is that even after dropping the serial correlation as much as they do in the most recent ver-
sion of their comment, the explanatory power of the nonlinear terms in their experiments
is less than half of what we found in the data. If one adopts more realistic assumptions on
the persistence of the driving processes, and uses the values assumed in CW (2002), there is
essentially no gain from adding higher moments to their regressions. Again, and contrary to
their claim, there is no false positive finding even when applying our methodology to highly
unrealistic data.
The other two paragraphs in their conclusion carry the implicit messages that “the gap ap-
proach” is voodoo-economics and that they are ready to deliver a superior gap-free alternative.
First, what they call “the gap approach” has been derived formally by us and many others before
us from at least as sophisticated microeconomic models as the one they present (for this, see the
extensive literature on the optimality of (S, s) models).13 Second, and perhaps more importantly,
the methods derived from dynamic optimization that do not “rely on gap measures” already exist
in published work. In fact, the difficulties in measuring gaps was our own motivation for Caballero
and Engel (1994, 1999).14
To end on a more positive note, CW’s approach contrasts with more constructive and interesting
recent developments in the literature on the macroeconomic implications of lumpy microeconomic
adjustments. For example, Kahn and Thomas (2003) conclude that within an otherwise standard
RBC model, fixed costs of adjusting capital do not have a significant impact on aggregate invest-
ment.15 This finding has been misinterpreted by many as a demonstration that fixed costs do not
matter for actual investment. But this is not what they did. In fact, they also show that the aggre-
gate data generated by such a model misses important features of actual aggregate data, such as the
skewness caused by investment spikes. And that such spikes can be generated by microeconomic
fixed costs if the interest rate is not endogenized (confirming the results in Caballero and Engel,
1999). This finding points to an interesting and fruitful area of research: How does the RBC model
need to be modified for it to capture the nonlinearites that are observed in aggregate investment?
Let us hope that energy will be spent on this type of question.
13The first proof of optimality of (S,s) policies is in Scarf (1960). For important extensions, relevant to the models
discussed in this reply see, among others, Harrison, Sellke and Taylor (1983), Grossman and Laroque (1990), and the
pedagogical survey in Dixit (1993).
14Inthese papers we extended the (S, s) literature to incorporate stochastic adjustment costs and estimate a structural
model via maximum likelihood.
15See Veracierto (2002) for a similar conclusion in a model of irreversible investment.