摘要：This paper attempts to evaluate the yield forecasting ability of alternative time series models using monthly data for debt securities in India with residual maturities ranging between 14 days to 25 years. The study period stretches from April 1996 to March 2010. Two univariate models namely Exponential Smoothing Method (ESM) and ARIMA as well as a multivariate VAR model are used for this purpose. The authors find that conventional method like ESM does a better job for both short (three months) as well as long range (twelve months) forecasting of yields compared to more complex and informationally expensive models like ARIMA and multivariate VAR. It is also observed that level of interest rate volatility impacts the yield forecast accuracy for a given period. Short-term yields are more difficult to forecast than yields for securities with longer maturities. Short range forecasting is better than long range forecasting in high interest rate volatility period while there is no such clear pattern,using different time series model,for low interest rate volatility period. The findings have strong implications for both policymakers and debt market players such as bankers, insurance companies and debt funds. The former uses yield forecast information for developing policy formulation while the later employs it for their asset-liability management as well as portfolio management strategies. This research contributes to financial econometrics as well as debt market literature for India which is a fast emerging economy.
关键词：interest rates;residual maturity;debt market;time series models;commercial banks.
