Pareto Type I: Support, Quantiles, and Tail Probabilities

Introduction

The Pareto Type I distribution describes positive quantities whose possible values begin at a fixed minimum and whose upper tail decreases according to a power law. It is useful when the question is not only how likely a particular value is, but how likely an observation is to exceed a high threshold. Examples include modeling large losses, unusually large claims, or other positive measurements with a long upper tail.

Two parameters determine the law: a minimum value \(x_m\), which sets the support and measurement scale, and a positive tail index \(\alpha\), which controls how quickly exceedance probabilities decline. This tutorial derives the distribution function, density, quantiles, and tail probabilities, then applies them to threshold-risk calculations. It does not require calculating moments.

Definition and Support

The support says that values below \(x_m\) are impossible, while values at or above it are possible. The distribution has no point mass at the boundary: substituting \(x=x_m\) gives \(F(x_m)=0\). The parameter \(x_m\) has the same units as \(X\); the index \(\alpha\) is dimensionless. Larger \(\alpha\) means faster tail decay when \(x_m\) is held fixed.

Deriving the Density and Survival Function

For \(x>x_m\), differentiate the CDF. On the region below \(x_m\), the density is zero. The result is the Pareto Type I density.

The density formula is consistent with the CDF because, for \(x>x_m\), $$ \frac{d}{dx}\left[1-x_m^\alpha x^{-\alpha}\right] =\alpha x_m^\alpha x^{-(\alpha+1)}. $$ At the single endpoint \(x_m\), the value assigned to a density does not affect probabilities; the displayed convention gives the right-hand formula there. To check that the density integrates to one, use \(\alpha>0\):

The survival function, or upper-tail probability, is \(P(X>x)=1-F(x)\). Therefore,

At \(x=x_m\), this probability is one: \(X\) is at least \(x_m\), and it has no atom at \(x_m\). Since the distribution is continuous, \(P(X>x)=P(X\ge x)\) for every fixed \(x\). This equality is useful in threshold calculations, but it would not hold automatically for a distribution with point masses.

Quantiles and Upper-Tail Risk Thresholds

A quantile reverses the CDF: it gives the value below which a specified fraction of the distribution lies. For \(0<p<1\), define the \(p\)-quantile by \(q_p=\inf\{x:F(x)\ge p\}\). Because the Pareto CDF is continuous and strictly increasing on \([x_m,\infty)\), this quantile is the unique value satisfying \(F(q_p)=p\).

To obtain the formula, solve the CDF equation: $$ p=1-\left(\frac{x_m}{q_p}\right)^\alpha \quad\Longrightarrow\quad \left(\frac{x_m}{q_p}\right)^\alpha=1-p \quad\Longrightarrow\quad q_p=x_m(1-p)^{-1/\alpha}. $$ The solution is at least \(x_m\), since \(0<1-p<1\) implies \((1-p)^{-1/\alpha}>1\).

At the boundary, the quantiles approach \(x_m\) as \(p\) decreases to zero, and increase without bound as \(p\) approaches one. Thus it is convenient to use \(q_0=x_m\) as the lower endpoint by continuous extension; there is no finite \(q_1\). For a high confidence level \(p\), the formula also gives a risk threshold: the probability of exceeding \(q_p\) is \(1-p\). For instance, \(q_{0.99}\) is the threshold exceeded with probability \(0.01\).

Exceedance Probabilities and Conditional Tails

For a threshold \(t\), the probability of exceeding it is particularly simple when \(t\) is within the support. If \(t\ge x_m\), it is \((x_m/t)^\alpha\); if \(t<x_m\), exceedance is certain. This makes it possible to answer either of two common questions directly: “What is the chance of exceeding this value?” and “How high must the threshold be to make exceedance this unlikely?”

The same survival formula gives a conditional exceedance probability. For \(t\ge u\ge x_m\), divide the probability of exceeding \(t\) by the probability of exceeding \(u\): $$ P(X>t\mid X>u) =\frac{P(X>t)}{P(X>u)} =\frac{(x_m/t)^\alpha}{(x_m/u)^\alpha} =\left(\frac{u}{t}\right)^\alpha. $$ The denominator is positive because \(P(X>u)=(x_m/u)^\alpha>0\). This calculation describes the chance of going past a higher threshold among observations already known to exceed \(u\).

How the Parameters Shape Tail Calculations

The formulas clarify the separate roles of the two parameters. Multiplying \(x_m\) by a positive constant multiplies every quantile and threshold by that same constant; it changes the scale without changing the tail index. Holding the threshold as a multiple of the minimum, say \(t=cx_m\) for \(c\ge1\), gives $$ P(X>cx_m)=c^{-\alpha}. $$ For fixed \(c>1\), increasing \(\alpha\) decreases this probability because \(c^{-\alpha}\) becomes smaller. At a fixed percentile \(p\), the multiplier from the minimum is \((1-p)^{-1/\alpha}\), which likewise decreases as \(\alpha\) increases.

These statements concern tail probabilities and quantiles only. They do not imply that all summaries of the distribution exist or are finite; moment existence requires a separate calculation. Keeping those questions distinct prevents a valid threshold calculation from being mistaken for a statement about an average or another moment.

Common Mistakes

• Using the wrong tail expression. The CDF is \(1-(x_m/x)^\alpha\), while the exceedance probability is \((x_m/x)^\alpha\) for \(x\ge x_m\). Check whether the question asks for probability below or above the threshold.
• Reversing the quantile exponent. Solving the CDF gives \(q_p=x_m(1-p)^{-1/\alpha}\), not \(x_m(1-p)^{1/\alpha}\). A high percentile must exceed \(x_m\), which the correct expression guarantees.
• Forgetting the support case. The power formula for the survival probability applies when \(t\ge x_m\). For \(t<x_m\), \(P(X>t)=1\), since all possible values exceed \(t\).
• Confusing percent and probability. A 1% exceedance probability is \(0.01\), so its threshold is \(q_{0.99}\), not \(q_{0.01}\).
• Confusing conditional and unconditional risk. \(P(X>t\mid X>u)\) is the fraction exceeding \(t\) among observations already exceeding \(u\); it is not the unconditional probability \(P(X>t)\).