Technology, Employment, and the Business Cycle: Do Technology Shocks Explain Aggregate Fluctuations?
📄 Summarized from the full manuscript · Human-reviewed for faithfulness before publication
In brief
Standard business-cycle models assume a surge in technology makes firms hire more workers along with producing more. This 1999 paper finds the opposite in US and other G7 data: once monetary-demand shocks are statistically separated out, a positive technology shock is followed by a persistent *fall* in employment, while demand shocks (not technology) are what drive employment and productivity up together. The pattern fits a model with sticky prices and variable work effort better than it fits the standard real-business-cycle story, reshaping how economists think about what actually drives recessions and booms.
What this paper finds — and why it matters
Gali decomposes measured labor productivity and employment into a technology component and a non-technology (“demand”) component, using a bivariate structural VAR identified by a single long-run restriction: only technology shocks can have a permanent effect on the level of labor productivity. Applied to postwar U.S. data (1948-1994), this decomposition shows that technology shocks are associated with a large negative conditional correlation between employment and productivity (average -0.78 across specifications), while demand shocks are associated with a positive conditional correlation (average 0.38); the unconditional correlation between employment and productivity is small and, on average, negative (-0.14). The identified technology shocks also produce a persistent decline in employment, and demand shocks produce a temporary rise in measured productivity – both contrary to the predictions of standard real-business-cycle (RBC) models in which a positive technology shock raises both variables together. Gali shows the same three qualitative results hold for most of the other G7 countries (all but Japan) and are robust to using hours versus employment, to a larger five-variable VAR, and to first-differenced versus HP-filtered data. He argues the pattern is instead consistent with a stylized sticky-price, monopolistic-competition model with variable effort, in which a positive technology shock that is not fully accommodated by monetary policy initially lowers labor demand because desired output rises by less than productivity, while a demand (monetary) expansion raises measured productivity through increased unobserved effort.
Summary of a classic paper, AI-assisted and human-reviewed. See the linked original for the authoritative claims and full conditions.
Questions & answers
Q1. What puzzle in RBC models motivates the paper, and what is the paper’s main objection to the standard defense of RBC theory?
The paper’s point of departure is that calibrated real-business-cycle models driven by technology shocks predict a strong positive correlation between labor productivity and hours/employment, whereas the unconditional correlation observed in the data is close to zero or slightly negative; researchers have defended RBC models by arguing that this reflects technology shocks (positive correlation) being offset by other, non-technology shocks that act as labor-supply shifters (negative correlation). Gali’s objection is that simply matching the unconditional correlation is not a strong test: if technology shocks are a quantitatively important source of fluctuations, RBC models should also correctly describe how employment and productivity respond specifically to a technology shock, i.e., their conditional comovement – and it is this conditional prediction the paper goes on to test directly.
Q2. How does the paper identify technology shocks separately from demand shocks?
The identification strategy assumes a bivariate process for employment growth and productivity growth, restricted so that only technology shocks are permitted to have a permanent (long-run) effect on the level of labor productivity, implemented as a zero restriction on the relevant long-run multiplier in a structural VAR’s moving-average representation (a Blanchard-Quah-style long-run restriction). This restriction is satisfied by a broad class of models – including RBC models and the sticky-price model Gali develops in Section 2 – so the identification itself does not presuppose which class of model is correct; what differs across model classes is the sign of the conditional comovement each type of shock produces.
Q3. What are the three central empirical results for the United States?
For the U.S. (1948:1-1994:4), Gali reports: (a) the conditional correlation between employment and productivity is negative for technology shocks and positive for demand shocks; (b) employment declines persistently after a positive technology shock; and (c) measured productivity rises temporarily after a positive demand shock. Quantitatively, the unconditional correlation between employment and productivity averages -0.14 (range -0.29 to 0.28 across labor-input measures and filters), while the average conditional correlation is -0.78 for the technology-driven components and 0.38 for the demand-driven components – the opposite sign pattern from what multiple-shock RBC models predict, in which technology shocks generate the positive comovement and non-technology shocks the negative one.
Q4. Are these findings robust to how labor input and detrending are measured?
Yes – the three main results are reported as robust to using either employment or total hours as the labor-input measure, to first-differenced versus Hodrick-Prescott-filtered data, and to extending the bivariate VAR to a partially-identified five-variable system that adds interest rates, money, and prices.
Q5. How does international evidence (Section 5) bear on the U.S. findings?
Using data for Canada, the U.K., Germany, France, Italy, and Japan, Gali finds the same three qualitative results hold for every country except Japan, and reports that the fluctuations in output and employment attributable to demand shocks are strongly positively correlated and clearly tied to the major postwar business-cycle episodes, while the fluctuations attributed to technology shocks display neither property.
Q6. What alternative model does the paper offer to explain these facts, and what is its intuition?
Gali develops a stylized monetary general-equilibrium model with monopolistic competition, sticky prices (firms set prices before shocks are realized), and variable, unobserved labor effort, and shows it can generate a negative conditional comovement between employment and productivity for technology shocks and a positive one for demand shocks, under empirically plausible parameter ranges (short-run increasing returns to labor, ρ ∈ (1,2), and a not-too-strong endogenous monetary response, γ ∈ [0,1)). The intuition: aggregate demand in the short run is pinned down largely by real money balances; with sluggish prices and limited monetary accommodation, a positive technology shock raises desired output by less than it raises multifactor productivity, so firms cut employment even as productivity rises. A monetary expansion, by contrast, raises output partly through increased unobserved effort, so measured productivity can rise together with (observed) employment.
Q7. What does the paper find about short-run increasing returns to labor (SRIRL), and why does it matter for interpreting productivity regressions?
The paper argues that failing to separate technology from demand-driven variation in standard production-function regressions biases estimates of returns to labor, and proposes using the structural VAR’s demand-driven component of output as an instrument; applying this approach to U.S. and international data, Gali finds evidence of significant short-run increasing returns to labor in most cases. This matters because the sign of SRIRL (ρ > 1) is also a key parameter governing whether the sticky-price model in Section 2 can generate a positive productivity response to demand shocks.
Q8. What is the paper’s overall conclusion about the relative merits of RBC versus sticky-price/imperfect-competition models?
Gali concludes that, for most G7 countries, the estimated effects of identified technology shocks are hard to reconcile with RBC models – positive technology shocks are followed by a decline in employment and a negative comovement with productivity – while the same evidence is consistent with a class of models featuring imperfect competition, sticky prices, and variable effort, implying a different account of the sources of aggregate fluctuations, and different implications for the potential role of stabilization policy, than the RBC paradigm.
Key terms in this paper
Definitions below follow the paper's own usage.
- Conditional correlation (technology- versus demand-driven)
- the correlation between the component of employment growth and the component of productivity growth attributable to one specific identified shock (technology or demand), as opposed to the unconditional correlation between the raw series; the paper's central object of interest because the two shock types are shown to imply opposite-signed conditional comovements even when their sum roughly matches a near-zero unconditional correlation.
- Long-run identifying restriction
- the assumption, used to separate technology from demand shocks in the structural VAR, that only technology shocks can have a permanent effect on the level of labor productivity, implemented as a zero constraint on the (1,2) element of the long-run multiplier matrix; satisfied by both RBC models and the paper's own sticky-price model, so the restriction alone does not favor either class.
- Short-run increasing returns to labor (SRIRL)
- the parameter ρ relating output to effective labor input (ρ > 1 meaning output rises more than proportionally with effective labor in the short run, e.g. through labor hoarding or variable capital utilization); in the paper's sticky-price model, ρ ∈ (1,2) combined with a muted endogenous monetary response is sufficient for a demand expansion to raise measured productivity even though employment also rises.
- Variable effort
- unobserved variation in work intensity per worker (or per hour) that drives a wedge between employment/hours (observed) and effective labor input; in the paper's model this wedge is what allows measured labor productivity to rise following a demand expansion even without a true technology improvement.