To run a Monte Carlo simulation, you must have at least one continuous chance node in your model. Once you've introduced a continuous event you'll notice that the default evaluation method indicated within the top half of the the Decision Analysis split button within the Home | Run group will update to Monte Carlo Simulation. In fact, if your model has one or more continuous chance nodes, it is the only evaluation method available. To run the simulation click Home | Run | Decision Analysis or press F10 to run a Monte Carlo simulation on the active model in your workspace. DPL will check the model for correctness and consistency. Many of the distribution and policy outputs within the Home | Run group can be generated with a Monte Carlo Simulation run. For a description of each output option see Decision Analysis Options.
Note: If you want to run a Monte Carlo simulation but all the chance nodes in your model are discrete, choose Hybrid Discrete Tree Simulation.
Number of Monte Carlo samples
You can set the number of random samples drawn for your run by either choosing a preset number of samples from the two drop down boxes under Monte Carlo samples within the Home | Run Group on the ribbon or by choosing your own number of samples (between 1 and 2,000,000,000) within the Run Settings dialog (Home | Run | Options).
Initial This is the number of random samples you want for your Monte Carlo simulation. A sample is a random draw from each of the chance nodes in the model. You will want to have enough samples so that the results are stable and do not vary significantly from run to run.
Minimum at decisions (referred to as Restart on the ribbon) When DPL encounters a decision, it needs to determine which alternative is best. For an initial decision (a decision at the far left of the tree), it samples each alternative by the initial number of samples. When DPL encounters a decision deep down in the tree, it has to do more than send one or a few samples down each branch to see which one is best. The "Minimum at decisions" parameter tells DPL how many samples it needs to send down each branch to ensure the decision is simulated with a reasonable degree of accuracy.
Distribution Outputs
Typically the result of most interest for a Monte Carlo simulation is a probability distribution that displays the range of uncertainty for a given metric. During a Monte Carlo Simulation run the continuous chance nodes(s) are simulated by drawing random samples from named distributions with specified parameters. This produces approximate expected values and Risk Profile percentiles. Risk profiles are displayed in frequency histogram format by default. This can be changed to a cumulative display within the Chart | Format | Display group.
Policy Outputs
Note that for Monte Carlo simulation, the Policy Tree is mainly used with small samples to gain intuition into how DPL evaluates the model, particularly when downstream decisions are involved. The first node in the Policy Tree has as many branches as the initial samples parameter. DPL will warn you if you ask for a large Policy Tree.
If you request a Policy Summary and/or Policy Tree, use the Levels drop-down to indicate the number of levels you wish to save in the policy data. For Policy Summaries, usually with Monte Carlo simulation you will only need to save the number of levels to reach the "deepest" decision.
Versions: DPL Professional, DPL Enterprise, DPL Portfolio
See Also