A Copula-Based Selectivity Model: Applications to Older Workers in Thailand

Abstract

This dissertation applied a copula-based selectivity model to study the labor force participation of older workers in Thailand. Importantly we paid attention to three selectivity models including: 1) sample selection model, 2) endogenous switching model, and 3) polychotomous choice selectivity model. Moreover, we considered different distributions for margins, such as normal, logistic, and Student’s t distributions, in all of empirical studies. Also, we considered various copulas, both radially symmetric and asymmetric copulas, such as, the Gaussian, FGM, and Archimedean copulas (AMH, Clayton, Frank, Gumbel and Joe copulas). The summary of findings for each study is presented. For the first empirical study, we applied a copula approach to a sample selection model to determine the factors affecting the labor force participation and working hours of older workers in Thailand. We found that the copula approach to the sample selection model, which allows for flexibility in dependence structure and relaxing the joint normality assumption, works in this context and performs better than the BVN bivariate normal model) especially, the Frank (L-t) copula model. Furthermore, the results obtained show the presence of significant positive dependency of unobserved factors between two error terms. In addition, the policy recommendation is that the stakeholders should consider having a program that can increase the working hours of some worker groups, such as female workers, or workers who are not single or workers who do not reside in the Bangkok province. For the second empirical study, we applied the copula approach to an endogenous switching model to determine the public and private sector wage for older workers in Thailand. This study demonstrates that all of the copula-based models perform better than the standard model in this context, especially the Frank-Gaussian (L-t-t) model. Furthermore, these results show the presence of significant negative dependency of unobservable factors between the switching regression and the wage regression. In addition, the concerned policymakers should facilitate campaigns that encourage the public sector to retain the older workers with high wages, especially those older workers who are individuals with high education and skills. The copula approach was applied to a polychotomous choice selectivity model for the third empirical study to estimate the wage determination according to the occupational choice for older workers in Thailand. This study demonstrates that a polychotomous choice selectivity model based on the copula approach performs better than the standard one. Likewise, it is evident that among the copula-based models, the Frank copula–based model provides the best fit. Also, these results show the presence of highly significant dependency of unobservable factors between the occupational choice regression and the wage regression for unskilled and skilled workers, which implies that the selectivity bias exists. The empirical results show that the older workers who live in Bangkok earn higher wages than those who live outside Bangkok. The gender variable has no impact on wages for the highly skilled older workers. For the unskilled, the male worker earns more. A surprised result is that the experience has a significantly negative impact on the wages regardless of the level of the skill of the older workers. This finding requires in-depth further research to explain this phenomenon. Finally, we found that copula-based selectivity models that allow for flexibility in dependence structure and relaxing of the joint normality assumption perform better than the conventional model in all of the empirical studies. Thus, applying copula approach in selectivity model should be taken into consideration in order to reduce the risk that may occur from some strong assumptions and attain the efficient estimators.