Applies to: @RISK 5.5 and Up

 

Using RiskIndepC and RiskDepC vs. RiskCorrmat

Question:
I’m using RiskIndepC and RiskDepC functions to correlate inputs, but the manual recommends RiskCorrmat. Does it really matter?

Answer:

• Single Dependent Variable:
  If you have only one RiskDepC per RiskIndepC, it doesn’t matter. @RISK effectively creates a simple 2×2 correlation matrix for each pair.

• Multiple Dependent Variables:
  Problems can arise when multiple RiskDepC functions are associated with a single RiskIndepC. Here’s why:

  1. @RISK creates a separate correlation matrix for each RiskIndepC, with the first row for the independent variable and subsequent rows for the dependent variables sharing the same identifier.

  2. Correlations you specify go in the first column; correlations between RiskDepC variables default to zero, because the old scheme cannot assign them.

• Why This is a Problem:

  • You may unintentionally create a non-self-consistent matrix. For example, two dependent variables correlated 0.9 and 0.5 with the same independent variable cannot simultaneously have zero correlation with each other. @RISK will adjust the matrix, potentially altering your specified correlations.

  • Even if the matrix is self-consistent, zero correlations between dependent variables may be unrealistic, reducing model accuracy.

Recommended Solution: Use RiskCorrmat

• Starting with @RISK 5.5, RiskCorrmat allows you to:

  • Specify an adjustment weights matrix to preserve desired correlations as much as mathematically possible.

  • Automatically assign valid correlations between dependent variables.

Steps to Convert:

1. Construct the correlation matrix for each RiskIndepC variable.

2. Construct an adjustment weights matrix: use 100s in the first column, zeros elsewhere.

3. Use the RiskCorrectCorrmat array function to compute a self-consistent adjusted correlation matrix.

4. Replace RiskIndepC and RiskDepC with RiskCorrmat in formulas:

   • Argument 1 = former independent variable

   • Arguments 2…n = former dependent variables

5. Optionally, copy the adjusted matrix and Paste Special → Values to replace formulas with numbers, then delete the original matrix and adjustment weights matrix.

Result:
The simulated correlations will match your original specifications as closely as mathematically possible, while avoiding inconsistencies and unrealistic zero correlations.

 

Attachments:
KB1537_Removing_RiskDepC.xlsx