Are Financial Markets Good Predictors of R-Star?

Sophia Cho and John C. Williams

Recently, there has been renewed attention on the natural rate of interest—often referred to as “r-star”—and whether it has risen from the historically low levels that prevailed before the COVID-19 pandemic. The natural interest rate is the real (inflation-adjusted) interest rate expected to prevail when supply and demand in the economy are in balance and inflation is stable. Some commentators claim that the prior decline in r‑star has reversed, pointing to the recent rise in future real interest rates implied by the bond market. But before declaring the death of this “low r‑star” era, a natural question to ask is: how reliable are market-based measures of r‑star? In this Liberty Street Economics post, we evaluate whether such measures provide additional information on future real interest rates beyond what is already contained in macroeconomic model-based estimates of r-star. Our findings suggest they do not, and we conclude that reports of the death of low r-star are greatly exaggerated.

Shifting Trends

Real interest rates in many countries exhibited a sizable downtrend over the quarter-century leading up to the COVID-19 pandemic (Williams 2017). This pattern is seen in U.S. data shown in the chart below. The blue line represents the ex post real federal funds rate, defined as the effective federal funds rate minus the four-quarter percent change in the price index for personal consumption expenditures excluding food and energy (“core PCE inflation”). The red line represents a market-based measure of longer-term real interest rates: the five-year, five-year-forward real yield implied by U.S. Treasury Inflation-Protected Securities (TIPS), measured at the end of each quarter. The shaded areas indicate recessions, as determined by the National Bureau of Economic Research (NBER) Business Cycle Dating Committee. As seen in the chart, short- and longer-term real interest rates averaged around 4 percent in 2000, around 2 1/2 percent in 2007, and only about 1/2 percent in 2019—all periods of a cyclically strong U.S. economy.

Over the past year, both short- and longer-term real interest rates have risen well above levels that prevailed in the years leading up to the pandemic. In particular, recent TIPS yields have been between 2 1/4 and 2 1/2 percent, comparable to levels seen directly before the global financial crisis. Looking at TIPS yields alone, one might conclude that the low r‑star era has indeed come to an end.

TIPS: A Leading Indicator of R-Star?

In theory, market-based measures of r-star should be very informative about future interest rates and should outperform macroeconomic model-based estimates that rely on only a limited number of indicators. Indeed, market participants have strong financial incentives to incorporate all available information—including that from models—to make highly informed forecasts of future interest rates. In practice, however, it is not that simple. For example, model-based forecasts can perform as well as or better than direct reads of future interest rates from longer-term yields (Taylor, Brandt, and Dotta 2025).

To shed further light on this subject, we evaluate various measures of r‑star based on their ability to predict future real interest rates. To conduct an “apples-to-apples” comparison, we compare market-based measures of r-star with the corresponding real-time estimates from the Holston, Laubach, and Williams (HLW) model described in our previous post. The chart below displays four-quarter moving averages of the real-time HLW estimate (blue line) and the five-year, five-year-forward TIPS yield (red line). TIPS yields are based on CPI inflation, while HLW estimates are based on core PCE inflation, so the levels are not perfectly comparable. This difference, however, does not affect the broad movements over time (Williams 2015).

As seen in the chart, the overall pattern of declining TIPS yields is similar to that of real-time HLW estimates prior to the pandemic. That said, two key differences stand out. First, TIPS yields tend to be much more volatile, evident in the sharp rise and reversal over 2013–2014 and the decline and reversal over 2019–2021. Second, downward movements in TIPS yields lag those in real-time HLW estimates, both in the early 2000s and early 2010s.

In contrast to the real-time HLW measure of r-star, the TIPS-based measure has essentially no predictive power for real rates three years in the future. This can be seen in a simple regression of the real federal funds rate on the five-year, five-year-forward TIPS yield from three years earlier (and a constant). The estimated coefficient on the lagged TIPS yield is very small and statistically indistinguishable from zero. This result also holds for TIPS yields at shorter forecast horizons, such as four or five years in the future. In comparison, the real-time HLW measure is positively correlated with future real rates, consistent with Taylor, Brandt, and Dotta (2025), who conduct a more extensive assessment of the predictability of nominal rates using current, rather than real-time, measures of r-star.

Is There a Better Market-Based Measure of R-Star?

The lack of predictive power of longer-term TIPS yields for future real rates suggests that TIPS yields have not been a reliable guide to r-star. One plausible explanation for this poor forecasting performance is that TIPS yields do not necessarily correspond to market expectations of short-term interest rates, owing to liquidity and risk premiums embedded in these yields. To address these factors, we turn to term structure models and a survey-based measure of r‑star.

A number of term structure models have been developed that adjust for liquidity and risk premiums, with the goal of producing market-based estimates of real interest rate expectations that correspond more closely to r-star. Among these is the D’Amico, Kim, and Wei (DKW) model, which provides estimates of the expected real short rate five-to-ten-years ahead, measured at the end of each quarter. These estimates include the period before TIPS were first issued, allowing our analysis to extend back an additional five years. Note that we use current, revised DKW estimates, which are not directly comparable to real-time HLW estimates but should still provide reasonable proxies for market expectations of real rates.

We also evaluate the Blue Chip survey of long-run forecasts. This survey directly asks about expectations of interest rates and inflation, making it immune to concerns about liquidity and risk premiums. It is conducted twice a year, typically in early June and again in early December. We construct a Blue Chip measure of r-star by subtracting the consensus projection for the GDP chained price index inflation rate from the projection for the federal funds rate, for the five-year period furthest into the future. To be roughly consistent with the timing of the real-time HLW measure, we treat the first Blue Chip long-run forecast of each year as being based on first-quarter data and the second as being based on third-quarter data.

The DKW and Blue Chip measures of r-star differ significantly in the two decades before the onset of the pandemic but are broadly similar since then. The chart below compares the real-time HLW (blue line), DKW (red line), and Blue Chip (gold line) measures of r-star. The rise and reversal in r-star over the late 1990s and early 2000s is evident in the real-time HLW and Blue Chip measures but less so in the DKW measure. Following the global financial crisis, the decline in the Blue Chip measure is far more gradual than in the real-time HLW measure. During the pandemic, the DKW and Blue Chip measures fall to around 0 percent but then rise to somewhat above 1 percent.

Based on simple regression analysis, the DKW measure of r-star has some predictive power for future real interest rates, but it does not add much information beyond what is already contained in the real-time HLW measure. The table below reports the regression results: column (1) for the real-time HLW measure, column (2) for the DKW measure, and column (3) for both measures. The DKW and real-time HLW measures both have predictive power for future real interest rates, with the real-time HLW measure providing somewhat better forecasting performance. That said, the DKW measure does not improve forecasts if the real-time HLW measure is already included, seen by the nearly identical root mean square forecast errors in columns (1) and (3). We find similar results when we extend the real interest rate forecast horizon to four or five years.

The Blue Chip measure of r-star also has some predictive power for future real rates, but it is clearly dominated by the real-time HLW measure. The table below reports the regression results. Adding the Blue Chip measure to the real-time HLW measure does not improve forecasts, seen by comparing columns (1) and (3).

Whither R-Star?

Given past forecasting performance, one should look first to macroeconomic models for guidance on r-star, rather than to market-based measures. So what do these models tell us about r-star’s movement since the pandemic?

As shown in the table below, relative to the real-time estimate from the third quarter of 2018 (a point in time for which we have real-time r‑star estimates from a variety of models), the HLW estimate has risen by about 1/4 percentage point. Looking more broadly at the five macroeconomic models, the typical increase over this time period is between 1/4 and 1/2 percentage point, with a median increase of 30 basis points. Although estimates from the six term structure models are more dispersed than those from the macroeconomic models, their median increase of 1/2 percentage point is significantly smaller than the 1 1/2 percentage point rise in the longer-term TIPS yield over the same period.

While r-star is famously difficult to estimate with precision and estimates vary across models, model-based estimates remain valuable for predicting future real interest rates, especially compared with market-based measures. Drawing on evidence from a variety of models, a reasonable estimate is that r-star has risen by a relatively modest 1/4 to 1/2 percentage point from its 2018 level. Thus, despite the recent rise in TIPS yields, the evidence suggests that the low r-star era is far from over.

Sophia Cho is a research analyst in the Federal Reserve Bank of New York’s Research and Statistics Group.

John C. Williams is the president and chief executive officer of the Federal Reserve Bank of New York.  

How to cite this post:
Sophia Cho and John C. Williams, “Are Financial Markets Good Predictors of R‑Star?,” Federal Reserve Bank of New York Liberty Street Economics, August 25, 2025, https://libertystreeteconomics.newyorkfed.org/2025/08/are-financial-markets-good-predictors-of-r-star/.