Rural financing in Thailand

Abstract

Rural financing in Thailand is heavily dependent on bank lending. Therefore, understanding the determinants of bank lending in the rural sector is an important element for promoting the development of credit accessibility to Thai farmers in the rural regions. Appropriate bank lending decisions would reduce lending costs and increase repayment rate and profits to the banks. Thus, a well-developed rural financial market would lead to sustainable development in the rural sector.

The purpose of this research is to identify critical factors in the bank lending decision and to investigate what factors affect the credit availability and loan price in rural lending in Thailand. This research also investigates the impact of the relationship lending (i.e., the relationship between the bank and the borrower) and the predictive power among the different estimation techniques in predicting the bank lending decision, amount of credit granted, and interest rate charged.

The data used in this research are obtained from the Bank for Agriculture and Agricultural Cooperative (BAAC). During the period of 2001 to 2003, a total of 18,798 credit files under the normal loan scheme are made available. The credit files are analyzed using the logistic regression (Logit), multiple linear regression (MLR), and four different types of the artificial neural networks (ANN), namely multi-layer feed-forward neural networks (MLFN), Ward networks (WD), general regression neural networks (GRNN), and probabilistic neural networks (PNN).

The results show that the total asset value (Asset), value of collateral (Collateral), and the length of the bank-borrower relationship (Duration) are crucial factors in determining bank lending decision, amount of credit granted, and interest rate charged. As expected, Asset has a positive impact on the bank lending decision and the amount of credit granted, while Collateral has a positive and a negative influence on the amount of credit granted and the interest rate charged, respectively. However, Collateral has no significant impact on the bank lending decision, while Asset has a significant negative impact on the loan price in some specifications.

Duration has a significant negative impact on bank lending decision, amount of credit granted, and interest rate charged, which implies the importance of relationship lending in the Thailand rural financial market. However, the negative relationships between Duration and the bank lending decision, and between Duration and the amount of credit granted, contradict the postulated hypothesizes. The results imply that the bank uses information from the borrowers and monitors the lending risk via the lending decision and amount of credit granted. On the other hand, the relationship lending benefits the borrowers via loan pricing since the borrowers with a long term relationship with the bank receive a lower lending rate.

The predictive results of both in-sample and out-of-sample on bank lending decision, amount of credit granted, and interest rate charged show that in terms of predictive accuracy, most of the artificial neural networks models outperform the logistic and the multiple regression models. The empirical results also show the superiority of using the PNN model to classify and screen the loan applications, and the GRNN model to determine the amount of credit granted and interest rate charged.