Forecasting government bond yields in Thailand

Abstract

This dissertation aims to comprehensively investigates key macroeconomic factors and bond yield interaction in Thai bond market, and analyzes the impacts of both domestic and international economic factors on the fixed income yields in different maturities of 1-, 3-, 5-, 7-, and 10-year movements, as well as forecasts the future bond yields of different maturities with macroeconomy, including comparing the best predictive yields with each model. Through estimating VAR, Bayesian VAR, and Single Equation (SE) approaches as well as Random Walk (RW) forecast as the benchmark, the results show several key findings:

Overall, fixed income yields generally respond strongest to the economic factor shocks. Explicitly, yield in various maturities responds directly to positive and negative shocks in macroeconomic indicators (i.e., six key variables: fed rate, primary budget deficit, commodity price, capital inflow, VIX index, and liquidity). Generally, it finds that both domestic and international macroeconomic factor shocks have a significant impact on the fixed income yields of various maturities with all models. Regarding the macro shocks from fed rate, commodity price, VIX index, capital inflow, primary budget deficit, and liquidity have a strong impact on the bond yields in all maturities and the impact is transitional, usually dies out after 5 to 10 quarters but the effects of Bayesian VAR approach seem to be long lasting more than 10 quarters. Interestingly, from the results, evidence shows that new economic variables intended into this study, commodity price and capital inflow have a quite strong impact on the bond yields with all maturities as well.

Last, VAR, Bayesian VAR, and SE models are built to forecast future bond yields in various maturities with macroeconomy. It finds that the static forecast (in-sample) with all of three models, the strong evidence results show that the most case of the BVAR model produces the best predictivity of bond yields in a different maturity, except for only 10-year maturity that the RW forecast beat BVAR model. Most figures of BVAR model of these statistical functions measured are the lowest and for RMSE and Theil of evaluations of all yields are smaller than one signals that the model under consideration strongly outperforms the SE, VAR, and RW models, but the figures of RW forecast (10-year maturity yield) beat all the models. The results reflect that the performance in-sample forecasting is quite good. Besides, for the dynamic rolling forecast (out-of-sample) with both VAR and BVAR models, the evidence results confirm that BVAR model is the best performance in dynamic rolling-window forecasting the bond yields with various maturities for 2-, 4-, and 8-quarter rolling ahead. The statistical evaluations show that the most figures of BVAR with all rolling forecasts appear the lowest and outperform the VAR at all maturities. Hence, the BVAR model produces generally more accurate forecasting future yields than those competitive model in a robust way.