Weekend Reading: Discussion of J. Bradford DeLong and Lawrence H. Summers: "Fiscal Policy in a Depressed Economy
Robert Hall observed that a better title for the paper would be “Eta,” since the paper’s surprising results all stem from the authors’ beliefs about the value of their hysteresis parameter η. The other parameter values the authors used for their simulations seemed mostly reasonable and uncontroversial to Hall. He noted that although Valerie Ramey had estimated a relatively low value for the multiplier on fiscal spending, the standard error on her estimate was large and did not rule out the possibility that the authors’ baseline value of 1.5 was correct. Hall also observed that some alternative ways of analyzing government spending data from World War II generated higher estimates of the multiplier. He found the authors’ value for the growth rate reasonable, and although he shared Ramey’s concern about the authors’ real interest rate assumptions, he thought their baseline value might be reasonable as well.
For the most important parameter, η, however, Hall felt that much more work was needed to arrive at a credible estimate. He noted that for the interesting cases in the authors’ analysis, r - g is small, which makes the present value of extra output due to avoided hysteresis significant for decades into the future. In such cases, then, the appropriate value for η would be an average not just over the next decade but over many decades.
Econometrics, however, simply cannot answer the question of whether hysteresis effects, or the effects of avoided hysteresis, are significant far into the future, Hall argued. The “unit root” literature of the late 1980s had found that it was impossible to precisely estimate the persistence of shocks to GDP, yet small differences in the persistence of government spending shocks had very different implications for the analysis. And even if it were possible to estimate the long-run effects of such a shock, the United States has not experienced a government purchases shock in many years. The 2009 stimulus package did not constitute such a shock, as the positive effect of the package on government purchases was slightly more than offset by negative effects from other sources. Hall was skeptical of the suggestion of comments and discussion using CBO revisions to GDP estimates as a source for estimating h, since these revisions were not made in response to fiscal shocks. Finally, Hall thought that in their discussion of η, the authors had written with great enthusiasm about the forces that might make η strongly positive while neglecting forces that might make it less positive or negative.
Eric Swanson thought the authors were correct to use the long-term interest rate at which businesses and households can borrow as the relevant one for their analysis of the fiscal multiplier. However, he felt that the authors’ view of monetary policy was too narrow in that it equated the stance of monetary policy with the current level of the federal funds rate. In his view a better measure was a medium-term interest rate such as the two-year Treasury yield, which reflects not just where the federal funds rate is at present, but also where it is expected to go over the next several quarters. This view was shaped by discussions Swanson had had with Brian Sack and research he had conducted with Sack and Refet Gürkaynak, in which they found that Federal Open Market Committee statements had large effects on the yield curve above and beyond the direct effect of changes in the federal funds rate. These effects seemed to be driven by changes in financial market expectations about the future path of the funds rate in response to forward-looking language in the FOMC statement. Thus, the FOMC appeared to have the ability to directly affect medium-term interest rates such as the two-year Treasury yield through its statements, consistent with the Eggertsson-Woodford view of monetary policy.
Taking the correct view of monetary policy mattered, Swanson continued, because if it was true that a better measure of monetary policy is the two-year Treasury yield, it was unclear that the zero lower bound was a meaningful constraint on monetary policy in 2008, 2009, and much of 2010, since yields at that maturity were consistently between 80 and 100 basis points during that period. Away from the zero lower bound, government spending can crowd out private investment by raising interest rates.
Indeed, in a recent working paper with John Williams, Swanson had found that between 2008 and 2010, two-year Treasury yields were just as sensitive to economic announcements as they had been in the 1990s, when the zero bound was clearly not a constraint. Thus, the fiscal multiplier was likely to have been no larger than normal during this period. By late 2011, two-year Treasury yields had fallen below 30 basis points and were then, according to the same working paper, about half as sensitive to economic news as in the 1990s, suggesting that crowding-out effects were half as important as usual. Even in late 2011, however, corporate yields remained as sensitive to economic news as in earlier years, suggesting that crowding out might be as important as ever. What appeared to be driving this effect was that until very recently, the private sector appeared resolute in its belief that a recovery or inflation or both would occur within the next few quarters, forcing the Federal Reserve to raise interest rates. The most recent FOMC commitments to keeping rates low through mid-2013 and late 2014 appeared to have finally altered those beliefs.
Arvind Krishnamurthy, citing work he had done with Annette Vissing-Jorgensen, thought one reason Treasury yields were currently so low was that a global shortage of safe and liquid assets had led investors to place a high safety and liquidity premium on Treasury bonds. Krishnamurthy warned that this premium could easily disappear in the next few years, which made it risky for the government to finance significant new spending with short-term bonds. Financing with long-term bonds would avoid this risk but would command higher interest rates, reflecting market expectations that the safety and liquidity premium on Treasury bonds would disappear over time. These higher rates, he thought, were the ones the authors ought to consider in their analysis. Pointing to his work with Vissing-Jorgensen on estimates of the elasticity of interest rates to the size of the government’s debt, Krishnamurthy suggested that the authors should account for the likelihood that increasing government debt would push up interest rates.
Justin Wolfers endorsed Hall’s view that η was the paper’s key parameter. He noted that the economics profession knows amazingly little about the degree and extent of hysteresis and that exploring the policy implications of this uncertainty was itself an interesting exercise. He thought that the nonzero probability of η being positive made Summers and DeLong’s analysis worthwhile. Today’s high long-term unemployment rate made Wolfers worry that η was indeed positive, although he felt the paper should equally consider forces that might make η low or negative, including those identified by Ramey.
Wolfers also suggested that uncertainty about η had equally large implications to explore for monetary policy. Economists have changed their thinking about monetary policy after realizing that inflation can cast a long shadow as it shifts markets’ long-run inflation expectations, and the potentially long shadow of unemployment and other forms of hysteresis could be similarly important.
Martin Baily hoped the authors would do more to tie their analysis to the current situation in the United States. Specifically, he wondered whether their model implied that the federal government ought to undertake more stimulus now, even though politics made that possibility extremely unlikely. Baily had been worried that more spending would push the nation over a fiscal cliff, and he was therefore intrigued by DeLong and Summers’s suggestion that more spending today could actually reduce future deficits.
Baily also said that he agreed that the stimulus had improved the economy relative to the outcome with no stimulus, but it was surprisingly hard to show that the 2009 stimulus bill had been instrumental in turning the economy around from a decline in annualized GDP growth of 8.9 percent in the fourth quarter of 2008 to almost 4 percent positive growth in the fourth quarter of 2009. The automatic stabilizers, he thought, might have played at least as large a role in the turnaround, and growing exports and inventories had also contributed substantially. Meanwhile government purchases had not increased much over that period, and disposable income, which the stimulus bill had explicitly sought to boost, had continued to fall.
Ricardo Reis agreed strongly with Hall and Wolfers that this was a paper about h. Most macroeconomists who had studied the unit root literature of the late 1980s equated that root with technology shocks. But Reis saw little to no evidence that the sole determinant of long-run GDP was technology. He viewed η, then, as a parameter that could capture the effects of various short-run interventions on long-run output, where different types of interventions have different values of η, and the higher its η, the larger an intervention’s impact. He thought the authors should do more to explain their view that government spending, in general, has a high η, and that they would do well to differentiate among the ηs of different types of government spending, such as tax rebates, infrastructure spending, or investment in science research. Reis also agreed with Ramey that the authors ought to develop a model of what determines η, although he was indifferent as to whether this model was a dynamic stochastic general equilibrium model or something else. He noted that in recent years Philippe Aghion had made some progress in this area, modeling the mechanism through which recessions can induce lower research and development spending and so reduce potential output, as well as empirically measuring the long-run scars of recessions through R&D spending.
Christopher Carroll agreed with Hall that it was impossible to test for the existence of long-run hysteresis effects in aggregate data but was optimistic about the possibility of measuring it at the micro level. As an example, Steven Davis and Till von Wachter, in a paper in the previous issue of the Brookings Papers, had found that individuals laid off during recessions suffer long-run earnings losses over 50 percent larger than those experienced by individuals laid off during expansions. Carroll found this result suggestive of important long-run hysteresis effects. Hall disagreed, however, saying his view was that the result was driven by selection effects.
Carroll also singled out a result from the Stock and Watson paper, presented before the Panel the previous day, that he found striking: One of the major reasons the recent recession appeared different from previous ones was that the economy effectively suffered a major negative monetary policy shock when interest rates hit the zero lower bound. Because of this practical barrier to negative interest rates, the federal funds rate remained stuck at a value that was two or three percentage points too high. Once one accounted for this shock, the macrodynamics of the economy were similar to those of previous recessions.
This result led Carroll to wonder whether automatic fiscal stabilizers had played a more important role than previously understood in minimizing the long-term damage from economic downturns, and if so, whether in light of the recent experience it would make sense to put even more of these stabilizers in place, to further reduce the odds of hitting the zero bound. Additional automatic stabilizers might also help circumvent the political challenges of passing one-time stimulus bills in a timely manner.
Olivier Blanchard agreed with Hall that the Congressional Budget Office’s revisions to its economic forecasts could shed no light on hysteresis. In fact, all these revisions meant was that the CBO had discovered that the economy had experienced supply shocks, which affect long-run GDP growth, not that transitory shocks were having permanent effects. Blanchard also thought that pure time-series econometric techniques could not, on their own, be used to test for the existence of hysteresis, since they cannot distinguish between the effects of permanent shocks, such as a permanent increase in the price of oil, and the permanent effects of a transitory shock, such as an increase in consumption due to rising animal spirits. Finally, Blanchard thought it was a misnomer to call η a “hysteresis” parameter, since hysteresis implies a permanent effect, whereas the authors’ analysis would still hold if the effects of η were long-lasting but not permanent.
Betsey Stevenson concurred with Reis that different types of government spending could have very different effects and that it was important to distinguish among them. Building on Carroll’s point about automatic stabilizers, she wondered what types of stabilizers might be most desirable. For example, should education spending rise automatically when unemployment rises? Or would it be a better idea to spend more on highways?
Michael Woodford agreed with Blanchard that it was impossible to iso- late evidence of hysteresis in revisions to potential output during recessions. He saw the problem as fundamentally an issue of signal extraction: Even if short-run deviations from potential output had no effect on long-run potential output, one would expect to find a correlation between short-run deviations below full output and lower long-run potential output. And even if deviations from potential output really did cause long-run potential output to fall, government purchases during a recession would not necessarily prevent that. The channel through which deviations from potential output decrease long-run output might, for example, be decreased investment, in which case more government purchases would not improve long-run potential output; indeed, if these purchases negatively impact private investment, they might lower potential output. Finally, Woodford thought the authors were wrong to assume that hysteresis effects, if they exist at all, are permanent. He encouraged the authors to test the sensitivity of their results to different lengths of persistence of hysteresis.
David Romer agreed with those panelists who thought the authors should develop a more fully specified model, and he cited an example of an aspect of the economy that their model ignored that could affect their results. The authors had not specified what determines the rate of inflation. Suppose that it is determined by an accelerationist Phillips curve. In that case, higher output today would lead to higher inflation in the long run unless the Federal Reserve restricted output at some point in the future, and that, Romer suspected, would undermine the authors’ result that fiscal spending at the zero lower bound has long-run benefits. Romer doubted that, in fact, an accelerationist Phillips curve characterized the U.S. economy today, but the example indicated that it would be prudent for the authors to lay out a fuller set of assumptions about the structure of the economy. Without a more complete model, readers are left to guess about the conditions under which the paper’s results apply.
Martin Feldstein agreed with the authors that some aspects of cyclical weakness could have a negative influence on long-run potential output. For example, workers’ skills might decay when they are out of work, and business productivity might be harmed when their capital is underutilized. However, he thought it important also to consider aspects of cyclical weakness that might work to increase long-run potential output. In the current economy, for example, one reason for the persistently low labor force participation rate was that many younger people were staying in school and presumably building human capital. There was also evidence that many businesses, after realizing the downturn would last a long time, had taken the opportunity to figure out how to produce more with fewer workers, which could translate into long-run productivity gains.
Responding to the discussion, Lawrence Summers said he was well aware of the costs of long-run budget deficits that Feldstein had stressed in his formal comment, and he underlined that the key to his and DeLong’s argument was that, if their model is correct, short-run fiscal expansion actually reduces, not increases, the long-run deficit. He thought that investors in southern European countries appeared to be placing real stock in such a possibility today: concerns about long-run growth and competitiveness, which could be improved by fiscal expansion—and not concerns about fiscal profligacy—seemed increasingly to be a factor driving up interest rates on government bonds in the region. The fact that Spain had a smaller budget deficit than Germany just four years ago also lent support to the view that the prospect of weaker long-run growth, not deficits, was weighing most heavily on the minds of investors these days.
Summers also clarified that he and DeLong had not meant to argue that, so long as the interest rate is less than the growth rate, government ought to spend limitlessly—a view that Summers saw as akin to the fallacy that since stocks, on average, return more than bonds, one ought to try to borrow without limit to buy stocks. Rather, he and DeLong had tried to demonstrate that a combination of hysteresis and multiplier effects made government spending look attractive under economic conditions like those prevailing today. Responding to Krishnamurthy, Summers said that he and DeLong had not intended to suggest a permanence to the liquidity premium on government bonds as grounds for fiscal expansion.
Summers said that he and DeLong would certainly study Ramey’s work on estimating the fiscal multiplier. However, he questioned how well the multiplier could be identified by examining the World War II period, since so much was going on in the economy alongside the burst of government spending at that time.
Summers agreed with Swanson that the federal funds rate is not a sufficient statistic for monetary policy and that, in theory, if the Federal Reserve responds to expansionary fiscal policy by doing less quantitative easing or less forward-looking signaling, it will offset expansionary policy much as it does in normal times. Summers thought that he and DeLong should consider the evidence Swanson had referenced suggesting that monetary offsetting of fiscal expansion was a real concern.
Summers agreed with the panelists who had suggested that the parameter η was of central importance to the paper. Responding to Reis and Stevenson, he did not doubt that the marginal productivities of, for example, investing in infrastructure and investing in the National Science Foundation differ in interesting and important ways, but he thought it beyond the scope of the paper to compare their long-run supply-side effects. The real impetus for the paper, Summers said, was to explore the implications of loosening an assumption that had become popular following the unit root debate, which was that only technology shocks have permanent effects on long-run poten- tial output. An alternative view, that reductions in aggregate demand could also have protracted effects, seemed to have substantial enough implications for optimal fiscal policy that it was worth modeling more carefully.
On whether it made sense to use revisions to CBO forecasts to test for hysteresis, Summers thought that if one added the identifying assumption that fluctuations in output between 2007 and 2010 were driven by aggregate demand shocks and not aggregate supply shocks, one could then develop an estimate of the hysteresis parameter using the CBO revisions. One might develop such an identifying assumption by more carefully specifying resource strengths.
Summers thought it reasonable to question the permanence of hysteresis effects and appreciated the skepticism that panelists had expressed about results that hinged on the weight of outcomes 75 years in the future. In fact, he and DeLong had included a decay rate on the effects of η in earlier drafts of the paper and might in the final draft either reinstate it or include an examination of debt-to-income ratios after 15 years.
Finally, Summers disagreed with Feldstein’s view that current deficits could significantly reduce private investment by making investors worry about higher future taxes on corporate profits and personal income. The value of new capital investment is closely related to the value of existing capital, and the strong growth in market value of existing capital over the past two years made it seem implausible to Summers that concerns about tax hikes were reducing investors’ expectations about the future profitability of capital.
#weekendreading #fiscalpolicy #macro #economicsgonewrong