Can Government Purchases Stimulate the Economy?

“I conclude that the multiplier for this type of spending [a temporary, deficit-financed increase in government purchases] is probably between 0.8 and 1.5” (Abstract, p. 673), a conclusion the author restates in the body text: “the range of plausible estimates for the multiplier in the case of a temporary increase in government spending that is deficit financed is probably 0.8 to 1.5” (Section 3, p. 680-681). She is explicit about the uncertainty surrounding this range: “reasonable people could argue that the multiplier is 0.5 or 2.0 without being contradicted by the data” (p. 681), and that if the spending increase occurs during a severe recession, estimates are likely to sit at the upper end of the 0.8-1.5 range (p. 681).

Because it “depends very much on the type of government spending, its persistence, and how it is financed” (Introduction, p. 673). She restricts her headline estimate specifically to spending that is temporary, deficit-financed (rather than tax-financed), and that “enter[s] separately in the utility function and have[s] no direct effect on private sector production functions” – i.e., not productive public investment or transfers, both of which she treats separately later in the essay (Introduction, p. 673).

In neoclassical models, government spending affects output only through its effect on equilibrium hours worked, which operates through a negative wealth effect: a rise in g reduces the value function’s implied wealth, and because leisure is a normal good, “the function h is strictly increasing in g” (Section 2.1, p. 674, following Aiyagari, Christiano, and Eichenbaum 1992). Absent distortionary taxes, Ricardian equivalence means financing is irrelevant; but Baxter and King (1993) show that once distortionary financing is introduced, the multiplier can be “as low as negative 2.5” for a temporary spending increase financed by concurrent distortionary taxes, “substantially below unity” for temporary spending financed by future lump-sum taxes, and around 1.2 in the long run for a permanent increase financed by lump-sum taxes, because “the greater negative wealth effect raises labor supply more and the steady-state capital stock rises” (p. 675). Burnside, Eichenbaum, and Fisher (2004) further show that when distortionary taxes rise only gradually (a hump-shaped path, as observed historically), intertemporal substitution toward present labor supply pushes short-run multipliers higher than under the lump-sum case (p. 675).

Standard New Keynesian models typically generate multipliers smaller than the traditional Keynesian cross would predict, “since the New Keynesian model builds a sticky-price edifice on a neoclassical foundation,” and Cogan et al. (2010) using the Smets-Wouters model find multipliers at or below unity (Section 2.2, p. 675-676). Galí, López-Salido, and Vallés (2007) obtain multipliers as high as 2.0, but only by assuming at least 50 percent rule-of-thumb consumers and fully demand-determined employment – assumptions that “essentially convert the New Keynesian model back into a traditional Keynesian model” (p. 676). The one channel that can raise multipliers without resorting to such assumptions is the zero lower bound: “a deficit-financed increase in government spending leads expectations of inflation to increase. When nominal interest rates are held constant, this increase in expected inflation drives the real interest rate down, spurring the economy,” and Christiano, Eichenbaum, and Rebelo (2011) show that if the nominal rate is held constant for twelve quarters while spending rises, the multiplier peaks at 2.3 (p. 676).

Ramey flags three omissions – productive government spending, transfers, and underutilized resources – and finds mixed evidence that any of them substantially raises short-run multipliers (Section 2.3, pp. 676-677). Baxter and King (1993) find that public capital raising private marginal products can generate long-run multipliers of 4.0 to 13.0, “but much lower in the short run,” so it “does not raise the predicted short-run stimulus effects.” Because most of the 2009-2010 U.S. stimulus was allocated to transfers rather than purchases, and standard models treat temporary transfers like a negative lump-sum tax with no effect under the permanent-income hypothesis and Ricardian equivalence, Oh and Reis (2011) explore relaxations of these assumptions but are “not able to generate much bigger effects” (p. 677). Whether underutilized capacity raises multipliers is, in Ramey’s assessment, “a promising area for more theoretical research” rather than a settled question (p. 677).

“Despite significant differences in samples, experiments, and identification methods, most aggregate studies estimate a range of multipliers from around 0.6 to 1.8,” with within-study ranges “almost as wide as the range across studies” and consistently large standard errors (Section 3, p. 679, Table 1). The evidence spans Evans’s (1969) large-scale Keynesian econometric models (multipliers “slightly above 2.0”), military-spending-instrument studies by Barro, Hall, and Barro-Redlick (0.6 to 1.25), Ramey-Shapiro-style narrative/anticipation-based VARs (0.6 to 1.57), Blanchard-Perotti SVARs (0.9 to 1.29), and Auerbach-Gorodnichenko’s regime-switching estimates, which imply multipliers around 2.2 in recessions versus -0.3 in expansions under one specification, and a more plausible 1.0-1.5 in recessions versus 0-0.5 in expansions when regimes are allowed to switch endogenously (pp. 679-680).

She “truncated the lower estimates because they were usually accompanied by increases in distortionary taxes” – for instance, her own lower estimates come from samples dominated by the Korean War, which was substantially tax-financed, and Barro and Redlick (2011), controlling for the average marginal tax rate, find a multiplier of only 0.6 – “and truncated the very highest estimates because of the various concerns” raised earlier, including Fisher and Peters’s (2010) finding of an unusually persistent spending shock (which a neoclassical model would predict raises the multiplier above the value relevant to a genuinely temporary stimulus) and possible anticipation effects contaminating Gordon and Krenn’s (2010) 1940-41 estimates (Section 3, pp. 679-681).

No – restricting the sample to 1939-1949, a period during which Treasury bill rates never rose above 0.38 percent despite six percent average inflation, Ramey (2011) finds a multiplier of 0.7, “though with even larger than normal standard errors,” so “I find no evidence of larger multipliers during the extended period in which interest rates were held virtually constant at the zero lower bound” (Section 3, p. 680). This is presented as evidence against, or at least not clearly in favor of, the theoretical prediction that multipliers should be systematically larger when the zero bound binds, though the author notes the small effective sample size limits how much weight this particular comparison can bear.

Cross-state income-multiplier estimates mostly fall between 0.5 and 2.0, with an implied cost of roughly $25,000-$35,000 per job created across several studies (Section 4, pp. 681-683, Table 2), and several papers (Shoag 2010; Suárez Serrato and Wingender 2011; Nakamura and Steinsson 2011) find multipliers are significantly larger during periods of greater economic slack. But Ramey illustrates with a simple example why these are not directly the aggregate multiplier: if the federal government transfers $1 to one state, financed by lump-sum taxes spread across all states, “the true aggregate multiplier is 0, since the taxes and transfers cancel in the aggregate,” yet a panel regression with time fixed effects – which nets out the economywide tax increase – recovers a spurious multiplier of mpc/(1 - mpc), e.g., 1.5 if the marginal propensity to consume is 0.6, “even though the aggregate multiplier for this experiment is 0” (Section 4, p. 682). Translating local multipliers into an aggregate figure therefore requires an explicit theoretical model of the kind developed by Nakamura and Steinsson (2011) and Shoag (2010), which treat individual states as small open economies within a currency union.

Despite the proliferation of multiplier estimates, “there is still no consensus on the mechanism by which government spending raises GDP”: some studies find consumption falls in response to a spending increase, consistent with the neoclassical negative wealth effect, while others find it rises, consistent with rule-of-thumb consumers, and Nekarda and Ramey (2011) present evidence that industry markups do not fall in response to government spending as the New Keynesian model requires (Conclusions, p. 683). She also stresses that “none of these estimates sheds light on the welfare consequences of temporary increases in government spending to stimulate the economy,” which would require both a resolved mechanism and explicit assumptions about whether government purchases enter household utility (p. 683).