What is the Generalized Pareto Distribution?
The Generalized Pareto Distribution (GPD) is a continuous probability distribution widely used in extreme value theory to model the tails of distributions, exceedances over a threshold, and heavy-tailed phenomena in finance, hydrology, and reliability engineering. It is described by three parameters: a location parameter mu, a scale parameter sigma (which must be positive), and a shape parameter xi that controls the heaviness of the tail. This is a pure mathematical tool with no regional or jurisdictional scope.
How to use this calculator
Choose the function you want: the probability density (PDF), the lower cumulative distribution (CDF), or the upper cumulative survival function. Enter the three parameters mu, sigma and xi. Then define the x sequence with an initial value, a step increment, and the number of points to evaluate. The calculator produces a table of (x, y) pairs and a line graph, plus the single function value at the first x for quick reference.
The formula explained
Let \(B = 1 + \xi\,\frac{x - \mu}{\sigma}\). When \(\xi\) is not zero, the density is $$f(x) = \frac{1}{\sigma}\,B^{-\frac{1}{\xi} - 1},$$ the CDF is $$P(x) = 1 - B^{-\frac{1}{\xi}},$$ and the survival function is $$Q(x) = B^{-\frac{1}{\xi}} = 1 - P.$$ When \(\xi\) equals zero, the distribution degenerates to the exponential form: $$f(x) = \frac{1}{\sigma}\,e^{-\frac{x-\mu}{\sigma}} \quad\text{and}\quad P(x) = 1 - e^{-\frac{x-\mu}{\sigma}}.$$ The support is \(x \ge \mu\) when \(\xi \ge 0\), and \(\mu \le x \le \mu - \frac{\sigma}{\xi}\) when \(\xi < 0\). Outside the support the density is 0, while \(P\) and \(Q\) clamp to their boundary values.
Worked example
Take the CDF with \(\mu = 1\), \(\sigma = 1\), \(\xi = 1\) at \(x = 2\). Then $$B = 1 + 1\cdot\frac{2-1}{1} = 2,$$ so $$P = 1 - 2^{-1} = 0.5.$$ The PDF at the same point is $$2^{-2} = 0.25,$$ and the survival function is $$Q = 2^{-1} = 0.5,$$ confirming \(P + Q = 1\).
How the Shape Parameter ξ Changes the Distribution
The table fixes \(\mu = 0\) and \(\sigma = 1\) and varies only the shape \(\xi\). With \(z = x\), the formulas are \(f(x) = (1+\xi x)^{-1/\xi - 1}\), survival \(Q(x) = (1+\xi x)^{-1/\xi}\) and \(P(x) = 1 - Q(x)\). For \(\xi = 0\) the limiting forms are \(f(x) = e^{-x}\) and \(Q(x) = e^{-x}\).
| ξ | Support | Tail type | x | PDF f(x) | CDF P(x) | Survival Q(x) |
|---|---|---|---|---|---|---|
| −0.5 | 0 ≤ x ≤ 2 (upper bound \(\mu - \sigma/\xi = 2\)) | Bounded (short tail) | 0.5 | 0.75 | 0.4375 | 0.5625 |
| 1 | 0.5 | 0.75 | 0.25 | |||
| 2 (endpoint) | 0 | 1 | 0 | |||
| 0 | x ≥ 0 | Exponential | 0.5 | 0.6065 | 0.3935 | 0.6065 |
| 1 | 0.3679 | 0.6321 | 0.3679 | |||
| 2 | 0.1353 | 0.8647 | 0.1353 | |||
| 0.5 | x ≥ 0 | Heavy (power-law) | 0.5 | 0.512 | 0.36 | 0.64 |
| 1 | 0.2963 | 0.5556 | 0.4444 | |||
| 2 | 0.125 | 0.75 | 0.25 |
Worked check for \(\xi = -0.5\), \(x = 0.5\): \(1 + \xi x = 1 - 0.25 = 0.75\). The exponent is \(-1/\xi = 2\), so \(Q = 0.75^{2} = 0.5625\) and \(f = 0.75^{1} = 0.75\).
Worked check for \(\xi = 0.5\), \(x = 1\): \(1 + \xi x = 1.5\). Then \(Q = 1.5^{-2} = 0.4444\) and \(f = 1.5^{-3} = 0.2963\).
Reading across the rows, the survival at \(x = 2\) is 0 for \(\xi = -0.5\), 0.1353 for \(\xi = 0\) and 0.25 for \(\xi = 0.5\). A larger \(\xi\) leaves more probability in the upper tail.
FAQ
Why must sigma be positive? Sigma is a scale parameter that divides several terms; a non-positive value is mathematically undefined, so the tool guards against it.
What happens when xi = 0? The GPD becomes the exponential distribution. The calculator automatically switches to the exponential formulas when \(|\xi|\) is below a tiny epsilon to avoid dividing by zero.
Can I scan x downward? Yes. Use a negative step increment to evaluate x values in descending order.