Estimation of penalized quantile regression with semiparametric correlated effects and an application

In this thesis, a recent semiparametric penalized panel quantile regression estimator and its practical parametric alternatives proposed by Harding and Lamarche (2017) are investigated and compared to their predecessors. Moreover, a tuning parameter optimisation algorithm for penalised panel quantile regression, a simulation based test and a Hausman-type misspecification test are introduced and applied on a Mincerian earnings model. The dataset covers a balanced panel of 2953 individuals between 2012 and 2015, which are extracted from "Income and Living Conditions Investigation" panel survey data collected by TurkStat. Test results show that, (i) Harding and Lamarche (2017) estimators are superior to their traditional counterparts; (ii) optimum λ for semiparametric estimator is 0.9, 1.4 for Lamarche (2010) and 1 for others; (iii) the model is correctly specified and introduced estimators are consistent in this manner. On the other hand, estimation results reveal that there is an inverse shaped relationship between earnings and experience; there may be a limited evidence on sheepskin effect; more schooling increases earnings; workers graduated from vocational high schools earn more than general high school graduates; marital status and sex have significant effects on income in all quantiles; health status has significant effects on income only in the lower quantiles and finally income differ according to line of work, except for the highest quantile.