One of the most important questions in macroeconomics is what leads to recessions? In the standard Real-Business Cycle (RBC) literature recessions are caused by large negative technology shocks. With the exception of the oil price shocks, however, it is difficult to identify negative technology shocks that are large enough to cause recessions. As King and Rebelo (1999) ask in their seminal paper, “if these shocks are large and important, why can’t we read about them in the Wall Street Journal?”An alternative explanation is put forward in Keynesian models of the business cycle, which suggest that recessions are mainly driven by monetary and fiscal policy shocks. Again, identifying these empirically has proven to be challenging. Hence, this major question remains fundamentally unanswered: Where are the shocks that drive the business cycle and, most importantly, where are the large negative shocks to technology that cause recessions?
This project investigates an additional mechanism: variations in the level of uncertainty. The general idea that links uncertainty to the business cycles is not new. John Maynard Keynes him-self argued that changes in investor sentiment, the so-called animal spirits, could lead to economic downturns. While this can be interpreted as an argument for the role of uncertainty, it has not traditionally played a large role in the theory of business cycles for two reasons. First, evidence on time series variation in uncertainty is scarce. The behavior of levels of variables over the business cycle (the first moment) is well documented, but the dispersion of these (the second moment) is much less well understood. Second, models with time varying uncertainty are theoretically challenging. In macroeconomics, the standard analytical and numerical solution techniques used in the RBC literature do not apply in this setup. One exception is Bernanke (1983), who models a single firm deciding on investment in energy efficient capital in the presence of oil-price uncertainty. He finds that higher uncertainty reduces investment as firms become more cautious. However, this paper is based on a stylized single-firm economy in partial equilibrium.
Recent work, however, has made progress on both issues. On the modeling side Bloom (2007) solves a model with heterogeneous firms and stochastic volatility and shows, in partial equilibrium, that high frequency fluctuations in uncertainty can cause large temporary drops and rebounds in aggregate output and employment. This occurs because higher uncertainty causes a large fraction of firms to temporarily pause their investment and hiring. The simulated impact of an uncertainty shock in this model compares favorably to vector autoregressive estimations on US data, showing a good match in both magnitude and timing. In this model, the pause in activity freezes the process of reallocation across units. Foster, Haltiwanger and Krizan (2000, 2006) show that reallocation can account for a significant fraction of aggregate total actor productivity (TFP) growth. This suggests that the effects of time variation in uncertainty could be large.
On measurement issues Schwert (1989) and Campbell et al. (2001) generate stock market measures of uncertainty show these are counter-cyclical. In this paper we gather evidence from aggregate, industry and firm-level datasets as well as forecasts and confirm this result for a much wider range of indicators. Volatility and uncertainty appear to rise strongly in recessions. One graphical demonstration of the greater volatility of output in recessions is Figure 1, which plots the quarterly growth rates of industrial production for 73 manufacturing industries for the period from 1962 to 2007 using data from the Federal Reserve Board Industrial Production Indices. The data is split into non-recession months (in black) and recession months (in grey). During a recession growth rates of industrial prduction are significantly lower on average (the first moment effect of recessions), and also significantly more dispersed (the second moment effect of recessions). In fact, the greater spread of industrial production during recessions is more striking at first sight than the difference in the means. This type of evidence suggests that this issue deserves further empirical investigation.
This project intends to build on the work cited above to answer the following research question: How large is the role of uncertainty in driving business cycles? To do this, the project is carrying out three interlinked pieces of analysis:
- (A) Constructing accurate time-series measures of uncertainty to quantify the size and significance of rises in uncertainty during recessions. This will use a variety of data sets, including the Census data, to build time series measures of cross-sectional volatility on a quarterly and annual basis linking them to GDP growth, industrial production and NBER recession indicators.
(B) Building a general equilibrium model that allows for shocks to both the level of technology (the first moment) as well as uncertainty (the second moment). The model can then be used to evaluate the impact of empirically calibrated changes in uncertainty. This could answer the question if these shocks are large enough to generate recessions and recoveries.
(C) Evaluating the empirical performance of this new type of business cycle model using Census micro data, with a particular focus on reallocation across establishments. One of the key ideas behind the impact of uncertainty is that it shuts down the process of reallocation. Firms become more cautious with productive firms expanding by less and unproductive firms contracting by less. Particularly at high levels of uncertainty the hiring and investment behavior of establishments should be less sensitive to productivity and demand conditions.
This project should provide both academic and policy benefits. On the academic front, providing a mechanism that is potentially able to explain recessions in a neo-classical framework addresses a major research question. On a policymaking front, understanding the factors that drive business cycles is critical in terms coming up with an optimal response.
Section 2 provides empirical evidence that uncertainty rises during recessions. It also provides an outline of the potential use for Census data to improve upon the existing uncertainty measures. Section 3 briefly introduces the theoretical model that we use to evaluate the mechanism we want to stress. Section 4 describes how the Census data can used to test the model’s predictions. Section 5 describes in detail the Census data sets that this project would use and comments on the risk for disclosure. Section 6 contains information on the Title 13 benefits. We discuss the proposed duration of the project as well as our sources for funding in Section 7. Section 8 concludes. Appendix A contains more detailed information on the data used in Section 2 and Appendix B provides a more rigorous definition of the model introduced in Section 3.
