Applies to:
@RISK 6.x and Up

Why Some Outputs Converge Slower Than Others in @RISK

When using convergence monitoring with the simulated mean, it’s common for outputs with very small means to converge slower than outputs with larger means, even if their standard deviations are similar.

Convergence in @RISK checks that the simulation has reached a specified confidence level (default: 95%) that the output mean is within a tolerance band (default: ±3% of the true mean).

• Example:

  • Output A mean = 200,000,000 → ±3% band = 12,000,000 units wide

  • Output B mean = 10 → ±3% band = 0.6 units wide

Output B takes more iterations to converge because the algorithm must narrow the estimate into a much tighter range relative to its mean.

Even moderately smaller means can show this effect. In the example, several outputs with means of $1,700–$6,200 converged, but an output with a mean of −$16 did not, despite similar standard deviations.

Key Takeaway:

Outputs with small means relative to their standard deviations may converge more slowly or appear unconverged in @RISK. This is normal behavior.

 

Related Article:

Convergence Monitoring in @RISK.

Attachments:
KB1740 Credit Risk 2 with Convergence Testing.xlsx