Trend Breaks, Long Run Restrictions, and the Contractionary Effects of Technology Shocks

Recent empirical work using structural VARs with long-run restrictions assesses whether hours worked per capita rises or falls following a technology improvement. This literature reaches divergent conclusions on the sign of this effect, depending on whether hours worked enters the VAR in log-levels or growth rates. In contrast, I find that once one allows for (statistically and economically plausible) trend breaks in labor productivity, it is unimportant whether hours enters the VAR in levels or growth rates: Hours worked falls by a statistically significant amount on impact following a technology improvement. These findings apply in both bivariate and larger VAR systems. Results are also robust to reasonable variation in dating the trend breaks. In the full sample, however, one must take out two breaks to get the contractionary effects of technology improvements, at least if hours enters in levels. Finally, but importantly, I discuss conjectures on what, statistically and economically, drives the sensitivity to trend breaks. I argue that the sensitivity is economically interesting and informative