Unfortunately for policymakers, investors, and consumers — all of whom might have been able to use such information to make better decisions regarding consumption, investment, and saving — the recession was not officially called until December 2008. Similarly, the four prior recessions were anywhere from five to nine months old before their onset was declared.1
Since recessions are critical events affecting the national economy, being able to “nowcast” a recession — that is, determine whether we are currently in one — would clearly be valuable. However, nowcasting recessions is not so easy. While the financial press routinely defines a recession as two consecutive quarters of a decline in the gross domestic product, the group that economists and policymakers rely on to call U.S. recessions, the Business Cycle Dating Committee of the National Bureau of Economic Research (NBER), follows a broader definition: “a significant decline in economic activity that spreads across the economy, which can last from a few months to more than a year.”2 It also examines a broader range of evidence. The NBER holds that a recession should also be visible in real income, employment, industrial production, and wholesale and retail sales. Yet, when initially released, these data are often noisy, ambiguous, incomplete, and subject to considerable subsequent revision. Furthermore, there is often a significant delay between the period being measured and the data’s release.3
One possible solution to the recession nowcasting conundrum is to use the Federal Reserve Bank of Philadelphia’s quarterly Survey of Professional Forecasters (SPF). The survey is already known for its ability to forecast an imminent recession via its “anxious index,” which is the mean of forecasters’ estimated probability that in the next quarter real GDP will contract — undoubtedly the most characteristic symptom of a recession.4 This report will evaluate a similar indicator derived from the survey: the mean probability of a contraction in real GDP in the current quarter, which we will call the anxious index nowcast. For some time now, it has been recognized that averaging numerous individual forecasts offers the best possible forecast from the available data.5 Using NBER recession quarters, this report will formally evaluate the anxious index nowcast’s predictive accuracy against that of the Conference Board’s composite index of coincident indicators, a popular business cycle indicator. By performing this comparison, we will be able to examine whether a completely unrevised, real-time probability assessment based on one variable can outperform a revised series based on multiple economic variables. The outcome should determine whether the public has the information it needs to nowcast recession quarters accurately in real time.
For instance, the January 1980 recession onset was announced in June 1980, and the July 1990 recession onset was announced in April 1991. See www.nber.org/cycles/main.html.
This is especially the case for GDP, which is first estimated for a given quarter roughly one month after the quarter ends. It is subsequently revised on a monthly basis, which results in large cumulative adjustments.
The one-quarter-ahead mean was first dubbed the anxious index in a 2002 New York Times article by David Leonhardt.
See Zarnowitz (1984).