Martingales and efficient forecasts of effective mortgage rates

Abstract

Economic theory and empirical evidence indicate that the optimal near-term forecast of a long-term rate is (approximately) today's rate; the no-change model should provide excellent near-term forecasts of a long-term rate. This article estimates the longest tbrecast horizon over which no-change predictions of each of three mortgage-related interest rates pass a series of quality tests. The empirical results reject the optimality of no-change predictions of the one-year Treasury bill rate for all forecast horizons. Since October 1979, the tests support the hypothesis that no-change predictions of the 30-year conventional home mortgage and GNMA rates are optimal for forecast horizons of up to threequarters-ahead.Interest rate forecasts are of more than passing concern to commercial banks, savings and loan associations, and many private and institutional players in the real estate market. Key financial decisions often hinge on the forecast that is deemed the best. At the same time, it is not always clear how to assess the merits of alternative interest rate forecasts. This paper is designed to help in such an assessment. Its primary contribution is an examination of the efficiency of no-change model forecasts.I f the market for fixed-income securities is efficient and if expectations of future short rates are embedded in current long-term rates, long-term rates should follow approximately a martingale sequence (Pesando, 1979)J In such an environment, the no-change model should produce efficient near-term forecasts of longer-term rates, with efficiency becoming less precise as the forecast horizon is lengthened (Pesando, 1981).I build on this research by estimating the longest forecast horizon for which nochange predictions of three mortgage or mortgage-related interest rates are efficient. The interest rates include the 30-year, fixed-rate conventional home mortgage rate; the current-coupon GNMA (Government National Mortgage Association) rate; and the one-year Treasury bill rate, which is often used as the base rate in adjustable rate mortgage agreements. The time periods examined are the eightyear periods bracketing the October 1979 change in monetary policy.

A random walk approach to predicting US 30-year home mortgage rates

“…We thus conclude that the random walk forecasts are unbiased for forecast horizons of up to three quarters ahead; see Reichenstein (1989) who reaches a similar conclusion.…”

“…1, it sets the forecast equal to the rate most recently observed at the time of the forecast (that is, the average rate for the first two full weeks of the month). The rationale for our random walk model is provided by term structure theory which suggests that, in an efficient bond market, the optimal forecast of a long-term interest rate is basically today's rate; see Pesando (1979) and Reichenstein (1989). and n 22 (n 12 and n 21 ).…”

Consensus vs. Time‐series Forecasts of US 30‐year Home Mortgage Rates

“…In line with this theory, Reichenstein and Shetty (1993) find that the near-term forecasts of the TBR from a simple random walk model are accurate. As we shall see, the HMR and TBR share a common trend (Figure 1) to which they converge in the long run, suggesting that the near-term random walk forecasts of the HMR should also be accurate (Reichenstein, 1989). With this in mind, we ask whether the consensus forecasts of the HMR from a panel of experts (Blue Chip) can outperform the random walk forecasts, and whether a more general and yet simple model produces more accurate forecasts of the HMR.…”