You can use Policy Summaries to understand what is going on in your Risk Profiles and Policy Trees. For example, assume you have a model that has a positive expected value, but that the Risk Profile shows a probability of losing money. By creating a Policy Summary Comparison from the Policy Tree (Policy | Policy Summary | Compare | Filter by Objective Function) you can examine the subset of optimal endpoints where the objective function (i.e., Profit) is below zero to see which events are contributing to the losses.
Or, assume you have a model with a downstream decision such as Enter New Market. In certain scenarios, the optimal policy is to enter the new market while for others it is not. Which uncertainties might be driving this? By using Policy Summary | Compare | Filter by Decision Alternative from the Policy Tree you can create a Policy Summary Comparison to examine the subset of optimal endpoints where the downstream decision alternative for Enter New Market is Yes to see which events are correlated with entering the new market.
Similarly you can create Policy Summary Comparisons that are filtered by a chance outcome by using Policy Summary | Compare | Filter by Chance Outcome from the Policy Tree.
Interpreting Policy Summary Comparisons
In a Policy Summary Comparison, two sets of bars are displayed on each event state (chance outcome or decision alternative). Blue indicates the policy dependent probabilities for the subset of endpoints that are part of the optimal policy for the full model. Red indicates the policy dependent probabilities for the subset of endpoints that are part of the optimal policy AND meet the filter criteria (e.g., also fall within the range of the objective function specified for the filter or also have the selected decision alternative/chance outcome).
- If the length of the red and blue bars are equal, the probability is unchanged and the chance outcome or decision alternative occurs with the exact same frequency in the set of optimal outcomes for the full model as in the set of filtered, optimal outcomes.
- If the blue bar is longer than the red bar, then the chance outcome / decision alternative occurs more often in the set of optimal endpoints for the full model than in the set of filtered, optimal endpoints .
- If the red bar is longer than the blue bar, then the chance outcome / decision alternative occurs less often in the set of optimal endpoints for the full model than in the set of filtered, optimal endpoints .
Events whose red and blue bars differ greatly in length are the interesting ones. These events have significantly different distributions in the filtered set of endpoints than in the full set of optimal endpoints. For example, if you are looking at scenarios where Profit is less than 0, event states with much longer red bars will be the key drivers behind losing money. Or if you are looking at all the optimal endpoints in which you Enter New Market, event states with much longer red bars will be those that make entering the new market a more attractive proposition.
Versions: DPL Professional, DPL Enterprise, DPL Portfolio
See Also