Applies to: @RISK (all versions supporting correlation matrices)

 

Question: How many input distributions can correlate? Is there a limit to the size of my correlation matrix?

Answer:

There is no fixed limit on the number of input distributions that can be correlated. However, as with all aspects of modeling, your available system resources (memory and processor) are a constraint.

Recommendations:

1. Partition large matrices:
   Avoid creating one enormous matrix if possible. For example, suppose you have 400 inputs that need to be correlated. A 400×400 matrix contains 160,000 cells. If those 400 inputs naturally fall into four groups of about 100 each, with correlations only within the groups, it’s better to use four 100×100 matrices (total 40,000 cells).

   • @RISK can validate smaller matrices much faster than a single large matrix.

   • If all 400 variables truly need to correlate with each other, then a single 400×400 matrix is necessary.

2. Reuse correlation matrices for similar groups:
   If you have several groups of variables that require the same correlations within each group, you can have each group refer to the same smaller matrix.

   • Typical examples:

     • Multiple time periods with the same correlations within each period, but no correlations across periods.

     • Geographical regions where factors are correlated in the same way within a region but not between regions.

Following these techniques reduces computational load and helps @RISK process correlation matrices efficiently.

References / Related Topics:

• Correlation Matrix Exceeds Excel's Column Limit

• Same Correlation Coefficients for Several Groups of Inputs

Related Articles:

 Correlation Matrix Exceeds Excel's Column Limit

Same Correlation Coefficients for Several Groups of Inputs