There’s a lot of talk about recession these days, despite the fairly rapid average growth of GDP in the past few quarters. Krugman (via DeLong) observes a slowdown is coming that might feel a lot like a recession. DeLong considers whether Fed policy has already raised rates to such a degree a recession is inevitable. Roubini bravely cites probabilities. My colleague James Hamilton provides a contrasting opinion, based upon his academic work with Chauvet [pdf].
What I want to focus on is DeLong’s assertion that “[f]orecasting recessions is a fool’s game. If there is enough solid economic information to make it appear highly likely that a recession is coming — that production, unemployment and consumer demand will actually fall — then it is highly likely that there already is a recession.” One is tempted to ask whether we would know if we were in a recession now, when so many macro indicators are blinking green. (To anticipate your questions, I don’t think we’re in a recession — this is an exercise in learning about the macro data.)
It’s useful to recall that, not only are almost all macroeconomic series announced with a lag, NIPA announcements are also are revised twice after the first announcement (“advance”, “preliminary”, “final”). There is also a comprehensive annual revisions that go back several years that occur in July. the NIPA announcements provide information on the size of the revisions (mean change, standard deviation). The latest NIPA release of July 28th reports that the standard deviation of revisions from advance to final is 0.4 percentage points on an annualized basis; and 1.0 from advance to latest.
Compounding the difficulties associated with tracking the cycles in the economy, these revisions appear to be larger around turning points. Recalling the the period before the last recession (dated by NBER Business Cycle Dating Committee as January March 2001 to November 2001), I thought it would be useful to compare the data of the time against what we now think are the measures of macroeconomic performance.
Figures 1-3 denote the annualized growth rate (in log terms) of real GDP, real consumption, and real business fixed investment; the blue (red) line is the May 2001 (May 2006) vintage of data as provided by the Philadelphia Fed’s realtime database.
Figure 1: Annualized quarter-on-quarter growth rate of real GDP (in log terms). Source: Philadelphia Fed’s realtime database, St. Louis Fed, and author’s calculations.
Figure 2: Annualized quarter-on-quarter growth rate of real consumption (in log terms). Source: Philadelphia Fed’s realtime database, St. Louis Fed, and author’s calculations.
Figure 3: Annualized quarter-on-quarter growth rate of real business fixed investment (in log terms). Source: Philadelphia Fed’s realtime database, St. Louis Fed, and author’s calculations.
As can be seen from these figures, growth rates were revised downward as more data were incorporated into estimates of macroeconomic aggregates. Figure 4 shows the nominal corporate profits as recorded in these two vintages of NIPA data. I am tempted to say that the 2000-01 experience with overstating corporate profits is a one-time affair, but I think I will resist.
Figure 4: Corporate profits in billions of $ (SAAR). Source: Philadelphia Fed’s realtime database and St. Louis Fed.
Quarterly data are pretty coarse, and the NIPA data come out with a considerable lag. What about monthly data. The NBER Business Cycle Dating Committee relies on several indicators, including monthly industrial production and payroll employment series. In Figures 5-6, the May 2001 vintages of these series (from St. Louis Fed’s ALFRED), as well as the November 2001 payroll series(drawn from Robert Hall’s database) are portrayed alongside the August 2006 series (drawn from the St. Louis Fed’s FRED II).
The Industrial Production series has been rebased, so the two vintages of data cannot really be compared (I used the log ratio in 1971m12 to rescale the two series so that they match exactly in that month and before). What is of interest is that in the November 2001 vintage series, IP hits a peak in September 2000; in the August 2006 vintage series, it hits a peak in June 2000. In contrast, nonfarm payroll employment peaks in February, or March, or February (2001), as one uses the May, November 2001 or August 2006 vinages. The key difference in payroll employment is that the most recent data exhibit a much more pronounced employment decline than the November 2001 vintage used to determine the onset of the last recession.
How do these patterns inform the debate about current events?
Turning first to real GDP, one sees that the most recent quarter’s growth rate of 2.5% seems a large distance above 0; in addition, DeutscheBank forecasts positive growth in the subsequent two quarters.
Figure 7: Annualized quarter-on-quarter real GDP growth (in log terms), and DeutscheBank forecasts. Source: St. Louis Fed and DeutscheBank.
However, the 2001q1 quarter-on-quarter annualized growth rate was revised downward by over 2 percentage points, and this suggests that similarly large revisions could change our perspective on output in the second quarter. On the other hand, industrial production and payroll employment (available up to June and July 2006 respectively) suggest continued positive momentum. John Kitchen’s real-time estimate of 3rd quarter annualized qoq GDP growth (as of 8/1) is about 3.5 percentage points.
Figure 8: Log industrial production and nonfarm payroll employment. Source: St. Louis Fed FRED II.
What lesson can one take from this exposition? I think it is that Brad DeLong is essentially right on this point: we won’t know whether we are in a recession until we are well into it. (Personally, I don’t think we’re in a recession now, but examining the data surely imparts a lot of humility.)
[late addition 8/10 7:10 pm Pacific]:
My former colleague during my time at the CEA, Phill Swagel, reminds me of this item from the 2004 Economic Report of the President, which argued that a case could be made that on the basis of the data available at end-2003, the recession of 2001 could have actually begun in 2000. (I have collected links to more documents on this debate here.) By the way, Kitchen’s real time estimates indicate the underlying growth rate (RTF-U) hit close to bottom in 2001q1; the quarterly variation measure (RTF-Q) goes negative the first time in that same quarter.