Prediction of individual JG bond prices via the TDM model

Kariya and Tsuda (1995) demonstrated the predictive power of TDM (time dependent Markov) model for individual bond prices with the end-of-month price data of JG (Japanese Government) bonds with initial maturities of 10 years. The model predicted well the monthly term structure of the individual JG bond prices for the period 1991.1-1992.12 though there are only four parameters in the model, where there are about 80 bonds for each month. In fact, the prediction standard error for the period is 0.9 yen while the estimation standard error is less than 0.3 yen, where the face value of a JG bond is 100 yen. We again test the prediction power of the TDM model with the end-of-month price data of JG bonds for the period 1993.1-1995.12 when the interest rate level was low, and observe that the model loses the predictive power when interest rates change volatilly even though the overall performance is good. The observation follows from the fact that the VAR (vector autoregressive) model for predicting four time dependent parameters in the model, which is modelled based on the cross-sectionally estimated parameters, fails to keep a stable prediction power for months of volatile interest rates. It is remarked that the TDM model is proposed by Kariya and Tsuda (1994) as a time series extension of the CSM (Cross-Sectional Market) model for individual bond prices Kariya (1993) formulated.