I analyze the business cycle implications of noisy economic indicators in the context of a dynamic general equilibrium model. Two main results emerge. First, measurement error in preliminary data releases can have a quantitatively important effect on economic fluctuations. For instance, under efficient signal-extraction, the introduction of accurate economic indicators would make aggregate output 10–30 percent more volatile than suggested by the post-war experience of the U.S. economy. Second, the sign — but not the magnitude — of the measurement error effect depends crucially on the signal processing capabilities of agents. In particular, if agents take the noisy data at face value, significant improvement in the quality of key economic indicators would lead to considerably less cyclical volatility.
From the GDP to M2, to productivity growth to the index of leading economic indicators, preliminary releases of economic data are routinely subject to sizable revisions as more information becomes available in subsequent periods. The existence of noise in these and other economic indicators has been the subject of several studies, but the associated literature is primarily statistical and of a partial equilibrium nature.1 In this paper I take a novel and complementary approach, examining the effects of indicator noise in a fully articulated dynamic macroeconomic model.
In choosing a theoretical framework for the analysis, I opted for the class of models in the real business cycle (RBC) tradition. Such a choice is motivated by two main factors. First, this modeling approach attempts to describe economic behavior from first principles, i.e., from individual optimization in a dynamic environment under uncertainty. Second, even if one disagrees with the main assumptions in the vast RBC literature, the RBC framework is well understood by the profession, and the results that I obtain under noisy indicators can be directly and quantitatively compared to the extant literature, which has generally abstracted from considering the economic effects of noisy information. Moreover, for the RBC skeptic, I review recent results in the monetary policy literature that suggest that the main thrust of this paper's findings should carry to a wide range of macroeconomic models and are thus not specific to the RBC structure.
This paper can be thought of as a well-defined sequence of computational experiments (Kydland and Prescott, 1996). Motivated by the findings of the empirical literature on indicator noise — which are summarized in Section 2 — I pose a very simple question: If economic data are as noisy as suggested by the statistical literature, what are the likely consequences for individual and macroeconomic behavior?2 To address this question, I introduce a version of the model first proposed by Baxter and King (1991), which I augment to include a noisy productivity indicator. In Section 3, I describe the model and lay out the solution to the representative agent's dynamic optimization problem. Such a solution, which assumes that agents use fully efficient signal extraction techniques, characterizes the business cycles of the artificial economy. Using the conventional tools of the quantitative approach to macroeconomics (King, 1995), I calibrate the model to the U.S. postwar data and run, in Section 4, a series of experiments designed to quantify the aggregate effects of noisy information. I find that the presence of measurement error in preliminary data can have a non-trivial effect on economic fluctuations. In particular, the introduction of more accurate economic indicators would make aggregate output 10–30 percent more volatile than suggested by the postwar experience of the U.S. economy. The results are supported by a battery of sensitivity tests on key model parameters.
The paper also highlights the key role played by agents’ information processing capabilities in business cycle fluctuations. As discussed in Section 4, if agents are boundedly rational and take the preliminary data at face value, although the aggregate effect of indicator noise remains sizable, its sign is reversed: Better economic indicators would lead to considerably less cyclical volatility. Concluding remarks are included in Section 5.
Traditional decompositions of sources of macroeconomic fluctuations tend to emphasize the importance of supply versus demand shocks, permanent versus transitory, monetary versus real, etc. The computational experiments run in this paper uncovered an additional factor underlying these fluctuations: the very nature of the economic indicators on which agents based their decisions. Under efficient signal-processing, the presence of noise in key economic data has a dampening effect on business cycle volatility. Accordingly, an appreciable improvement in the accuracy of economic indicators would likely contribute to significantly larger volatility in the economy.
Following Bomfim (1999), this paper also highlighted the importance of agents’ information processing capabilities in understanding the aggregate effects of noisy economic indicators. In particular, given the strikingly different conclusions that can be drawn from the fully rational and boundedly rational characterizations of the model, the results point to a novel way to empirically distinguish between these two views of the world.
The quantitative findings of this paper are especially relevant in light of the declining relative importance in the U.S. economy of sectors such as agriculture, mining, and manufacturing — for which we currently have more reliable data — and the growing importance of the harder to measure service sector. What this trend suggests is that the problem of noisy economic data is unlikely to disappear soon, making the need to understand its implications for business cycle fluctuations that much more important.
Finally, I should caution those who might feel tempted to interpret the rational-expectations based results to mean that better economic indicators are bad because they lead to higher macroeconomic volatility. In the model presented in this paper, all fluctuations are optimal responses to shifting opportunities in the leisure-consumption tradeoff. Therefore, there is an important sense in which noisy data are always bad because they make it harder for agents to fully identify and efficiently respond to these shifts.
