Pension Choices of Senior Citizens in Thailand:A Multi-Label Classification with Generalized Maximum Entropy

Thamonwan Ruanto, Supanika Leucharusmee, Warattaya Chinnakam

Abstract

Following the World Bank's five pillars conceptual framework, this study applied the Classifier Chain Generalized Maximum Entropy (CC-GME) method to examine individual characteristics of senior citizens with different choices of pension options in Thailand. The CC-GME model was developed for the multi-label classification problem, which can directly be applied to estimate a discrete choice model where each individual has more than one pension plan.  As the model is GME based, it benefits from the semi-parametric nature of the model and can predict a set of pension plans chosen by each senior citizen without making an assumption on the error distribution. Moreover, GME is robust to the multicollinearity problem allowing us to study correlated pension choice determinants. The results show that the majority of Thai senior citizens rely more on their saving, family and government universal supports as only a small percentage have social security benefits or workplace pensions.  The lack of financial stability problem is especially serious among people without high school degree and live in the rural area.