SELECTING A BOND PRICING MODEL FOR TRADING IN KAZAKHSTAN

Abstract

In this research, we fit term structure models of the interest rates to cross-sections of government bonds in Kazakhstan and investigate whether we can use the pricing errors as indicators of mispricing. We test whether we can generate excess returns by building portfolios of bonds based on the mispricing from the term structure models. We consider commonly used specifications of term structure models: Cox, Ingersoll and Ross (“CIR”) and Vasicek as well as basic spline function. The abnormal returns are measured against a duration-based benchmark. We construct zero-yield curves based on the cross-sectional estimates from the term structure models and compare this to term structure plots based on tenure (duration/convexity). Trading test results show that CIR and Vasicek bond-pricing models efficiently detect mispricing and generate 14.4% and 15% cumulative abnormal returns (“CAR”) over 6 years if the data would be actually tradable. On the other hand, spline model overfits the data and does not detect mispricing as efficiently as both economic models. Experiments with less knot points did not help to improve the issue of overfitting. Because of limitations to the number of cross-section estimations we are unable to show that the CARs for CIR and Vasicek are statistically significant. However, the pattern seems similar for both term structure models.