A major topic of empirical macroeconomics is the analysis and measurement of shock response. Typically, aggregate data and a variety of tools, such as the Impulse Response Function (here-inafter, IRF) and other related scalar measures, are used to perform the analysis.
A recent strand of literature has emphasized that this approach can be problematic when the representative agent hypothesis is violated. According to this view, in the dynamic case when the coefficients differ across individuals, persistence estimates based on aggregate data are significatively higher than those derived from disaggregate data.
Two explanations have been provided to account for this phenomenon. Firstly, building on the results of Pesaran and Smith (1995), it has been argued that estimates computed with aggregate data are biased and that this bias translates into an overestimation of persistence measures (see Imbs et al., 2005). Secondly, other authors state that the aggregation of heterogeneous dynamic processes is not an innocuous operation and that it may tend to increase overall shock response (see Altissimo et al., 2006a, 2006b, among others).
Several empirical studies corroborate these arguments by finding estimates of persistence that vary considerably across aggregation levels and are, in general, higher the higher the level of aggregation (in addition to the above mentioned articles, see Altissimo et al., 2007, Clark, 2006, Lünnemann and Mathä, 2004, Abraham and White, 2006, etc). The conclusions of these articles are drawn by comparing the average across individuals of some scalar measures of persistence, typically the sum of the autoregressive coefficients (SAC), the largest autoregressive root (LAR) or the first autocorrelation, with the values of the same measures computed with aggregate data.
Dynamic heterogeneity has been found to be relevant in a wide variety of contexts, such as in the speed of reversion to PPP (Imbs et al., 2005 and Crucini and Shintani, 2006), in the speed of reversion of income shocks, (Hu and Ng, 2004), in the dynamics of saving behavior (Haque et al., 2000), in inflation dynamics (Altissimo et al., 2006b, Angeloni et al. (2005), in labor demand across firms (Zhang and Small, 2006), etc. One could argue that the existence of some degree of heterogeneity across individuals is likely to be the rule rather than the exception in most contexts.
If the criticisms above are confirmed, they could be very harmful because they imply that the standard approach of assessing the response of the economy to shocks using aggregate data would yield, on many ocassions, non-reliable estimates that would tend to systematically amplify the impact of shocks on the economy.
