Gamma Distribution: Shape, Rate, and Connection to Poisson
Study the gamma distribution, its parameters, and its connection to the Poisson process.
Definition and Parameters
If X ~ Gamma(alpha, lambda), the PDF is f(x) = (lambda^alpha / Gamma(alpha)) * x^(alpha-1) * e^(-lambda*x) for x > 0. Here alpha is the shape parameter and lambda is the rate parameter. The mean is alpha/lambda and the variance is alpha/lambda^2. The gamma function Gamma(alpha) = integral of t^(alpha-1) * e^(-t) dt from 0 to infinity, with Gamma(n) = (n-1)! for positive integers.
Some sources use the scale parameterization with theta = 1/lambda, giving mean alpha*theta and variance alpha*theta^2. Always check which form a problem uses.
The gamma function has several useful properties. Gamma(1) = 1, Gamma(1/2) = sqrt(pi), and Gamma(alpha) = (alpha - 1) * Gamma(alpha - 1) for alpha > 1. This recursion means that for positive integers, Gamma(n) = (n-1)!. For half-integer values, you can build on Gamma(1/2) = sqrt(pi): for example, Gamma(3/2) = (1/2) * Gamma(1/2) = sqrt(pi)/2 and Gamma(5/2) = (3/2) * Gamma(3/2) = 3*sqrt(pi)/4.
The MGF of the gamma distribution is M(t) = (lambda / (lambda - t))^alpha for t < lambda. This generalizes the exponential MGF, which has alpha = 1. Recognizing the gamma MGF on Exam P allows you to identify the distribution and read off both parameters.
Special Cases
The exponential distribution is Gamma(1, lambda). The chi-squared distribution with k degrees of freedom is Gamma(k/2, 1/2). When alpha is a positive integer, the gamma distribution is also called the Erlang distribution. These connections are tested on Exam P.
The sum of n independent Exponential(lambda) random variables has a Gamma(n, lambda) distribution. More generally, if X ~ Gamma(alpha1, lambda) and Y ~ Gamma(alpha2, lambda) are independent (same rate), then X + Y ~ Gamma(alpha1 + alpha2, lambda).
The shape parameter alpha controls the shape of the PDF. When alpha < 1, the PDF is decreasing and unbounded near zero. When alpha = 1, the PDF is a decreasing exponential. When alpha > 1, the PDF has a mode at (alpha - 1)/lambda and is bell-shaped but right-skewed. As alpha increases, the gamma distribution becomes more symmetric and approaches a normal distribution (by the CLT, since it is a sum of exponentials).
Connection to the Poisson Process
In a Poisson process with rate lambda, the waiting time until the alpha-th event follows a Gamma(alpha, lambda) distribution. This connection is useful on Exam P: the probability that the alpha-th event occurs after time t equals P(Gamma > t) = P(Poisson count in [0,t] < alpha). This relationship lets you convert between gamma probabilities and Poisson sums, which can simplify calculations when alpha is a small integer.
Example: Claims arrive according to a Poisson process with rate 2 per month. Find the probability that the 3rd claim arrives after month 2. The waiting time for the 3rd claim is Gamma(3, 2). P(T > 2) = P(Poisson count in [0, 2] < 3) = P(N < 3) where N ~ Poisson(4). This equals P(N = 0) + P(N = 1) + P(N = 2) = e^(-4)(1 + 4 + 8) = 13 * e^(-4) = 0.2381.
This Poisson-gamma duality is one of the most powerful computational shortcuts on Exam P. Instead of integrating the gamma PDF (which requires the incomplete gamma function), you simply sum a few Poisson probabilities. This works whenever the shape parameter alpha is a positive integer.
Scaling and Transformation Properties
If X ~ Gamma(alpha, lambda), then cX ~ Gamma(alpha, lambda/c) for c > 0. The shape parameter stays the same, but the rate parameter scales inversely with c. In the scale parameterization, cX ~ Gamma(alpha, c*theta), so the scale parameter scales directly with c.
If X ~ Gamma(alpha, lambda), then 2*lambda*X ~ Chi-squared(2*alpha) when 2*alpha is a positive integer. This connection to the chi-squared distribution is occasionally useful for looking up probabilities in chi-squared tables, though Exam P rarely requires this approach.
The ratio of independent gamma variables with the same rate has a Beta distribution: if X ~ Gamma(alpha, lambda) and Y ~ Gamma(beta, lambda) independently, then X / (X + Y) ~ Beta(alpha, beta). This result connects the gamma and beta families and appears in some Exam P problems.
Insurance Applications
The gamma distribution models aggregate claim amounts in some actuarial applications. If individual claims follow an exponential distribution and a policyholder submits a Poisson-distributed number of claims, the total loss can be related to gamma distributions in certain cases.
The gamma distribution also serves as a mixing distribution in Bayesian models. For example, if individual claim frequency follows Poisson(lambda) and lambda varies across the population according to a gamma distribution, the resulting marginal distribution of claims is negative binomial. This Poisson-gamma mixture is a cornerstone of credibility theory, which you will study in more advanced actuarial exams.
In the context of Exam P, gamma distribution problems typically involve computing expected values, variances, or probabilities using the Poisson connection. You may also be asked to identify a gamma distribution from its MGF or to determine the distribution of a sum of independent exponential variables.
Common Pitfalls
The biggest source of confusion is the parameterization. Some problems use the rate parameter lambda, while others use the scale parameter theta = 1/lambda. If you use the wrong formulas, your mean and variance will be incorrect. Before starting any gamma problem, identify which parameterization is being used by checking whether the given parameter appears in the exponent as e^(-lambda*x) (rate) or e^(-x/theta) (scale).
Another common mistake is trying to integrate the gamma PDF directly to find probabilities. For non-integer alpha, the integral involves the incomplete gamma function, which cannot be evaluated by hand. Instead, use the Poisson connection (for integer alpha) or the MGF to identify the distribution. On Exam P, if you find yourself trying to integrate a gamma PDF, look for a simpler approach.
Study Tips
Memorize the PDF, mean, variance, and MGF of the gamma distribution in both parameterizations. Know the special cases (exponential, chi-squared, Erlang) and the reproductive property (sum of independent gammas with the same rate is gamma). Practice the Poisson-gamma conversion, as this is the most commonly tested aspect of the gamma distribution on Exam P. Work through at least five problems that require converting gamma tail probabilities to Poisson sums. On exam day, recognizing the gamma distribution quickly from its PDF or MGF form can save you significant time.