CONCLUSIONS

1. Virtually all commonly used probability distributions admit of CE-Value functions obtained

• In closed form, or
• As a series expansion, or
• through numerical integration

2. Value additivity enables the value of information concepts EVPI and EVSI to be extended to the risk-averse case, yielding CEVPIt and CEVSIt .

3. The CE Value of Perfect and Sample Information may exceed EVPI and EVSI respectively, in some cases by a wide margin. A risk averse D.M. may value sample information much more highly than a risk neutral one does, hence EVSI and EVPI may NOT be taken as upper limits on amounts spent on sample information.

4. For asymmetric return distributions and for scenario optimizations,

The Mean-Variance Efficient Frontier is replaced with the CE-Maximality frontier as a set from which to choose the ‘optimal’ solution.

5. Backtest results indicate that CE-Maximal solutions MAY perform better than Mean-Variance efficient ones in actual practice.

6. Entire "callable library" should be available soon. Keep watching www.matpro.com/cevalues.