Choosing Between Non-Normal Cost Distributions for Risk-Averse Decisions

Summary

The document presents a decision problem in which Monte Carlo simulation has produced cost distributions for two alternatives. Their benefits are assumed equal but are not specified, and the decision maker is described as risk averse. Because the simulated cost distributions are not normal, the questioner doubts that a comparison based only on expected cost minus a variance penalty is adequate. The central issue is how to rank uncertain costs when the distribution’s shape may matter to the stakeholder’s preferences.

No answer, proposed decision rule, simulation output, or distributional evidence is included, so the document does not resolve the comparison. It identifies a useful distinction: expected cost and variance may fail to capture preferences over asymmetry, tail losses, or other features of non-normal outcomes, but the preferred criterion depends on how risk aversion is defined. A decision analysis would need a stakeholder utility or an agreed risk measure and enough information about the simulated outcomes to apply it. Without those details, neither alternative can be selected from the document alone.

Key ideas

• The problem compares two alternatives with simulated, non-normal cost distributions and assumed equal benefits.
• A mean-and-variance score may not represent preferences when cost distributions are non-normal.
• Choosing a risk-averse decision rule requires a defined utility function or agreed risk measure.
• The document provides no simulation results or decision rule, so it cannot identify the preferred alternative.

Tags

# Comparing cost of two alternative given their distribution


# Comparing cost of two alternative given their distribution












I have distribution for cost of two alternative through Monte Carlo simulation. The distributions are not normal. Given the benefit of the two alternatives is the same but ungiven, I want to choose the alternative with less cost for a risk aversive stakeholder. (I can't simply choose the one with smaller E(c)-a*var(c) since the distribution is not normal)

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.