Risk & Modeling

Fernandez-Steel Distribution

TL;DR

The Fernandez-Steel transformation adds asymmetry (skewness) to the Student's t-distribution, allowing one tail to be heavier than the other. This models the real-world observation that market crashes are sharper and more extreme than rallies of equal magnitude.

The Fernandez-Steel distribution is a method for introducing skewness into symmetric distributions like the Student-t. Developed by Fernandez and Steel (1998), it applies a skewness parameter (gamma) that stretches one tail relative to the other, breaking the symmetry while preserving the fat-tail properties controlled by degrees of freedom.

How It Works

The transformation works by scaling the distribution differently on each side of the mode:

  • Returns below the mode are scaled by gamma
  • Returns above the mode are scaled by 1/gamma
  • When gamma = 1: the distribution is symmetric (standard Student-t)
  • When gamma < 1: negative skewness — the left tail (losses) is stretched, making large losses more probable
  • When gamma > 1: positive skewness — the right tail (gains) is stretched

This elegant approach preserves the kurtosis properties of the underlying Student-t while adding a single interpretable parameter for asymmetry.

Why It Matters for Retirement Planning

Equity markets exhibit both fat tails and negative skewness — two properties that a normal distribution cannot capture. The Fernandez-Steel skewed Student-t handles both:

  • Degrees of freedom controls tail thickness (kurtosis)
  • Gamma controls asymmetry (skewness)

This combination is critical for retirement simulation because retirees are asymmetrically exposed to downside risk. A model that correctly weights the probability of sharp downturns — as the Fernandez-Steel distribution does — produces more realistic success rate estimates than symmetric alternatives.

Retirement Lab uses this distribution as its pro-tier fat-tail model, generating skewed Student-t random variables through the Marsaglia Polar Method and Marsaglia-Tsang gamma sampling.

Frequently Asked Questions

What does the gamma parameter do in the Fernandez-Steel distribution?
Gamma controls asymmetry. When gamma = 1, the distribution is symmetric. When gamma < 1, the left tail is stretched (negative skewness — crashes are more extreme). When gamma > 1, the right tail is stretched (positive skewness). For equity returns, gamma is typically set below 1 to capture the empirical observation that market downturns are sharper than upswings.
Why not just use a normal distribution with adjusted parameters?
A normal distribution cannot produce fat tails or skewness regardless of its parameters — it is always symmetric with thin tails. The Fernandez-Steel skewed Student-t captures both excess kurtosis (via degrees of freedom) and asymmetry (via gamma) in a single, mathematically coherent model.