Decision Analytics vs. Monte Carlo Simulation

A Clear Case for a Decision Analytic Approach

Critical business decisions deserve the most rigorous approach to analyzing potential risks and enumerating possible alternatives. While Monte Carlo simulation provides a quick method for assessing risk and helping the user make a go/no go decision, it does not provide a system for structuring a decision problem and mapping out alternatives.

The Limitations of Monte Carlo Simulation

Monte Carlo simulation is limited in that it requires a pre-specified decision rule of thumb be applied regardless of the situation. While this rule of thumb may make sense in a given scenario, there is no guarantee that it is appropriate across a range of scenarios. Monte Carlo takes a less rigorous approach than Decision Analysis which allows for a more comprehensive framework for structuring uncertainties, defining dependencies and explicitly considering alternatives in order to arrive at the most well conceived decision.

Below are some key reasons why a Decision Analytic Approach holds the advantage over Monte Carlo:

 DECISION ANALYSIS MONTE CARLO Advantage of DA DECISIONS Decisions are modelled explicitly (e.g., we can build, abandon, or wait) Most decisions are modelled as rules (e.g., =IF(A42>B42,C42,0)). It's about the decisions. UNCERTAINTIES Most uncertainties are modelled with discrete outcomes. Probabilities and values are assessed. Most uncertainties are modelled with common statistical distributions. Distribution parameters are assessed. Decision makers have more intuition for specific outcomes than for variance, skewness, etc. DEPENDENCIES Dependence between uncertainties is usually handled explicitly ("if we know natural gas prices are high the likelihood changes to ...") Dependence between uncertainties is usually handled with a correlation coefficient. (In practice, it is often ignored.) Even technical people have little intuition for correlation coefficients. VALUE OF INFORMATION The importance of an uncertainty (& the value of learning more about it) can be calculated and shown in a tree. Value of information can't be calculated directly (no explicit modelling of decisions). Value of information has relevance to real world actions (should we pay for a market study) and can be used to determine when to move forward. SOLUTION METHOD Results come from a decision policy tree, whose paths are specific scenarios that can be inspected. Results come from a number of equally likely random draws of the input distributions. Portions of the tree are meaningful and can build understanding. With Monte Carlo, each draw is independent, and the results don't become stable until there are a large number of draws (say 10,000). RESULTS A key result is a Policy Tree™ both part of the solution method and a "strategic road map" for making decisions and managing value going forward. The other main result is a Risk Profile showing the overall range of outcomes. Pretty much just an output distribution. An outcome probability distribution is usually not enough to give a decision maker confidence that a thorough analysis has been performed and a decision can be made.