About Monte Carlo Simulation (MCS)

First-things-first; there is no Monte Carlo QRA method. There are only models of risk behavior to which Monte Carlo simulation is applied. Why is this important to understand? Because the most common QRA method in use called line-item ranging (LIR) with MCS, mislabeled as “the Monte Carlo method”, does not work (research shows LIR is a disaster if there are systemic risks). And because of that failure MCS has earned a bad reputation (voodoo) among some. The reason LIR fails is that the underlying model is not a model of risk behavior; it is a cost estimate. Or activities in a schedule. It is not risk driven.

The second thing to know is that there are two methods of probabilistic QRA; regression and MCS. They are similar. Regression uses a dataset of actual project data to develop a parametric risk model directly. MCS on the other hand creates a dataset of simulated data (the many iterations) built upon a hypothesized risk model (e.g., A CPM or Expected Value model).

To do QRA, one must use one, and optimally, both of these methods. Regression works best with systemic risks data which are consistent in nature and impact across project systems from industry to industry. MCS works best with risk events which are unique to a project. ValidRisk uses both parametric modeling for systemic risk, and expected value with MCS for risk events. The MCS capability is built into ValidRisk without the need for any add-ons.

In ValidRisk, the potential intricacies of MCS are minimized. For one thing, the method is designed to minimize the “correlation problem” of MCS. Systemic risks are independent of the risk events, and the few critical risk events are typically independent of each other (but correlation can be defined). The other MCS intricacy is the selection of “distributions” to apply to a variable. ValidRisk uses the 3-point triangular distribution for schedule and cost impacts. First of all, as a 3-point distribution, it is easy for a team to define. But also, ValidRisk partner experience shows that because of the primacy of systemic risks, and the fact that one does not know which critical risk(s) events might occur (among other stochastic assumptions), dwelling on exact distributions is not a value adding exercise. The team is better off focusing on cost/schedule trading, considering secondary risk impacts, planning risk responses and so on. In the end, every research study finds cost growth and schedule slip outcomes fit a lognormal distribution, and that will be true for the outputs of ValidRisk.