even more helpful to understand the point. The opposite holds with a proper prior. In
this case all the expected values have the "correct" sign, but some variances are still
quite large (even if three times lower than with a diffuse prior). In South there are still
large regions of theoretical irregularity, as we can see from Fig. G.
Nonetheless this analysis is conceptually anomalous, since it is conditional on
actual factor share. This is not very meaningful, as in our translog model, we should
condition on the triplet (Y,t,w/r). Moreover marginal distributions don’t fully account for
regularity conditions, that must be imposed at the same time on all the elasticities. These
tasks are accomplished in two steps as discussed above. First we perform a Monte Carlo
composition method to get a sample of 10.000 observations from the joint distributions
of parameters and predicted. Then we use this sample to derive elasticities conditional
on (Y,t,w/r) in 1989. Subsequently we impose concavity and plot the resulting
distributions. All these are plotted together to check different behaviors. Distributions
conditional on labor shares are plotted with a dotted line, while I have plotted with a
solid line those conditional on (Y,t,w/r) with and without imposing concavity.
13