Zero-Adjusted Log-Symmetric Quantile Regression Models

This paper proposes zero-adjusted log-symmetric quantile regressions to deal with the issue of regression estimation when there are many zeros in the dependent variable. We introduce the zero-adjusted log-symmetric distributions that accommodate the presence of zeros and are consistent with heteroscedasticity. The model builds on a conditional quantile distribution and the parameters are estimated by maximum likelihood. The quantile approach is more flexible to analyze the effects of the explanatory variables on the dependent variable, increasing the relevance of the proposed models to the existing literature on zero-adjusted log-symmetric regression models. We apply Monte Carlo simulations to evaluate the quality of the estimates and compare alternative model specifications. As an illustration, the proposed models are confronted with actual data on hourly wages of retired people and time spent on extramarital affairs. The empirical results indicate that the proposed models provide a better fitting to the data. Particularly, the zero-adjusted log-symmetric quantile model outperformed the existing zero-adjusted gamma and zero-adjusted inverse Gaussian regression models.