The support search is repeated R = 3 times, and both the AMMSE and MAP estimates are calculated.
Compared with the MAP estimate, the AMMSE estimate has an improved performance as expected.
Here we still adopt the AMMSE estimate in the two BMP methods.
The differences between the MELO estimator and AMMSE arise because his marginal posterior for [Mathematical Expression Omitted] is based on a diffuse prior that has been updated with sample information assumed to be generated by a process that has independently distributed normal errors with zero mean and constant variance.
To fully describe the comparative performance of OLS versus AMMSE in small samples would require extensive research along the lines of Kling's recent experimental comparisons (1988a, 1988b) of random utility versus conventional travel cost demand models.
Each experiment involves 500 independent replications where OLS and AMMSE (with the OLS estimates as the starting values) are applied to the task of estimating s using samples of 100 observations.
In terms of mean squared error, the AMMSE estimator of consumer surplus offers an improvement over both the OLS and the inequality-constrained OLS estimates for all but the case of the smallest (in absolute magnitude) own-price coefficient.
However, this improvement develops because the AMMSE estimator accepts a large bias.