Interesting paper by Michael Bauer at the San Francisco Fed:
….The difficulty of predicting changes in interest rates mainly arises from two features that characterize their evolution over time. First, like other financial variables, interest rates vary widely from day to day, which makes them difficult to link to economic fundamentals such as monetary or fiscal policy. This well-documented “excess volatility,” was first pointed out in Shiller (1979), and it reflects the importance of frequent changes in investor sentiment due to a never-ending stream of economic data releases and other news.
Second, as evident from 10-year Treasury yields since 1971, seen in Figure 1, interest rates have not fluctuated around a stable average level over this period. Instead of “mean reversion” around a constant average, they exhibit slow-moving trends, such as the rise during the “Great Inflation” period of the 1970s, and the long-lasting decline since then.
….the gap model does not assume that the level of the series will revert to some constant mean, but instead that the gap between the series and its trend component will revert to zero. Estimating trend components and gaps underlies most macroeconomic forecasting, and Faust and Wright (2013) recently demonstrated the gap model’s excellent performance for inflation forecasting.
….Since inflation is ultimately determined by monetary policy, the long-run inflation trend corresponds to the perceived inflation target of the central bank. This can be estimated reasonably well from surveys. Figure 1 plots the publicly available and mostly survey-based inflation trend estimate (red line) that underlies the Federal Reserve Board’s structural model of the U.S. economy, FRB/US. For the trend in the real interest rate, also called the natural or equilibrium real interest rate, Laubach and Williams (2003) suggested a way to estimate it from macroeconomic data and popularized its use in policy analysis (see also Williams 2016). Figure 1 includes an estimate of the equilibrium real interest rate (green line) taken as the average of several popular estimates, as discussed in Bauer and Rudebusch (2017).
Figure 1 also plots the sum of these two trends (red line); this estimate of the trend component in interest rates has exhibited a very pronounced decline since the 1980s. The 10-year yield generally fluctuated near this trend, and both are currently very low in historical comparison, with important consequences for policymaking (Williams 2016). Figure 1 suggests that it may be useful to take into account the level of the trend when forecasting interest rates.
….the final piece required for a practical forecast rule is an assumption about the transition of interest rates to their trend. Based on how quickly interest rates have historically reverted back to the trend, a reasonable assumption to make for this forecasting exercise is that 20% of the remaining gap is closed each quarter. But the precise speed of reversion to the trend is typically not crucial for forecasting performance (Faust and Wright 2013). Furthermore, it becomes essentially irrelevant for long-horizon forecasts, since forecasts are approximately equal to the estimated trend…..