The accuracy of predicted wages of the non-employed and implications for policy simulations from structural labour supply models

The main focus of this paper is on the accuracy of predicted wages for the nonemployed. We first examine whether the three groups of non-employed–the unemployed, the marginally attached, and the not in the labour force–should be modelled separately or together. We conclude that these are three distinct states and that they should not be pooled in modelling wages. We predict wages separately for the three non-employed groups using a range of two-state and four-state sample selection models.We predict wages separately for the three non-employed groups using a range of two-state and four-state sample selection models. Using a panel data set from Australia, we test the accuracy of predicted wages for the non-employedby focusing on those individuals who subsequently enter employment. We find that conditional predictions, which incorporate the estimated sample selection correction, perform poorly for all groups, especially for the marginally attached and the not in the labour force. Unconditional predictions from the sample selection models perform better but never out-perform a simple linear regression. These results may have important implications for policy simulations from structural labour supply models.