Measuring the Output Responses to Fiscal Policy

Questions & answers

Q1. What identification approach does the paper build on, and what three extensions does it add?

The paper’s “starting point is the classic paper by Blanchard and Perotti (2002),” which identifies government-spending and tax shocks in a VAR using the assumption that discretionary fiscal policy cannot respond to output within a quarter, combined with auxiliary estimates of the automatic (non-discretionary) response of taxes and spending to output. Auerbach and Gorodnichenko extend this in three ways: (1) a regime-switching VAR that lets the entire dynamic and contemporaneous response structure differ between recession and expansion rather than assuming one fixed multiplier, (2) disaggregating government purchases into components (e.g., defense vs. non-defense, consumption vs. investment) that “behave differently in relation to aggregate fiscal policy shocks,” and (3) using real-time forecast data to “purge fiscal variables of ‘innovations’ that were predicted by professional forecasters,” producing a more precise measure of genuinely unanticipated fiscal shocks.

Q2. How does the smooth-transition VAR (STVAR) distinguish “recession” from “expansion,” and why not simply split the sample?

The authors use a smooth-transition autoregression in which a continuous transition function F(z) determines how much weight the recession-regime versus expansion-regime coefficients receive at each date, allowing “not only differential dynamic responses but also differential contemporaneous responses to structural shocks.” Estimating separate linear SVARs for each regime directly would leave very few effective recession observations, making estimates “unstable and imprecise”; the smooth-transition structure instead pools information across the whole sample while still letting the estimated responses vary with the degree of slack, at the cost of biasing the estimates “towards not finding differential fiscal multipliers across regimes,” since each regime’s estimate is partly informed by dynamics from the other regime.

Q3. How large are the estimated government-spending multipliers in recession versus expansion, and how does this compare to existing theory?

In the baseline STVAR, “the impact output multiplier is about 0.5” in both regimes, but the responses diverge over time: in expansion the response soon falls below zero, while in recession it “rises steadily, reaching a value of over 2.5 after 20 quarters,” with the cumulative-multiplier metric giving “between 0 and 0.5 in expansions and between 1 and 1.5 in recessions.” The authors note this spread is “largely consistent with” both the old Keynesian IS-LM-AS intuition (where slack makes the aggregate-supply curve flatter) and recent zero-lower-bound DSGE models (Christiano et al. 2009, Woodford 2010) that predict multipliers of 3-5 once nominal rates are constrained, though their own estimates stay below that upper range.

Q4. Why might the true regime-dependent multiplier differ from the raw regime coefficients, and what does the historical simulation show?

Because the economy can transition out of a recession during the horizon over which a multiplier is measured, “the actual multipliers starting in recession (or expansion) should be a blend of those estimated for the separate regimes” rather than a pure recession or expansion number. Incorporating these endogenous regime shifts, the authors compute historical dollar-for-dollar multipliers that vary by date depending on how deep the prevailing slack is – finding, for example, that a spending increase in 1985 “would have barely increased output,” while the same dollar increase in 2009 “could raise output by about $1.75” – and note that the multiplier changes quickly once the economy starts growing out of a trough, which they say makes the timing of discretionary spending critical to its effectiveness.

Q5. How do multipliers differ across components of government purchases, and what does the military-spending result show?

Breaking spending into defense and non-defense purchases, the linear-model response to a unit defense-spending shock “rises immediately by just over 1 … and then gradually falls, becoming negative”; once split by regime, the defense-spending multiplier shows much stronger cyclicality, “peaking at nearly 4 in the fifth quarter after the shock” in recession but resembling the linear estimate in expansion. Non-defense spending’s regime difference instead shows up mainly in timing – a faster peak response in expansion, a delayed peak in recession – “starts smaller but eventually exceeds 1” in the linear model, but without the same swing in overall size across regimes. A separate breakdown into consumption versus investment purchases finds investment-type spending has a markedly stronger output effect than consumption-type spending in the linear model.

Q6. Why does the paper try to purge fiscal shocks of their predictable component, and what does this do to the estimated multipliers?

The authors argue that because much of what shows up as a VAR fiscal “innovation” may already have been anticipated by private agents, the standard SVAR shock is contaminated with predictable variation – so they augment the VAR with real-time government-spending forecasts from the Michigan RSQE model, the Survey of Professional Forecasters, and Federal Reserve Board staff projections, finding these forecasts “help explain a considerable share of the fiscal innovations.” Using Ramey’s (2009) narratively identified military-spending news shocks as a complementary check, the paper finds the recession-regime multiplier rises from “the impact multiplier of 0.8 and the average multiplier of 2.2” in the baseline VAR to “the multiplier on impact is about 2 … and the average multiplier over 20 quarters is 3.7” when isolating the unanticipated component – evidence the authors read as corroborating that unanticipated shocks have systematically larger effects in recessions than the full, partly-anticipated shock series suggests.

Q7. What do the results imply about the effect of tax shocks, and how confident are the authors in this part of the analysis?

In the linear (non-regime) model, a tax increase lowers output “from an initial impact of -0.2 … reaching -1.0 by the end of five years, which is similar to results reported in Blanchard and Perotti (2002),” but once split by regime the pattern becomes harder to interpret: in both recession and expansion the output response to a tax increase is “less negative” than in the linear model, and in the recession regime it turns “generally positive,” a result the authors say is “sensitive to using alternative measures of the elasticity of tax revenue with respect to output” and for which they “can offer no obvious explanation.” The authors are explicit that this tax-side anomaly is unresolved, which is itself informative about the limits of what the regime-switching SVAR cleanly identifies – the paper’s confident, well-supported claims are on the spending side, not the tax side.

Q8. What limitations do the authors themselves attach to the exercise?

The authors note that their regime split “lump[s] together” historically distinct types of downturns – “the recent one, associated with financial market disruptions and very low nominal government interest rates,” versus “other recessions induced by monetary contractions” – that may have different true multipliers, and that their multiplier estimates are “tied to historical experience concerning the persistence of policy shocks,” so may not generalize to future policies more or less persistent than those in the postwar sample. They also caution that their treatment of tax shocks remains “too simple” to capture the complex ways structural tax-policy changes affect the economy, and that the postwar U.S. experience may not forecast well into “a period of unprecedented long-run budget stress.”