Deficits are Raising Interest Rates. But Other Factors are Lowering Them.

Long-term nominal interest rates in the US have fallen since 2001 even as federal deficits and debt have risen, in particular since the Great Recession. This has led some to question the pre-crisis conventional economic wisdom about a link between deficits and interest rates. In this blog post, I update the empirical analysis of Warnock & Warnock (2009) to show that deficits and debt, both actual and expected, still have positive and statistically-significant impacts on interest rates at the margin. A one percentage point increase in the ongoing cyclically-adjusted federal deficit raises the 10-year Treasury yield by 18 basis points, and the ACM 10-year term premium by 20 basis points. Yet other factors with downward effects on yields such as growth and productivity, monetary policy, demography, and foreign demand for US Treasuries have ultimately dominated this upward fiscal effect since 2001.

Introduction

In a recent Foreign Policy article, economists Jason Furman and Larry Summers highlight an economic mystery, some might even say contradiction. Federal deficits and debt have risen substantially over the past 20 years. But interest rates, rather than rising as conventional economic theory would have predicted, have fallen.

This miss has not only sparked a deeper — and in my opinion, healthy — debate about the economic tradeoffs of deficits and debt, it has also influenced, implicitly if not explicitly, public policy already. The $1.5 trillion 2017 tax cuts. The string of increasingly-expensive bipartisan spending deals since 2014. The Green New Deal that its advocates, at least in its current drafts, propose financing entirely through deficits. Policymakers not only appear to be eschewing fiscal offsets for short-term, countercyclical policy, but also for long-term, permanent policy.

Every change in fiscal policy must be financed in some manner, even if it’s just pulling the leaves off trees and declaring them legal tender (though I wouldn’t recommend that). And each of these means of financing — whether it’s raising taxes, cutting spending, selling bonds in the open market, or simply printing money — poses tradeoffs.

By now, a decade after the Great Recession, it’s clear that the tradeoffs of deficit financing are more complex than the pre-crisis conventional thinking, at least in the way that conventional thinking was commonly understood, and that the short-run debt anxiety in the years after 2008, when the US was still so far from full employment, was in retrospect grossly misplaced.

But “complexity” does not mean that the tradeoffs have disappeared entirely, or even that economists mismeasured them in the first place. As I hope to show in this (primarily empirical) blog post, higher deficits probably have been exerting upward pressure on interest rates this whole time, and at magnitudes consistent with the pre-crisis literature. But a multitude of other factors have also been counteracting and ultimately dominating this deficit effect over the same period.

The conventional economic story is that deficits push up interest rates, which then bear down on investment…

Let’s start in the early- to mid-2000s. In those years before the Great Recession, the conventional economic thinking was that a key tradeoff of deficit-financing a change in fiscal policy was that it would put upward pressure on long-term interest rates.¹ ²

That’s because a fall in public savings (another way of describing an increase in the federal deficit or, equivalently again, a fall in the government surplus) would have to be offset in an accounting sense by some combination of either a rise in private savings, a fall in private investment, and/or an increase in net capital inflows from foreign investors:

This relationship is an accounting identity, and while accounting identities are not causal models, they must hold true by definition. One way to re-balance this relationship after a decline in the government’s budget would be through higher interest rates, which might increase private savings and reduce domestic investment. I should note here that it’s not the only way to balance this relationship: if, to provide one example, there were a global savings glut, and/or high demand worldwide for safe assets, then an increase in deficits might be offset by a rise in foreign capital flows into the US as hungry foreign investors snatch up the new Treasuries being issued.

Economists worry that higher interest rates would be a drag on economic activity and investment, particularly over time. Lower investment means a lower productive capacity for the economy and, therefore, slower wage growth, slower rises in consumption, and overall just a more sluggish pace in the rise in well-being over time.

And the link between the fiscal trajectory and interest rates was more than just a theory. A broad empirical literature supported this hypothesis, including Feldstein (1986), Wachtel & Young (1987), Cohen & Garnier (1991), Elmendorf (1993), Canzoneri et al (2002), Gale & Orszag (2003), Engen & Hubbard (2004), Dai & Philippon (2005), Laubach (2003), and Warnock & Warnock (2009). Some analyses like Ardagna et al (2004) and Faini (2006) even attempt to exploit international variation, though this often proved difficult without the consistent and repeated projections of an independent fiscal monitor like CBO. Other analyses are less motivated by fiscal policy per se and more concerned about the dynamics of Treasury supply. Krishnamurthy & Vissing-Jorgensen (2012), for example, finds that an increase in Treasury supply lowers the safety and liquidity advantages of Treasuries over AAA corporate debt (what they term the “convenience yield):

…but this story doesn’t match a quick eyeball of the recent historical data.

But at first glance, the actual economic data seem to contradict this conventional framework.

Way back in March 2001, when the economy peaked at the precipice of the dot com bust, the 10-year Treasury traded at 4.9%, we were running a structural surplus of 2% of GDP, publicly-held federal debt stood at 33% of GDP, and CBO was projecting the entire debt would be paid off within a decade. In other words, both the economy and the fiscal trajectory were extraordinary.

Today, as we inch ever close to another business cycle peak, the structural deficit is now 5% of GDP, debt is 76% of GDP, and CBO projects that it will reach 105% of GDP in a decade if current policies stay in place. Today’s fiscal trajectory is almost a photographic negative of the outlook from early 2001.

And yet, the 10-year Treasury now sits more than 2 percentage points lower than it did back in March 2001.

To be fair here, the pre-crisis conventional thinking on government debt was never as simple as the overly-blunt “debt is always and everywhere bad”. But it is also probably fair to say that many a public finance expert circa 2001 would have been gobsmacked had a visitor from our time gone back and told them the debt and interest rate outcomes prevailing near full employment two decades later.

The conventional story may have been wrong all along. Or maybe other factors have been pushing down interest rates.

What explains this mystery? I think that there are two possibilities.

One is that this conventional framework is wrong, full stop, and deficits and debt have no relationship at all with interest rates.

A second possibility is that fiscal policy does affect interest rates, and that rising deficits since 2001 have put upward pressure on them, but countervailing downward pressures have dominated this fiscal effect.

Long-term interest rates, after all, are influenced by a complex litany of factors. The Council of Economic Advisers in 2015 published a helpful overview of what drives long-term interest rates, drawing on both theoretical and empirical insights.

So what factors have arisen since 2001 that may have offset any possible upward interest rate pressure from growing deficits? The CEA report mentions two candidates: an aging population (demography) and slower productivity growth. Nonfarm business productivity growth was 1.0% in 2017, versus 3.1% in 2001. And as Baby Boomers have aged, the share of the US population that’s 65 or older has risen 3 percentage points over the same period.

In mainstream economic models and assumptions, lower trend productivity growth is associated with lower real interest rates because it implies lower consumption growth over time. Slowing population growth may also feed into this by reducing the incentives to invest in R&D and innovation since the return to scale of new knowledge is lower, which in turn implies less adoption of productivity-enhancing innovations. An older population is also more weighted to high-saving middle-aged cohorts than high-spending younger cohorts, which puts downward pressure on interest rates. Some economists see these phenomena as evidence that the US is going through a period of “secular stagnation”, or persistent shortfalls of aggregate demand.

Another potential factor is foreign demand for US debt, which was already rising in 2001 and since then has grown to historically high levels, though in the last couple of years foreign demand has fallen off. Foreign investors both official and private were more prominent purchasers of our debt in the post-Great Recession debt expansion than they were in the 1980s debt expansion.

Economists Ricardo Caballero and Arvind Krishnamurthy have shown that foreign demand for US Treasuries has become an important factor in our own interest rates. As growth has accelerated in emerging market economies such as China and India, financial institutions in these countries are increasingly looking to hold safe assets as leverage. Investors consider US Treasuries to be among the safest assets in the world thanks to the US’s economic largesse, its stability, and the fact that they are denominated in US dollars, the paramount world reserve currency and over which the US has sovereign control.

In other words, if there’s increasing thirst for US debt from foreign investors, that could put downward pressure on our interest rates. Identifying any impacts of federal debt on interest rates means it’s important to account countervailing effects from foreign demand and capital flows.

Warnock & Warnock (2009) provides a simple empirical framework for a first approximation of the effects of deficits and foreign demand.

One example of an empirical framework that has already accounted for this is Warnock and Warnock (2009) which lays out a simple empirical approach to estimating the joint effect of foreign capital flows and deficits. They model 10-year Treasury yields as a function of inflation and growth expectations, current policy rates, variation in rates (to capture risk), cyclically-adjusted deficits, and foreign purchases.³ They then estimate this model using ordinary least squares (OLS).

The advantages of the Warnock approach make it a good starting place for a first approximation of fiscal effects on the margin, keeping the limitations of OLS in mind here.⁴

The Warnocks’ original analysis looked at monthly data from January 1984 through May 2005; in my update, I extend the analysis through December 2018. I use all the original measures from the original paper⁵, plus I also add in measures of growth in the dollar and the elderly population share:

Demographics will partially be captured by changes in growth expectations, but that will also reflect changes in underlying productivity, so to more fully capture the effects of aging I explicitly include changes in the elderly population share. I also include growth in the US dollar’s value to help adjust for the position of the US relative to other world economies and capture another influence on the relative favorability of US assets to foreigners that may not be fully explained by the interest rate alone.

It turns out, deficits have indeed put upward pressure on interest rates, but other factors like demography and foreign demand have more than made up for deficit effects since 2001.

The first set of results are below. As you can see, the deficit variable, cadef1, is positive and highly significant, suggesting that for every 1 percentage point increase in the structural federal deficit (relative to potential GDP), the 10-year Treasury yield rises by around 18 basis points, adjusting for all other factors. This is quite close to the estimates found in other studies, such as Laubach (2003), Gale & Orszag (2004), and Engen & Hubbard (2004).

However, the results also show that other factors can more than offset the impact of higher deficits. If for example that same rise in the structural deficit were accompanied by a fully equal 1 percentage point of GDP rise in foreign purchases (foreign2g), the net result would be a decrease in the 10-year yield of 10 basis points, adjusting for all other factors.

Indeed, even though federal deficits have been putting upward pressure on interest rates since 2001, a litany of other changes have been lowering the 10-year, among them conventional monetary policy⁶, demographic shifts, volatility, and foreign purchases.

These fiscal effects remain statistically significant, and consistent with the prior literature, when using debt instead.

There isn’t a consensus even in economic theory, let alone in economic practice, as to whether the deficit or debt is the more relevant metric for interest rates. So I respecify the model to use ongoing total publicly-held federal debt stock as a percent of GDP (phfdgdp) and federal debt held overseas by foreigners, also as a percent of (US) GDP (foreigndebtgdp), rather than the deficit and purchases flow measures.

The results below show that debt stock and foreign holdings also have statistically-significant effects in my update of the Warnock model.⁷ Each percentage point increase in debt-to-GDP raises the 10-year yield by 4.2 basis points, all else equal. This is close to the effect estimated in Laubach (2003) and Gale & Orszag (2004). Meanwhile each percentage point increase in foreign holdings relative to US GDP lowers the 10-year by 11.2 basis points.

Deficit effects on interest rates have been falling over time.

I perform three additional alternatives analyses for robustness.

The first explores the possibility that the effect of deficits and debt on interest rates may have been changing over time. The core Warnock OLS models are not set up to answer this question, but we can modify them by running the regressions on a recursive sample; that is, keeping the sample start fixed at 1984 and gradually expanding it to the present day to see how the deficit coefficient changes.

The results below show that, indeed, the marginal effect of structural deficits on interest rates have fallen substantially over the past 25 years. Over the 1984–94 period, my modification of the Warnock approach suggests that each percentage point of debt-to-GDP raised the 10-year yield by around 65 basis points. But expanding the sample beyond the late 90s, this effect began declining until it plateaued at around 20 basis points beginning in 2010 (and then saw a temporary rise between 2011–2015). In all cases these effects are statistically significant.

Deficit effects are almost identical when using CBO fiscal expectations instead.

The second alternative analysis swaps out my monthly ongoing structural deficit measure with the five-year-ahead deficit-to-GDP projected by CBO under their current law baseline. This is the same measure used in Laubach (2003) and similar analyses. Since the 10-year is a forward-looking instrument, arguably deficit expectations rather than coincident deficit outcomes are the more germane measure of the fiscal trajectory. However, CBO only updates its projections infrequently, typically in January and August, and there are no other consistent or publicly-available fiscal forecasts over time. This severely curtails my sample to 72 months over the last 35 years. Nevertheless, as you can see below the marginal effect of deficit expectations on the 10-year is almost exactly the same as in my baseline model.

Deficits and foreign purchases primarily affect interest rates through the ACM term premium.

Finally I go back to my original structural deficit measure and ask a different question: can we shed light on whether the deficit effect on interest rates flows through policy expectations (i.e. the market expects that the Fed will offset deficit-driven inflationary effects) or through the term premium (i.e. deficits raise the risks associated with holding longer-term securities).

Policy expectations and the term premium are unobservable concepts; one needs a term structure model to decompose the 10-year yield into these two component parts. The breakdown of Adrian, Crump, and Moench (2013, “ACM”) has the advantage of 1) being published by a reputable source (the Federal Reserve Bank of New York), 2) seeing frequent updates, 3) having a deep history going all the way back to the 1960s, and 4) being widely used by market participants and policymakers alike.

The ACM model however is far from the only approach to decomposing the yield curve, nor is it without its critics. In the end, all this analysis can tell us definitively is how the deficit effect breaks down within the framework of the ACM model, not how it affects the term premium in any broad, universal sense.

What the results below show is that deficits and foreign purchases entirely affect the 10-year through the ACM term premium (facm10tv) rather than through ACM policy expectations (rn10). Every 1 percentage point increase in the structural deficit increases the ACM term premium by 20 basis points and lowers policy expectations by 2 basis points, though the latter effect is not statistically significant. Likewise, a 1-percentage-point-of-GDP increase in foreign purchases of US Treasuries lowers the term premium by 31 basis points and raises policy expectations by a statistically insignificant 3 basis points.⁸

Given that the Warnock approach is partial equilibrium, not dynamic, the way to interpret these results is that there are two channels by which deficits affect interest rates. The first is the policy channel: a fiscal impulse may put upward pressure on interest rates by raising growth expectations, inflation expectations, and/or, in a regime where a central bank operates under an inflation target, raising expectations that the central bank will offset these effects with higher policy rates. The ACM regressions do not show a significant effect of deficits on policy expectations, but since the specification adjusts for the inflation and growth expectations channels as well as the policy rate channel, the lack of a statistically significant effect from deficits is not surprising.

But the other channel is a risk channel. Higher deficits may heighten tail risks of negative outcomes: of higher taxation or greater spending cuts in the distant future, lower liquidity of Treasury markets, or of higher inflation or higher policy rates. There may also be a portfolio rebalancing effect that is driven by Treasury supply in the domestic market, similar to one of the mechanisms commonly theorized in the QE literature (and which would explain why foreign demand also operates exclusively through the term premium to counteract these effects). There may even be some effect from an increase in default risk stemming from higher deficits, though given that the US is a sovereign issuer in its own currency it’s not obvious this would be meaningful in magnitude.

Conclusions

So where does this analysis leave policymakers? I think there are nine major takeaways here:

1. In terms of simple rule-of-thumb tradeoffs, the evidence suggests pre-crisis consensus was largely correct: an increase in structural budget deficits puts upward pressure on interest rates, all else equal.
2. However, all else may not always be equal, particularly in periods where there is high foreign demand for US debt and/or high demand for safe assets in general. Over the last 20 years, this has helped blunt the effects of deficits on interest rates. By the same token, if foreign / safe asset demand dynamics subsided in the future, then structural deficits could shift to having a larger net upward effect than they have in recent years.
3. These deficit effects appear to have fallen substantially over the last two decades, but are still positive and statistically significant.
4. The upward interest rate effect from structural deficits and the countervailing downward effect from foreign purchases both appear to flow materially through term/risk premia at least as measured by the ACM model, though given that the ACM model is not the final word on term structure decompositions this is not dispositive. Still, this finding makes it more likely that the interest rate impacts of higher deficits reflect more than just policy choices by the Federal Reserve.
5. Other structural and cyclical factors can depress interest rates as well. In the US context, these factors have more than offset the influence of higher deficits over the last 20 years. Some examples include the aging of the population and slower productivity growth.
6. Policymakers may exploit the interest rate room carved out by these countervailing factors by running higher deficits, particularly as a means of financing policies designed to combat such factors as e.g. secular stagnation.
7. Because these structural factors are unlikely to reverse themselves suddenly, and because US interest rates are already relatively low given the state of the business cycle, deficits and debt should not be a near-term worry, nor should they impede counter-cyclical policy when the economy is far from full capacity.
8. However, deficit-financing still poses a tradeoff in the form of upward interest rate pressure. This has implications for investment, productivity, wages, and consumption in the long-term that policymakers ought to weigh against the specifics of the policy being financed.
9. These are distinct questions from many other recent fiscal policy debates; for example, whether there are states of the economy, such as when r < g (interest rates are lower than the rate of economic growth), when debt-financing certain policies is more likely to be welfare-enhancing. Note though that my analysis implies that deficits and debt can be endogenous to the determination of r over the medium- to long-run. In other words, the more that policymakers exploit an r < g state and deficit-finance changes in fiscal policy, the less likely it will be that the r < g state persists over the medium- to long-term.

[1] ^ Throughout this post I will use the term “interest rates” to refer to long-term interest rates, and in particular the 10-year nominal Treasury.

[2] ^ There is no consensus in economic theory on whether the deficit or the debt stock is the more relevant metric for interest rate determination. Elmendorf & Mankiw (1999) is a good overview of the conventional thinking going into the 21st century. To name but two examples, the neoclassical growth model equates the real interest rate to the marginal product of capital, so the question in that context hinges on the extent to which federal debt competes with private capital. The Keynesian IS/LM model meanwhile determines the real interest rate via the flow equilibrium of aggregate demand and supply; in that model, then, it is the deficit that affects interest rates through a shift in the IS curve. The IS/LM model is strictly-speaking a short-run model, though with stronger assumptions about inflation dynamics it is sometimes used as a longer-run framework as well. Since deficits and debt are highly correlated, particularly the expectations of each, in practice it proves difficult for an empirical model to arbitrate between these views. Some economic rubrics see the choice of policy financing as not relevant to long-term interest rates at all.

Throughout this blog post, my analysis will focus on flow measures such as the deficit

[3] ^ The authors impose the restriction based on Mehra (1998) that the coefficients on the 3M rate and long-run inflation expectations must sum to 1. This restriction ensures that these variables are cointegrated with the 10-year Treasury yield, which is necessary in OLS estimation since the 10-year yield is nonstationary.

[4] ^ The Warnock specification has advantages and disadvantages. On the plus side, it’s a relatively simple specification with minimal structural assumptions, as well as independent variables that are widely-understood and interpretable. Because it uses monthly ongoing data, it doesn’t suffer from the severely limited sample sizes of approaches that rely on CBO or OMB projections, like Laubach (2003). The downside of the Warnock analysis is that theirs is not a dynamic approach and can only give us time-invariant partial equilibrium effects; the Warnock model cannot tell us, for example, how the other independent variables might react to a deficit or debt shock and how these effects net out in general equilibrium. Further work in this area should explore dynamic approaches to the question, e.g. such as by estimating a VAR model.

[5] ^ Unlike the original Warnock paper I do not lag the structural deficit measure.

[6] ^ It’s also plausible that unconventional monetary policy actions such as forward guidance and large-scale asset purchases (Quantitative Easing/Tightening) have affected interest rates. Thoroughly and properly adjusting for the effects of each of these actions is a priority of future research.

[7] ^ None of the results show evidence of a spurious relationship; the augmented Dickey-Fuller test strongly rejects the null hypothesis that the residuals of each model are nonstationary.

[8] ^ Note that the ACM decomposition is linear: the sum of the term premium and policy expectations terms equals the actual observed 10-year Treasury yield. Since I distribute the Warnock model restrictions between the term premium and policy expectations regressions, then, the sum of the coefficients on each variable across both regressions will equal the coefficients on those same variables in my original 10-year Treasury specification.