Stochastic Processes for Actuarial Risk Modeling
Study stochastic processes including Poisson processes and Brownian motion for Exam MAS-I.
What Are Stochastic Processes
A stochastic process is a collection of random variables {X(t), t in T} indexed by a parameter t, usually representing time. In actuarial applications, stochastic processes model the evolution of uncertain quantities over time: the number of claims received, the insurer's surplus, investment returns, or the state of a policyholder's health. Understanding stochastic processes provides the mathematical foundation for dynamic risk modeling.
Stochastic processes are classified by whether the time parameter is continuous or discrete, and whether the state space (the set of possible values) is continuous or discrete. A random walk is a discrete-time, discrete-state process. A Poisson process is a continuous-time, discrete-state process. Brownian motion is a continuous-time, continuous-state process.
The Homogeneous Poisson Process
The homogeneous Poisson process is the most fundamental model for claim arrivals. It models events occurring randomly over time at a constant average rate lambda. The process N(t) counts the number of events in the interval [0, t]. Its defining properties are: (1) N(0) = 0, (2) the process has independent increments (counts in non-overlapping intervals are independent), and (3) the process has stationary increments (the distribution of N(t+s) - N(t) depends only on s, not on t).
Key distributional results follow from these properties. The count in any interval of length t follows a Poisson distribution: N(t) ~ Poisson(lambda * t). The times between successive events (interarrival times) are independent and identically distributed Exponential(lambda) random variables. The waiting time until the k-th event follows a Gamma(k, lambda) distribution. These results are heavily tested on Exam MAS-I.
Properties of the Poisson Process
Several important properties of the Poisson process have actuarial applications. The superposition property states that the sum of independent Poisson processes with rates lambda_1 and lambda_2 is itself a Poisson process with rate lambda_1 + lambda_2. This means that if claims from two independent sources arrive as Poisson processes, the combined claim arrival process is also Poisson.
The thinning property states that if each event in a Poisson process with rate lambda is independently classified as type A with probability p and type B with probability 1-p, then the type A events form a Poisson process with rate p*lambda and the type B events form a Poisson process with rate (1-p)*lambda, and the two are independent. This is useful for modeling different claim types (large vs. small, or by cause of loss).
Nonhomogeneous Poisson Processes
The nonhomogeneous Poisson process allows the arrival rate to vary with time: lambda(t) can change over the day, season, or year. The count in the interval (s, t) follows Poisson(integral from s to t of lambda(u) du). Interarrival times are no longer identically distributed; the memoryless property of the exponential no longer applies to the interarrival times directly.
In insurance, nonhomogeneous Poisson processes model seasonal patterns in claim arrivals (more auto claims in winter, more property claims during hurricane season), time-of-day patterns (more accidents during rush hours), or growth trends in claim frequency over calendar years. The integral of the rate function, Lambda(t) = integral from 0 to t of lambda(u) du, called the intensity function, transforms the nonhomogeneous process into a homogeneous one through a time change.
The Compound Poisson Process
The compound Poisson process models aggregate losses by combining a Poisson process for claim counts with random claim sizes. If N(t) is a Poisson process with rate lambda and X_1, X_2, ... are i.i.d. claim sizes independent of N(t), then S(t) = X_1 + X_2 + ... + X_{N(t)} is a compound Poisson process. The mean aggregate loss is E[S(t)] = lambda * t * E[X] and the variance is Var(S(t)) = lambda * t * E[X^2].
This is the continuous-time analog of the collective risk model studied in ratemaking. The moment generating function is M_{S(t)}(r) = exp(lambda * t * (M_X(r) - 1)), where M_X(r) is the moment generating function of the individual claim size. This formula is useful for computing tail probabilities and risk measures of the aggregate loss distribution.
Brownian Motion
Brownian motion (also called the Wiener process) W(t) is a continuous-time process with continuous sample paths and normally distributed increments. Its defining properties are: (1) W(0) = 0, (2) it has independent increments, (3) W(t) - W(s) ~ N(0, t - s) for t > s, and (4) sample paths are continuous. Brownian motion models the continuous random fluctuations in financial quantities like investment returns or insurance surplus.
Brownian motion with drift is U(t) = mu * t + sigma * W(t), which has E[U(t)] = mu * t and Var(U(t)) = sigma^2 * t. In the surplus model for an insurer, U(t) = u + c * t - S(t), where u is the initial surplus, c is the premium rate per unit time, and S(t) is the aggregate claims process. When S(t) is approximated by a Brownian motion with drift, the probability of ruin (the surplus becoming negative) can be computed in closed form.
Ruin Theory Basics
Ruin theory studies the probability that an insurer's surplus falls below zero. For the Brownian motion surplus model U(t) = u + mu * t + sigma * W(t) with positive drift mu (premiums exceed expected claims), the probability of eventual ruin is psi(u) = exp(-2 * mu * u / sigma^2). This shows that ruin probability decreases exponentially with initial surplus u and with the safety loading (the excess of premiums over expected claims).
For the compound Poisson surplus model, exact ruin probabilities are available only for certain claim size distributions (such as exponential). The Cramer-Lundberg approximation gives psi(u) approximately C * exp(-R * u), where R is the adjustment coefficient found by solving lambda * (M_X(R) - 1) = c * R. The adjustment coefficient R plays a central role in ruin theory and risk management.
Common Pitfalls
A frequent mistake is confusing the rate parameter lambda with the mean of the Poisson distribution. For a Poisson process, N(t) ~ Poisson(lambda * t), not Poisson(lambda). The rate lambda is per unit time; the mean count over an interval of length t is lambda * t. Another error is applying the memoryless property of exponential interarrival times to non-Poisson processes, where it does not hold.
With compound Poisson processes, students sometimes use Var(S) = lambda * t * Var(X) instead of the correct formula Var(S) = lambda * t * E[X^2]. The difference is that E[X^2] = Var(X) + (E[X])^2, so using Var(X) instead of E[X^2] omits the (E[X])^2 term.
Study Tips
Exam MAS-I covers Poisson process properties, compound Poisson moments, and basic Brownian motion concepts. Memorize the key distributional results for Poisson processes: Poisson counts, exponential interarrival times, and Gamma waiting times. Know the superposition and thinning properties and how they apply to insurance scenarios. Practice computing E[S(t)] and Var(S(t)) for compound Poisson processes with various claim size distributions. Understand the surplus model and the concept of ruin probability, even if detailed ruin theory calculations are not heavily tested at this level.