Working Paper: NBER ID: w9898
Authors: David Berger; Ricardo J. Caballero; Eduardo Engel
Abstract: The estimated persistence of macro aggregates involving lumpy microeconomic adjustment is biased downward when inferred from VAR estimates. The extent of this “missing persistence bias” decreases with the level of aggregation, yet convergence is very slow. Paradoxically, while idiosyncratic shocks smooth away microeconomic non-convexities and are often used to justify approximating aggregate dynamics with linear models, their presence exacerbates the bias. We propose a method to estimate the true speed of adjustment and illustrate its effectiveness via simulations and applications to real data. \nThe missing persistence bias is relevant for macroeconomists on many grounds. First, when calibrating or estimating models via simulation based methods, macroeconomists should pay attention to the number of agents used in simulations for otherwise they are likely to obtain systematic biases in their parameter estimates. Second, results purporting to find persistence measures that vary systematically with levels of aggregation should be examined with care since the differential speeds may disappear when using estimation methods robust to the missing persistence bias. To illustrate the latter, we show that the difference in the speed with which inflation responds to sectoral and aggregate shocks (Boivin et al 2009; Mackoviak et al 2009) disappears once we correct for the missing persistence bias.
Keywords: macroeconomics; lumpy adjustment; VAR estimates; persistence bias
JEL Codes: C22; C43
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| missing persistence bias (G41) | estimated persistence of macro aggregates (E10) |
| lumpy microeconomic adjustments (D59) | estimated persistence of macro aggregates (E10) |
| conventional VAR estimates (C51) | underestimating the true speed of adjustment (C51) |
| missing persistence bias (G41) | impulse response functions (C22) |
| impulse response functions (C22) | erroneous conclusions about the dynamics of inflation (E31) |
| sectoral data (R10) | missing persistence bias (G41) |
| idiosyncratic shocks (D89) | missing persistence bias (G41) |