Is the technology-driven real business cycle hypothesis dead? Shocks and aggregate fluctuations revisited

The paper opens by noting that “a trio of recent papers has called into question the notion that technology shocks have anything to do with business cycles”: Galí (1999, long-run-restriction SVAR), Shea (1998, patents and R&D data), and Basu, Kimball, and Fernald (1999, Hall-style regressions with utilization proxies) all independently found that “positive technology shocks appear to lead to declines in labor input.” The authors state plainly that “when we initially set out on this project, we were somewhat skeptical that the shocks identified using long-run restrictions could plausibly be interpreted as technology shocks,” motivating the paper’s extensive plausibility testing before taking the result at face value.

Using a standard neoclassical growth model (an extension of the working-paper version of King, Plosser, Stock, and Watson 1987) as a source of auxiliary qualitative predictions, the paper derives two additional long-run restrictions beyond Galí’s original identification and tests them formally using Hansen’s J-statistic. The paper reports “Hansen’s J-statistic is 0.117, with a p-value of 0.73,” concluding “there is no evidence against the overidentifying restriction with respect to hours” and that “all three identification schemes are mutually consistent” – i.e., the extra restrictions implied by the growth model are not rejected by the data, lending credibility to interpreting the identified shock as a genuine technology disturbance.

Following the logic of Hall (1988) and Evans (1992) – who had shown that money, interest rates, and government spending Granger-cause the conventional Solow residual, casting doubt on it as a measure of technology – the paper regresses the identified technology shock on three independently-dated sets of “exogenous” shock dummies: Romer and Romer’s (1989) monetary policy indicators, Hoover and Perez’s (1994) oil shock dummies, and Ramey and Shapiro’s (1998) war dates. The paper reports that “none of the sets of dummy variables is significant in explaining the technology shock,” with “the lowest p-value… for Ramey-Shapiro war dates” at only 0.12, concluding “there is no evidence that the technology shock identified using long-run restrictions is correlated with any of these other three shocks” – so the shock “passes the Evans’ test” and appears to be “a better candidate than the Solow residual for a true technology shock.”

Calibrating habit persistence and investment adjustment costs to the high values used by Boldrin, Christiano, and Fisher (2001) to match asset-pricing facts (b = 0.9, θ = 4.348), the paper finds that “it is the combined sluggishness in investment (due to investment adjustment costs) and consumption (due to habit persistence) that drives the decline in labor. Without both effects, labor either rises or declines very little.” The intuition is that a permanent technology improvement creates an immediate wealth effect, but habit persistence makes consumers reluctant to raise consumption quickly, and high adjustment costs make investment an expensive way to absorb the extra resources in the short run – so “households ‘spend’ the new wealth on the only remaining alternative: leisure,” reducing labor supply even though real wages are rising. The authors note “the high values of b and θ are not necessary”: a milder calibration (b = 0.5, θ = 2) still generates a roughly two-year decline in hours, with hours falling 2.5% on impact for a 5% technology shock.

The paper also builds a one-sector variable-utilization model with short-run fixed-proportions (Leontief) production, adapted from Ramey and Shapiro’s (1998) model of military buildups, exploring it specifically “because of its improved predictions for wages” relative to the simplest versions of a negative technology-hours link. In this model, as in the dynamic sticky-price model of King and Wolman, hours fall because “current wages are low relative to expected future wages,” inducing an intertemporal-substitution-driven reduction in current labor supply – a different mechanism from the habit-formation/adjustment-cost model, where hours fall because leisure is the only margin that can readily absorb the technology shock’s short-run wealth effect.

Yes: the paper notes “the habit formation-adjustment cost model, real wages jump so far on impact that they actually overshoot their long-run level slightly,” whereas “in the sticky price model… real wages do not adjust at all on impact, and then rise steeply… There is a similar pattern in the Leontief model.” However, the paper is explicit that its own empirical estimates cannot adjudicate between these competing wage predictions: some specifications (“the first-difference specifications… as well as the VECM labor share specification”) show real wages jumping immediately, “more consistent with the habit formation-adjustment cost model,” while others (the VECM with an estimated cointegrating vector) show “a slower response of real wages, favoring the sticky price model and Leontief model” – so “the data do not give robust results” on which mechanism is correct.

The paper is explicit that reproducing the qualitative labor-input response is not the same as rehabilitating technology shocks as a central driver of business cycles: “despite their ability to explain these effects, the modified models do not resuscitate technology shocks as the driving force of business cycles. In that sense, the original technology-driven real business cycle hypothesis does appear to be dead.” The deeper problem the paper documents is that the identified technology shock produces “very un-business cycle like characteristics” on its own – “hours move in the opposite direction of output, productivity, consumption and investment” – so even a model that matches this pattern is, by construction, not a model in which technology shocks alone generate the standard positive comovement among output, hours, consumption, and investment that defines an actual business cycle.

The paper reports that the second, non-technology shock in its system – the one that “produces positive comovements between output, productivity and hours,” matching the textbook business-cycle pattern – “also raises consumption and investment, but not real wages.” Testing this shock against the same oil-event and military-event dummy variables used for the exogeneity check on the technology shock, the paper finds that “almost 20 percent of the variation in this shock can be explained by simple dummy variables capturing oil events and military events” – in contrast to the technology shock, which showed no significant relationship with any of the three sets of exogenous dummies tested (Q3).

The paper highlights Campbell and Ludvigson’s (2001) home-production model, in which a low intertemporal elasticity of substitution in consumption – needed there to be consistent with balanced growth under home production – combined with a highly persistent technology shock, “produces a negative correlation between technology and hours worked.” The authors note that Campbell and Ludvigson themselves treated this prediction as “a fundamental difficulty of the model,” but comment dryly that “in light of the empirical results it should be considered a strength of the model” – underscoring the paper’s broader point that multiple distinct modeling devices, not only sticky prices, can match the empirical fact once it is taken seriously.

The paper’s conclusion states the authors “were surprised to find that further study of the data supported” the interpretation of Galí-style identified shocks as genuine technology shocks, despite having begun “somewhat skeptical” – but that this very success in validating the shock’s identity makes its empirical behavior more puzzling, not less, since “what appears to have all of the right properties for a technology shock leads to very un-business cycle like characteristics.” The paper’s own summary verdict is twofold: technology shocks, properly identified, are real and distinguishable from other known disturbances, but they are not the force generating the standard comovements of the business cycle – that role falls to a separate shock, partially explained by oil and military events, leaving the search for its full structural interpretation as an open question.