Measuring Variability and Stationarity of Term Premia for Interest Rate Forecasting

We study a series of weekly term premia extracted from U.S. Treasury bill quotations from 1970-1982. We choose this period because it is characterized by high and variable inflation. We find that spot and forward rates are cointegrated and that the series of their differences is stationary. This implies that term premia are also stationary, which has implications for researchers seeking to improve interest rate forecasting. It also rules out several variables as determinants of term premia. We use a variance estimator that is consistent against autoregression to compute variance bounds. This lets us use overlapping observations, thus conserving degrees of freedom and letting us use a finer sampling interval. This estimator shows that both forecast errors and term premia are more variable than reported previously. Those earlier results used periods of low inflation, though, so our results are not necessarily inconsistent with them. Variation in the premium rises with the level