Why Random Number Sequences Differ Between Simulations

 

Problem:
When comparing random number sequences for the same model, differences can appear in these situations:

• Simulating the model with @RISK vs. a Custom Development API run with DCR or SDK.

• Simultaneously simulating the same model in two open workbooks.

 

Explanation:
Differences occur because distributions may be defined differently:

• Number of distributions: Adding or removing distributions changes how the seeded list is sampled.

• Order of distributions: Changing the order alters which numbers are assigned to which distribution.

 

Example 1 – Single distribution

Iteration	Value of RiskUniform(0,1)
1	0.1
2	0.2
3	0.3
4	0.4

 

Example 2 – Two distributions

Iteration	First RiskUniform(0,1)	Second RiskUniform(0,1)
1	0.1	0.2
2	0.3	0.4
3	0.5	0.6
4	0.7	[next value in seed]

 

Example 3 – Different order with RiskPoisson(5) added

Order A:

Iteration	First RiskUniform(0,1)	Second RiskUniform(0,1)	RiskPoisson(5)
1	0.1	0.2	7.0
2	0.4	0.5	3.0
3	0.7	[next value]	[next value]

Order B:

Iteration	First RiskUniform(0,1)	RiskPoisson(5)	Second RiskUniform(0,1)
1	0.1	4.0	0.3
2	0.4	1.0	0.6
3	0.7	[next value]	[next value]

• The same fixed seed always produces the same underlying random numbers, but their assignment changes with the number or order of distributions.

• All models ultimately converge to the same statistical distributions, making the results correct and comparable.

• Latin Hypercube sampling can also produce differences because each bin must be sampled, so sequences may not match exactly.

 

Solutions if identical results are critical:

Approach	Description
Monte Carlo	Use Monte Carlo instead of Latin Hypercube to preserve sequence identity.
Supply variable data	Provide pre-defined sequences (e.g., monthly interest rates).
RiskSeed property	Assign a unique RiskSeed to each distribution so sequences are independent of order.