Literature Review: R* – Part II

(Note: for those interested in a comprehensive analysis, the April IMF WEO featured a chapter on the natural rate of interest. The chapter is here).

Paper #1: Determinants of Global Neutral Interest Rates

A new paper (“Determinants of Global Neutral Interest Rates”) by Thiago Ferreira (Fed Board) and Samer Shousha (Insper) argues that the supply of safe government debt is an important determinant of global neutral interest rates. The authors also stress the importance of spillover effects in explaining the strong co-movement of neutral rate trends across advanced economies.

What the paper does

The paper studies the drivers of neutral interest rates through an empirical model that simultaneously accounts for many commonly discussed drivers of neutral rates: productivity growth, population aging, convenience yields, supply and demand of safe assets, and global spillovers from productivity and demographics in other countries. The supply of safe assets is defined as the amount of U.S. government debt available to the private sector and foreign governments as a share of world GDP. The demand for safe assets is measured through the accumulation of international foreign exchange reserves in U.S. Treasuries by foreign governments and private sector demand for U.S. Treasuries following regulatory changes such as the Dodd-Frank Act and Basel III. The model is estimated for 11 advanced economies over the period 1960-2019.

Main results

Safe asset supply has important implications for the neutral rate. According to the paper, since the 1960s, supply of safe assets has determined a significant share of the trends in U.S. neutral rates. Figure 1 shows the change in the estimated US neutral rate and its determinants by sub-periods. For instance, the paper suggests that ¼ of the change in the natural rate of interest during the 1960-1974 period can be attributed to supply of safe assets. Since 2008, most factors point toward a decline in the neutral rate. However, the strong increase in safe asset supply is estimated to have raised the neutral rate by 1.3pp offsetting the downward pressure from the other effects.

Other effects, such as global spillovers, also play an important role in explaining neutral rate trends. Another result of the paper is that global spillovers have contributed to changes in neutral rates. The authors estimate that the productivity slowdown and population aging in the rest of the world (“global spillovers” in Figure 1) reduced the U.S. neutral rate by 90bp during the 1995-2019 period. Finally, the authors show that correlations across neutral rates in advanced economies could largely be explained by spillover effects alone. Given domestic changes, investor decisions are sufficient for strong international interest rate co-movement if cross-country capital mobility is high enough.

Figure 1. Main result from Ferreira and Shousha (2023): Determinants of the U.S. neutral rate (%).

A critique of the paper

The paper accounts for many potential drivers of neutral rates in one unified framework. In particular, the emphasis on supply and demand forces is an important contribution. In doing so, the authors bridge the literature focusing on saving-investment frameworks to study neutral rates with papers that focus on demand-supply forces without considering long-run drivers of real rates. The results of the paper add a new interpretation to the historical narrative for neutral rates: While many estimates find a continuing decline in neutral rates in the 2010s, the authors argue that the strong supply of U.S. Treasuries has increased neutral rates globally during that period.

The main critique of the paper, as often happens in macroeconomics, is about endogeneity. For instance, it is hard to disentangle to which extent the supply of safe assets is driven by other underlying forces such as population aging or a productivity slowdown. Similarly, the demand for safe assets could also be driven by population aging, as the authors acknowledge. The authors’ focus on asset demand through regulatory requirements is a valuable attempt to disentangle this causal chain, although some question marks on the identification strategy remain.

The paper does not provide an idea of the estimation uncertainty. Many estimates of neutral rates are subject to very wide confidence bands, and the same is true for the decomposition into different drivers. Therefore, it is not clear how reliable the estimates presented in this version of the paper are.

Policy implications

Neutral rates can be endogenous to the government’s supply of debt instruments. The low interest rate environment of the last decade has led to arguments that increased debt issuance may not be problematic if neutral rates are low. This paper estimates how neutral interest rates adjust as the supply of government debt increases and show that this issuance has led to an increase in neutral rates that overcompensated the negative contributions from all other factors combined. While current concerns about U.S. Treasury issuance are focused on cyclical interest rate implications, the results from this paper add a long-term worry to the debate: According to the paper, neutral rates may increase significantly if excess supply is very large.

Paper #2: Measuring The Natural Rate Using Natural Experiments

The paper “Measuring The Natural Rate Using Natural Experiments” by Verónica Bäcker-Peral (Princeton), Jonathan Hazell (LSE), and Atif Mian (Princeton) uses extensions in UK property leases as natural experiments to estimate the natural rate of return on capital. This innovative approach allows the authors to get very precise estimates of the natural rate without concerns about model misspecification.

What the paper does

This paper studies trends in the natural rate of return on capital, defined as the return required to balance supply and demand in the UK property market, that is the price that clears this market for private capital in the long run. This natural rate differs from the neutral real rate (R*) through a natural risk premium on capital and long-run expected capital gains (but remains the “close cousin” of R* to use Mian’s words).

UK property lease extensions serve as a natural experiment to estimate the natural rate. Many properties in the UK are leaseholds, meaning they have very long but finite leases of decades or centuries. These leases can be bought and sold and are commonly extended far in advance before the end of the lease. The authors use these extension events to estimate the change in market value of a property due to duration extension by comparing the price before and after the lease extension with the change in price of an otherwise similar control group that does not get a lease extension. Given this price change, a discounted cash flow model can be used to estimate the natural rate.

Main results

According to the paper, the natural rate fell from 4.8% in 2003-2006 to 2.3% in 2022. 2023 has seen a small uptick in natural rate estimates (Figure 2). The 95% confidence bands around these estimates are tight, especially towards the end of the sample, when the width is often below 0.1pp. (the evidence in Figure 2 has been discussed on Twitter by one of the authors (Mian) here).

Figure 2. Natural rate of Bäcker-Peral, Hazell and Mian (2023).

A critique of the paper

The authors provide an innovative approach to estimate the natural rate that reduces estimation uncertainty and model misspecification risks. The authors’ method also ensures that the natural rate can be estimated in times of structural breaks such as the recent Covid period, which posed a challenge for established methodologies.

The drivers of the trends in the natural rate are unclear as well as the external validity. This paper focuses on estimating the level of the natural rate and does not provide an attempt to decompose the negative trend over the sample period into the contributions of different causal factors. Furthermore, the results are estimated for the UK property sector and may not translate to other asset classes. The authors argue that their natural rate estimate should relate to the broader economy since housing is a major asset class. How tight this connection actually is cannot be answered directly with the methods used in the paper. However, while the level of the natural rate may differ from R* due to risk premia or expected capital gains, both rates can have parallel/similar trends.

Conclusion

In these two notes, we have reviewed 4 papers that explore the drivers of R* and its recent evolution. Figure 3 shows the latest results of two publicly available models that estimate R*. The first one is the Lubik-Matthes (L-M) model maintained by the Richmond Fed (here), and the second one is the Laubach-Williams (L-W) model maintained by the NY Fed (here). The two models produced similar results pre-Covid but they have diverged in recent quarters with the L-M model suggesting an increase in R*. It should now be clear to the reader that real-time differences across models are “normal”, given estimation issues. We do not enter into the details (in case, please get in touch) but it is likely that the L-M model is overstating the possible increase in R* while the L-W model is probably not capturing part of the increase. For a discussion on this point, please see this document and this paper. The intuition is that the revised post-Covid L-W model (shown in Figure 3) allows for a component that captures high frequency movements of R*, therefore limiting the possibility of an upward drift of the low frequency component shown in Figure 3. The L-M model has also issues. In fact, if the L-M model is run assuming that the next 4 years evolve according to a recent Macroeconomic Advisers forecast, the ex-post blue line in Figure 3 (that is, what we will see in 4 years from now) will look flat rather showing an increase in R*. In other words, which one is “right”, the L-M or the L-W model? Probably neither of the two, and probably the truth is somewhere in the middle.

Figure 3. Latest estimates of R* from Lubik-Matthes (L-M) and Laubach-Williams (L-W) models

Note: “lb” and “ub” stand for “lower bound” and “upper bound” respectively.

Very difficult for Powell and other policymakers to draw a conclusion. Putting everything together, to us the literature seems to convey the following messages: (i) the estimation of R* is model-dependent, particularly prone to end-of-sample issues, and estimated with large uncertainty, (ii) the factors affecting the low frequency components of R* are known, although their contributions is subject to a debate, (iii) the factors that have pushed down R* over the last decades should remain in place going forward at least until demographic trends will revert and/or structural productivity will accelerate, (iv) supply of safe assets does play a role, (v) R* might have increased in recent quarters but we cannot be sure. Ultimately, to us, Powell’s answer at the last press conference reflects what we know from the literature and the uncertainty surrounding it.  

Contact

Want something more tailored?

We provide tailored consulting on ad-hoc projects.