starting in $i$ the process will go to state $j$ in $n + m$ transitions through a path which takes it into state $k$ at the nth transition. Summing over all intermediate states yields the total probability that the process will be in state $j$ after $n + m$ transitions.

If we let $P^{(n)}$ denote the matrix of $n$-step transition probabilities $P_{i j}^{n}$ then the Chapman-Kolmogorov equation asserts that:

$P^{(n + m)} = P^{n} \cdot P^{m}$

The $n$-step transition matrix can be obtained by multiplying the matrix $P$ by itself $n$ times.

Classification of States

State $j$ is said to be accessible from state $i$ if $P_{i j}^{n} > 0$ for some $n \geq 0$ . Two states $i$ and $j$ that are accessible to each other are said to communicate, and we write $i \leftrightarrow j$ .

Any state communicates with itself, since by definition $P_{i i}^{0} = 1$ .

The relation of communication satisfies the following 3 properties:

State $i$ communicates with state $i$ , for all $i \geq 0$
If state $i$ communicates with state $j$ , then state $j$ communicates with state $i$ .
If state $i$ communicates with state $j$ , and state $j$ communicates with state $k$ , then state $i$ communicates with state $k$ .

Two states that communicate are said to be in the same class. The concept of communication divides the state space into several separate classes. The Markov chain is said to be irreducible if there is only 1 class, that is all states communicate with each other.

Limiting Probabilities

Suppose we have a $P$ matrix such as:

$P^{(4)} = [\begin{matrix} 0.5749 & 0.5241 \\ 0.5668 & 0.4332 \end{matrix}]$

We can compute $P^{(8)}$ :

$P^{(8)} = [\begin{matrix} 0.572 & 0.428 \\ 0.570 & 0.430 \end{matrix}]$

Notice that $P^{(8)}$ is almost identical to $P^{(4)}$ , and each of the rows of $P^{(8)}$ has almost identical entries. It seems that $P_{i j}^{n}$ is converging to some value that is the same for all $i$ . There seems to exist a limiting probability that the process will be in state $j$ after a large number of transitions, and this value is independent of the initial state.

State $i$ is said to have period $d$ if $P_{i i}^{n} = 0$ whenever $n$ is not divisible by $d$ , and $d$ is the largest integer with this property. For instance, it may be possible for the process to enter state $i$ only at times $2, 4, 6, \dots, 8$ , in which case state $i$ has period. A state with period 1 is said to be aperiodic.

In a finite-state Markov chain, all recurrent states are positive recurrent. Positive recurrent, aperiodic states are called ergodic.

For an irreducible ergodic Markov chain $lim_{n \to \infty} P_{i j}^{n}$ exists and is independent of $i$ . Furthermore, letting

$π_{j} = lim_{n \to \infty} P_{i j}^{n}, j \geq 0$

then $π_{j}$ is the unique nonnegative solution of:

$π_{j} = \sum_{i = 0}^{\infty} π_{i} P_{i j}, j \geq 0, \sum_{i = 0}^{\infty} π_{j} == 1$

We can obtain an expression for $P (X_{n + 1} = j)$ by conditioning on the state at time $n$ :

$\begin{aligned} P (X_{n + 1} = j) & = \sum_{i = 0}^{\infty} P (X_{n + 1} = j | X_{n} = i) P (X_{n} = i) \\ = \sum_{i = 0}^{\infty} P_{i j} P (X_{n} = i) \end{aligned}$

These long run proportions $π_{j}, j \geq 0$ are often called stationary probabilities. The reason being that if the initial state is chosen according to probabilities $π_{j}, j \geq 0$ , then the probability of being in state $j$ at any time $n$ is also equal to $π_{j}$ .