Scenario simulations do not yield results stochastically consistent with alternative Monte Carlo results: U.S. nuclear plant decommissioning funding adequacy (2000)

For policymakers, the conventional wisdom has often been that policy analysis and recommendations derived from scenario simulation are preferable to those derived from stochastic (Monte Carlo) risk simulation because scenario analysis yields discrete value outputs that are considered of more practical use than are the probabilistic ranges of results from Monte Carlo analysis. This paper shows that policymakers can be fooled by the results from scenario analysis, and that Monte Carlo analysis generally should be preferred. To show this, I reveal the extreme “stochastic,” or percentile, variation in scenario simulation results - i.e., pessimistic, baseline (most likely), and optimistic, respectively - when compared to analogous percentage results obtained when using Monte Carlo (stochastic) methods. Most users of scenario simulation results are likely not aware of this large, inherent stochastic uncertainty, or risk, embodied in these results.Specifically, for 2000, I compared the simulated percentage adequacy of each of the 222 U.S. nuclear power plant decommissioning funds - for each of these three scenarios, respectively - with the results obtained from an equivalent analysis - with comparable assumptions - but using Monte Carlo methods to obtain percentile (likelihood) results for each result for funding adequacy. My results suggest that there are no consistent percentile (likelihood) results for a given scenario simulation. For example, 90% of my baseline scenario balance adequacies would be expected to range between the 47th and 83rd percentiles. Similarly, there is extreme percentile variation around the funding adequacy results for the pessimistic and optimistic scenarios, respectively.The importance of my results for policy purposes - whether for the nuclear industry, and its regulators, or for policymakers associated with any subject area - may include the following. Monte Carlo simulation is superior to scenario simulation because it reveals the specific likelihood of each individual result. By contrast, each set of scenario simulation results, respectively, reveals no consistent likelihood information. Intuitively, one would infer that most scenario simulation pessimistic results would approximately reflect the same statistical likelihood; most baseline results, the same likelihood; and most optimistic results, the same likelihood. They do not. One can only very approximately infer likelihood when viewing a given scenario result. However, there may be some circumstances (e.g., cost and data availability) in which scenario analysis may be the preferable simulation method over Monte Carlo analysis.