The gender pay gap in Hungary: new results with a new methodology

We estimate the gender pay gap with the traditional OLS based Blinder-Oaxaca decomposition, and with an extension using Random Forest (RF) regressions on Hungarian data for the years 2008-2016. Random Forests perform better as predictors out-of-sample and yield consistently lower estimates for the unexplained pay gap than OLS. Then we analyse the unexplained gaps obtained from the RF regressions with a CART (Classification and Regression Tree) analysis. It seems that sectoral and educational factors are most consistently involved, but some other factors like firm size, age or tenure are also important. There are indications that medium educational levels and small firm size together, in certain industries, are most conducive to small (or even negative unexplained gaps), while high educational achievement in certain other industries (including manufacturing) are responsible for the highest gaps. In the first years of our sample period it was true in particular for middle aged and older women. This seems to be in accordance with the idea that educated women may have had problems with accumulating human capital.