Pearson System Distribution Choice from Skewness and Kurtosis

Description

Place a sample's squared skewness and kurtosis on Pearson's chart to choose a density type, fit it from the moments, and use it for tail quantiles or simulation.

Pearson's system is a family of densities generated by one differential equation; it includes the normal, the beta (type I), the gamma (type III), the heavy-tailed type IV, the beta of the second kind and Student's t (type VII). Which member fits is decided by two numbers: beta1, the squared skewness, and beta2, the kurtosis. Plotting the sample point on the beta1-beta2 chart identifies the type, whose parameters are then fitted from the first four moments. Because the normal is the point (0, 3), the method returns the normal when the data are normal, which Vinod and Reagle describe as cheap insurance. The fitted density gives smoother tail quantiles than the empirical distribution and can feed simulations. Johnson's system plays the same role with transformations to normality.

Attributes

Loss type
Siloed
Effort
Hours

Steps

  1. Compute the sample's second, third and fourth central moments.
  2. Compute beta1 = m3 squared over m2 cubed and beta2 = m4 over m2 squared (optionally Pearson's kappa criterion).
  3. Locate the point on the beta1-beta2 chart to select the Pearson type.
  4. Fit that type's parameters from the moments using published formulas or software.
  5. Overlay the fitted and empirical densities, and run a goodness-of-fit test if the decision matters.
  6. Use the fitted density for tail quantiles, percentile targets or simulation inputs.

Sources

  • Preparing for the Worst: Incorporating Downside Risk in Stock Market Investments (Vinod & Reagle, 2005)

Also known as

  • Pearson curves
  • Pearson family of distributions
  • Beta1-beta2 chart
  • Skewness-kurtosis plot
  • Johnson system fitting
  • Choosing a distribution family from skewness and kurtosis (Pearson system fitting)

Categories

  • Analytics & Behaviour
  • Statistics & Modelling