Productivity Shocks and the Business Cycle: Reconciling Recent VAR Evidence

Gali (1999) used a VAR with productivity and hours worked to argue that technology shocks are negatively correlated with labor and are unimportant for the business cycle. More recently, Beaudry and Portier (2003) studied a VAR in productivity and stock prices. Remarkably, they found that the component which has a permanent impact on productivity is almost identical to that which has no immediate impact on productivity. Moreover, either of these components explains most business cycle variation. Like Gali's results, these observations are inconsistent with early RBC models, but on the other hand they contradict Gali's claim that technology shocks are unimportant for cycles. In this paper, we study trivariate VARs in productivity, hours worked, and stock prices to see how these apparently contradictory results can be reconciled. We find one VAR specification that qualitatively and quantitatively matches the findings of Gali (so that long-run technology shocks drive hours down), and a second specification that matches the main findings of Beaudry and Portier (so that long-run technology shocks increase hours, are similar to the short-run shock to stock prices, and play a major role in generating business cycles). Surprisingly, the difference between these two specifications has nothing to do with estimating in levels or in differences, or with running VARs or VECMs, or with the ordering of variables. The only difference between the two specifications lies in which productivity variable is used: labor productivity (to generate results like Gali's) or TFP (to generate results like those of Beaudry and Portier). Both the original Beaudry and Portier estimations, as well as our findings on the productivity specification, add to the evidence that Gali's findings are not robust. Apparently the cyclical role of technology shocks is only picked up when a sufficiently cyclical productivity series is used in the estimation.