How @RISK Computes Spearman Rank-Order Correlations with RiskCorrel()

Applies to: @RISK, all releases

Question:
How does the RiskCorrel() function compute the Spearman rank-order correlation coefficient?

Method

1. Ranking values:

   • Each number in the simulated distributions is replaced by its rank (1 to N, where N = number of iterations).

   • For continuous distributions, all ranks are distinct.

   • For discrete distributions with tied values, the midrank method is used: tied values receive the average of the ranks they would have had if slightly different.

2. Correlation calculation:

   • Once ranks are assigned, the Pearson correlation coefficient of the ranks is computed.

   • This is the Spearman rank-order correlation coefficient.

Usage in @RISK

• RiskCorrel() can return either Pearson or Spearman correlations.

• Spearman rank-order correlations are used both after a simulation and within a simulation when drawing numbers for correlated distributions.

Reference:
Press, W. H., Flannery, B. P., Teukolsky, S. A., & Vetterling, W. T. (1986). Numerical Recipes, Cambridge University Press, pp. 488+.

 

Related Articles:

How @RISK Correlates Inputs