Order Statistics: Min, Max, and Range for Actuarial Applications
Study order statistics and their role in actuarial modeling for Exam P.
Definition and Basic Formulas
Given a random sample X1, X2, ..., Xn from a continuous distribution with CDF F and PDF f, the order statistics X_(1) <= X_(2) <= ... <= X_(n) are the sorted values. X_(1) is the sample minimum and X_(n) is the sample maximum. Order statistics are fundamental in insurance for modeling extreme claims, estimating quantiles, and analyzing the range of losses.
The CDF of the k-th order statistic is F_{X_(k)}(x) = sum from j=k to n of C(n, j) * [F(x)]^j * [1 - F(x)]^(n-j). This is simply the probability that at least k of the n observations fall at or below x, which is a binomial probability with "success probability" F(x).
For the two most commonly tested order statistics, the formulas simplify considerably. For the minimum: F_{X_(1)}(x) = 1 - [1 - F(x)]^n, because the minimum exceeds x only when all n observations exceed x. For the maximum: F_{X_(n)}(x) = [F(x)]^n, because the maximum is at most x only when all n observations are at most x.
PDFs of Extremes
Differentiating the CDFs gives the PDFs. The PDF of the minimum is f_{X_(1)}(x) = n * [1 - F(x)]^(n-1) * f(x). The PDF of the maximum is f_{X_(n)}(x) = n * [F(x)]^(n-1) * f(x). For the general k-th order statistic:
f_{X_(k)}(x) = n! / ((k-1)! * (n-k)!) * [F(x)]^(k-1) * [1 - F(x)]^(n-k) * f(x).
This general formula has a natural interpretation. The factor [F(x)]^(k-1) accounts for k-1 observations below x, [1 - F(x)]^(n-k) accounts for n-k observations above x, f(x) is the density of the observation at exactly x, and the combinatorial factor counts the number of ways to assign the observations to these three groups.
A useful fact that appears frequently on Exam P: if X1, ..., Xn are independent Exponential(lambda), then X_(1) ~ Exponential(n * lambda). The proof uses the minimum CDF: P(X_(1) > x) = [e^(-lambda*x)]^n = e^(-n*lambda*x), which is the survival function of an Exponential(n*lambda). The mean of the minimum is therefore 1/(n*lambda), which is 1/n times the mean of a single observation.
Expected Values of Order Statistics
Computing E[X_(k)] requires integrating x * f_{X_(k)}(x) over the support, which can be complicated for general distributions. However, for the Uniform(0, 1) distribution, the formulas are elegant: E[X_(k)] = k/(n+1) and Var(X_(k)) = k(n-k+1)/((n+1)^2 * (n+2)).
For exponential order statistics, successive spacings provide a useful decomposition. Define D_i = X_(i) - X_(i-1) for i = 1, ..., n (with X_(0) = 0). Then D_i ~ Exponential((n-i+1)*lambda), and the spacings are independent. This means X_(k) = D_1 + D_2 + ... + D_k, so E[X_(k)] = sum from i=1 to k of 1/((n-i+1)*lambda) = (1/lambda) * sum from j=n-k+1 to n of 1/j.
For example, with n = 4 and lambda = 1: E[X_(1)] = 1/4, E[X_(2)] = 1/4 + 1/3 = 7/12, E[X_(3)] = 1/4 + 1/3 + 1/2 = 13/12, and E[X_(4)] = 1/4 + 1/3 + 1/2 + 1 = 25/12. These spacings results are tested on Exam P and provide an elegant way to compute moments of exponential order statistics.
The Sample Range
The range R = X_(n) - X_(1) measures the spread of the sample. For iid Uniform(0, 1), R has the Beta(n-1, 2) distribution (which can be derived using the joint density of the minimum and maximum), with E[R] = (n-1)/(n+1).
For exponential samples, the range R = X_(n) - X_(1) can be analyzed using the spacing decomposition. Since the spacings D_1, D_2, ..., D_n are independent and D_i ~ Exponential((n-i+1)*lambda), the range is R = D_2 + D_3 + ... + D_n (all spacings except the first). Each term is exponential with a different rate, so the distribution of R is a sum of independent exponentials, which does not simplify to a single named distribution unless n = 2.
Exam P may ask for E[R] or Var(R), which can be computed by summing the means and variances of the independent spacings. For exponential data with rate lambda and sample size n: E[R] = sum from i=2 to n of 1/((n-i+1)*lambda) = (1/lambda) * (1/(n-1) + 1/(n-2) + ... + 1).
Joint Distribution of Order Statistics
The joint density of all n order statistics is f_{X_(1),...,X_(n)}(x_1, ..., x_n) = n! * f(x_1) * f(x_2) * ... * f(x_n) for x_1 < x_2 < ... < x_n. The factor n! accounts for the number of permutations of the original sample that produce the same set of order statistics.
The joint density of two order statistics X_(i) and X_(j) with i < j is f_{X_(i),X_(j)}(x, y) = n!/((i-1)!(j-i-1)!(n-j)!) * [F(x)]^(i-1) * f(x) * [F(y)-F(x)]^(j-i-1) * f(y) * [1-F(y)]^(n-j) for x < y. This formula is useful for computing the joint distribution of the minimum and maximum (i=1, j=n) or the distribution of the range.
Actuarial Applications
Order statistics model extreme events in insurance: the largest claim in a portfolio X_(n) determines the maximum loss exposure, while X_(1) might represent the smallest deductible that would be triggered. The range X_(n) - X_(1) measures the dispersion of losses in a given period.
In reinsurance, excess-of-loss treaties are concerned with the largest claims. If an insurer retains the first d dollars of each claim, the largest loss to the reinsurer from n independent claims is max(X_1 - d, 0, ..., X_n - d, 0), which involves the order statistics of the original losses.
Order statistics also appear in non-parametric estimation. The sample median X_((n+1)/2) (for odd n) and sample quantiles are order statistics used to estimate population percentiles. These estimators are robust to outliers compared to the sample mean, which matters when dealing with heavy-tailed loss distributions.
Study Tips
For Exam P, memorize the CDF and PDF formulas for the minimum, maximum, and general k-th order statistic. Know the exponential minimum result (X_(1) ~ Exponential(n*lambda)) and the uniform order statistic expected values (E[X_(k)] = k/(n+1)). Practice the exponential spacings approach for computing moments. When tackling order statistics problems, always start by writing down the CDF of the parent distribution, as all order statistic formulas build on F(x) and f(x). These problems typically take three to four minutes on exam day, with the main challenge being careful algebraic manipulation.