Introduction to Stochastic Processes Notes

Fundamental Concepts of Stochastic Processes

A stochastic process is formally defined as a collection of random variables that change over time. It is mathematically denoted as Xt:t∈T{X_t : t \in T}, where $ X_t $ represents the specific value of the process at a given time $ t $, and $ T $ represents the index set, also referred to as the time set. The key idea underlying this concept is that it describes a system that undergoes random changes as time progresses. To illustrate this, consider the number of customers entering a shop at time $ t $. Because the number of customers changes in a random fashion, this situation is classified as a stochastic process.

Index Sets and Temporal Dimensions

The index set, denoted as $ T $, indicates exactly when observations of the process are made. This set is also known as the time value. There are two primary types of index sets: discrete time and continuous time. A discrete-time index set involves specific, countable values such as 1,2,3,…{1, 2, 3, \dots}. A real-world example of this is the recording of daily sales. A continuous-time index set involves any value within a given interval, such as monitoring temperature variations over a period of time. It is important to note that the index does not necessarily refer to clock time; it can represent any sequence or specific order of events.

State Space and Outcome Varieties

The state space is defined as the set of all possible values that a stochastic process can take. Similar to index sets, state spaces are divided into discrete and continuous categories. A discrete state space consists of countable values. Examples include a coin toss, where the state space is $ {H, T} $, or the number of customers in a store, typically represented as 0,1,2,…{0, 1, 2, \dots}. In contrast, a continuous state space involves real numbers. A stock price, for instance, can be any real number and thus exists within a continuous state space.

Sample Paths and System Behavior

A sample path represents one specific possible outcome or realization of the stochastic process over time. For example, if we observe the number of customers in a shop over the course of three days, a specific sample path might look like $ (2, 3, 10) $. By examining a sample path, analysts can see exactly how a process behaves and evolves throughout its duration.

Dependence and Predictive Structure

In many stochastic processes, the random variables involved are dependent on one another. This dependency means that future values are influenced by past values. Two prominent examples of processes with dependent variables are stock prices and weather patterns. The existence of this dependence is vital because it creates a specific structure and allows for a degree of predictability within the process.

Filtration and Information Accumulation

Filtration refers to the total amount of information that is available regarding a process up to a specific time $ t $. This is mathematically denoted by Ft\mathcal{F}_t. The key idea behind filtration is that information increases as time moves forward, which assists in making predictions. For instance, at time $ 3 $, we possess the information from observations at times 1,2, and 31, 2, \text{ and } 3, which can be represented as $ (X_1, X_2, X_3) $.

White Noise as Pure Randomness

White noise is a specific sequence of random variables defined by three characteristics: they are independent, they have a mean of $ 0 $, and they possess a constant variance, denoted as σ2\sigma^2. The primary significance of white noise is that it represents pure randomness that contains no discernible pattern. Common examples of white noise include measurement errors in scientific experiments and random shocks observed in financial data.

Increments and Process Changes

An increment is defined as the actual change in a process between two distinct points in time, calculated as $ X_t - X_s $. In a financial context, if a stock price is $ 100 $ on Day 1 and $ 105 $ on Day 2, the increment is $ 105 - 100 = 5 $. There are two main types of increments: independent increments and stationary increments. Independent increments occur when the changes do not depend on one another. Stationary increments occur when the distribution of the change remains the same over time.

The Markov Property and Memorylessness

A stochastic process is said to have the Markov property if its future value depends solely on its present value, and not on any of its past values. This concept is fundamentally known as the memoryless property. An example of this property is weather modeling, where tomorrow's weather is assumed to depend only on today's weather conditions, regardless of what occurred last week.

Random Walks in Modeling

A random walk is a process where each subsequent step is determined randomly. This is expressed by the formula $ X_n = X_{n-1} + Z_n $, where $ Z_n $ is the random step. In a simple example, a process might start at $ 0 $. Each step could be $ +1 $ with a probability of $ 0.5 $ or $ -1 $ with a probability of $ 0.5 $. A possible path for such a process might be 0→1→0→1→20 \rightarrow 1 \rightarrow 0 \rightarrow 1 \rightarrow 2 or 0→−10 \rightarrow -1. Random walks are frequently utilized to model phenomena such as gambling outcomes and stock prices.

Poisson and Compound Poisson Processes

A Poisson process is a model for the number of random events occurring over a span of time. Its properties include events occurring randomly, the presence of independent increments, and a constant average rate, symbolized by λ\lambda. Typical examples include counting customers entering a store or the number of incoming calls at a call center. A compound Poisson process models the total cumulative value of these events, calculated using the formula Yt=∑i=1NtXiY_t = \sum_{i=1}^{N_t} X_i, where $ N_t $ is the number of events and $ X_i $ is the size of each individual event. For instance, in insurance, if $ 3 $ claims occur with sizes of $ 100 $, $ 200 $, and $ 500 $, the total loss for the process is $ 800 $.

Time Series Analysis

A time series is a stochastic process that has been observed over time and recorded as data. Examples include daily stock prices, monthly rainfall totals, and annual GDP (Gross Domestic Product) figures. The primary uses for time series data are forecasting future outcomes, identifying specific trends within data sets, and assisting in general decision-making processes.

Categories of Stochastic Processes

Stochastic processes are categorized into four types based on whether time (index set) and state space are discrete or continuous:

  1. Discrete-time, discrete-state: In these processes, both the time and the values are countable. An example of this category is Markov chains.
  2. Discrete-time, continuous-state: In these processes, time is discrete, but the values are continuous. Time series data typically fall into this category.
  3. Continuous-time, discrete-state: Time is continuous, but the possible values are countable. A Poisson process is a classic example of this.
  4. Continuous-time, continuous-state: Both the time and the state values are continuous. Brownian motion is a primary example of this type of process.