With revised data, when did the last recession begin?

In a recent post, Greg Mankiw cites Hamilton and Chauvet in support of his view that a good argument could be made that the recession of 2001 actually began in 2000. Mankiw writes:

“Back in 2004, Michael Mandel of Businessweek gave me grief for saying that the 2001 recession began in late 2000, rather than at the official NBER date of March 2001. My view (and the view of the nonpartisan CEA staff) was that the data were substantially revised after the NBER committee made their call, and the March 2001 date no longer seemed right in light of the revised data.”

As some readers may recall, the 2004 Economic Report of the President contained this box, which states:

“The National Bureau of Economic Research (NBER) uses a variety of
economic data to determine the dates of business-cycle peaks and
troughs. This task is made more difficult because many of these data
series are subject to revision. For example, on November 26, 2001, the
NBER announced that a recession had begun in March 2001. Since
then, the four data series that the NBER used to determine the timing
of the recession have been revised. The revisions to these series
suggest that the recent recession began earlier than March 2001.

The four series cited by the NBER in their decision about the recent
business-cycle peak were revised as follows:

• Real personal income less transfers:When the NBER dated the
  recession, this series showed a generally steady rise throughout
  2000 and early 2001. Subsequent revisions reveal that income
  peaked in October 2000.
  
• Nonfarm payroll employment: The data at the time of the recession announcement showed employment growing at a substantial pace in early 2001, with 287,000 jobs added from December 2000 to its peak in March 2001. Revised data show that employment grew less than one-third of this amount in early 2001 and peaked in February 2001.
  
• Industrial production:The original data used by the NBER showed
  that this series peaked in September 2000. Revised data show that
  this peak came even earlier, in June 2000.
  
• Manufacturing and trade sales: Original data showed a peak in
  August 2000; the most recent data show a peak in June 2000.

Thus, the revised data show that the latest peak among the four
series was February 2001, with some series peaking considerably
earlier. Moreover, another data series, which the NBER has recently
announced it will incorporate into its business-cycle dating process,
also shows a peak before March 2001: monthly GDP reached a high
point in February 2001, according to the most recently available
estimates computed by a private economic consulting firm.

While some arbitrariness in determining the date on which a recession began is inevitable, revisions since the NBER made its decision for the most recent recession strongly suggest that the business-cycle peak was before March 2001. The median date of the peak for the five series discussed here is October 2000. Other data support the notion that economic activity had slowed sharply or even begun to decline by this point, including the stock market, business investment, and initial unemployment claims. For these reasons, the analyses throughout this chapter (including the charts that compare this recession to past recessions) use the fourth quarter of 2000 as the peak of economic activity and the start of the recession.

In October 2003, the NBER announced that it would defer consideration of whether the latest business-cycle peak should be revised until the results of the coming comprehensive revision of the National Income and Product Accounts were released. The major results of this revision were announced in December 2003, but the monthly manufacturing and trade sales data and some of the detail needed to estimate monthly GDP had not been released at the time this Report went to press.”

First let me make the observation that while Mankiw prefers the multivariate Markov Switching results in Table 6 of Hamilton-Chauvet, it is by no means the only set of estimates in the paper. Table 8, using the recursive estimation methodology indicates a March 2001 recession date — exactly the determination of the NBER Business Cycle Committee. (Caveat: We’ll have to wait for Jim to tell us which one he prefers — these are solely my views I’m reporting.) Indeed, using a similar methodology, Piger and Chauvet find that the recursive methodology identies the beginning of the 2001 recession in 2001q2 (according to real GDP) and September 2001 (payroll employment). (Using the full sample methodology retrieves dates in 2000 similar to those noted by Mankiw.)

In any event, we now have series that have been revised several times (see this post on data revisions) so that they are in final form. What do these data say? To answer this question, I refer first to what the NBER Business Cycle Committee states are the variables of importance:

“In choosing the dates of business-cycle turning points, the committee follows standard procedures to assure continuity in the chronology. Because a recession influences the economy broadly and is not confined to one sector, the committee emphasizes economy-wide measures of economic activity. The committee views real GDP as the single best measure of aggregate economic activity. In determining whether a recession has occurred and in identifying the approximate dates of the peak and the trough, the committee therefore places considerable weight on the estimates of real GDP issued by the Bureau of Economic Analysis of the U.S. Department of Commerce. The traditional role of the committee is to maintain a monthly chronology, however, and the BEA’s real GDP estimates are only available quarterly. For this reason, the committee refers to a variety of monthly indicators to determine the months of peaks and troughs.

The committee places particular emphasis on two monthly measures of activity across the entire economy: (1) personal income less transfer payments, in real terms and (2) employment. In addition, the committee refers to two indicators with coverage primarily of manufacturing and goods: (3) industrial production and (4) the volume of sales of the manufacturing and wholesale-retail sectors adjusted for price changes. The committee also looks at monthly estimates of real GDP such as those prepared by Macroeconomic Advisers (see http://www.macroadvisers.com). Although these indicators are the most important measures considered by the NBER in developing its business cycle chronology, there is no fixed rule about which other measures contribute information to the process.” [emphasis added]

Below are figures for the first three key variables: real GDP, payroll employment (with civilian employment based on the household survey thrown in for good measure), and real personal income less transfers. Gray shading indicates the recession dates according to the NBER.

Figure 1: Real GDP in billions of chained 1996$, SAAR. Source: BEA via St. Louis Fed FRED.

Figure 2: Payroll employment (red line) and civilian employment (over 16) (blue line), in thousands, seasonally adjusted. Source: BLS via St. Louis Fed FRED.

Figure 3: Personal income less transfers, in billions of 2000Ch$ (using PCE deflator), seasonally adjusted. Source: BEA via St. Louis Fed FRED and author’s calculations.

The peak for real GDP could be 2000q4 or 2001q2. The “sawtooth” pattern makes a determination difficult; John Kitchen’s “quarterly variation” growth measure first goes negative in 2001q1, while his “underlying growth” measure hits its trough in 2001q4. Payroll employment peaks in February 2001 (civilian employment in March 2001). Personal income less transfers peaks in March 2001. What about the auxiliary variables of interest, industrial production and real sales? (Note: I don’t have access to Macroeconomic Advisers’ monthly GDP series — contributions welcome!)

Figure 4: Industrial production, seasonally adjusted. Source: St. Louis Fed FRED.

Figure 5: Real sales of manufacturing and trade, in billions of 2000Ch$, seasonally adjusted. Source: BEA NIPA Supplementary Table 2BU.

The peak dates for these two less heavily-weighted series are June and September 2000 respectively.

So, while I think reasonable people can disagree on the starting point, for me, a revisiting of the data tells me the recession of 2001 took place in…2001.

[Addition, 8/19/06 11:20am Pacific]

Courtesy of Spencer England, I can now plot real monthly GDP as estimated by Macroeconomic Advisers:

Figure 5: Monthly index of real GDP, in billions of 2000Ch$, SAAR. Source: Macroeconomic Advisers via Spencer England.

The graph confirms Rex’s comment below that there are two peaks, solidly in the NBER recession period, both in 20001. Whether this plateau is consistent with one’s view of what defines a recession, I leave up to the reader. (Recall, however, NBER does not place central importance on the monthly GDP index.)

Technorati Tags: recession,
GDP,
payroll employment,
industrialproduction,
sales, and
personal income.