Sunday, July 27, 2008

Revisions to Macroeconomic Data: Do the initial releases tell the story?

From my email inbox or your comments: You ask and I respond. Aunt Jane emailed me the following question:

“Just thought of something.... The government (and other groups) come out with monthly/quarterly reports on different aspects of the economy like the jobs report, inflation, etc. Then, the next month they have a retake on the report and finally have a third take at it. We seldom get those last two revisits which can change the numbers drastically and the final call is benign or positive. Can you do a report on the reports and how much can we trust the initial ones? I used to be able to get it from Tom Sullivan when he was in Sacramento but, now that he is in NYC and has his own Fox Business show, I can't find them. Thanks for considering this.”

Is this (the picture) what you would expect from a U.S. labor market that has slashed 438,000 jobs since January? Many macroeconomics statistics are subject to large revisions - upward and downward.

Many of the most closely followed statistics, like employment or gross domestic product (GDP), are revised two to six times since their initial releases. Therefore, any individual following the statistics should be weary to put too much stock in any one initial release.

How much can we trust the initial releases? If the data is subject to revision (not all macro data are), then the initial release is a good indicator of the general trend. However, the data will change during the second or third revisions, so one cannot trust the initial release at all. One should not trust the initial release except to provide some insight in the general direction of the economy.

Data that are followed closely by the financial markets and the media:

GDP: an estimate of total domestic income/production that is released by the Bureau of Economic Analysis (BEA). It is subject to 3 consecutive revisions: Advanced, Preliminary, and Final, 3 annual revisions performed in July, and finally, 5-year revisions. Over the revision periods, the data change substantially.

Employment data: nonfarm payroll data is released monthly by the Bureau of Labor Statistics (BLS). It is subject to 3 consecutive monthly revisions, annual revisions, and some irregular revisions.. Over the revision periods, the data change substantially.

Consumer Price Index (CPI): is the data used to estimate inflation that is reported by the BLS. It is not subject to revision.

Industrial Production (IP): Industrial production and capacity utilization estimate production in the industrial sectors, mining, manufacturing, and utilities that are reported monthly by the Federal Reserve Board (Fed) and subject to 6 monthly revisions and annual revisions.

Personal Income (PI): Personal income and consumption expenditures are released monthly by the BEA and subject to 6 months of revisions, and the same revisions as in GDP (used to measure GDP).

Other data followed by the markets, trade (exports and imports), new home sales, surveys (e.g., University of Michigan consumer confidence survey), or even housing data (e.g., existing home sales), reported by the National Association of Realtors, are all subject to revisions. Non-revised data are few, and the revised data often are underreported in the media.

Employment data is subject to heavy revision

Nonfarm payroll employment continued to trend down in June (-62,000), while
the unemployment rate held at 5.5 percent, the Bureau of Labor Statistics of
the U.S. Department of Labor reported today”

That statement must be taken with a grain of salt, as the data will be revised two more times (1st release, 2nd release, and 3rd release). Since 2000, the absolute value (meaning everything is a positive number, or +62,000 for June) of the 3rd (final) nonfarm payroll release was on average 135,000 jobs. The average revision from the 1st release to that 3rd release was 44,000 jobs, ranging from 1 to 152,000.

One should expect that the initial release is trustworthy anywhere between +/- 152,000 jobs.

GDP data is subject to heavy revision

GDP data is one of the most important series reported: it tells us the value of overall production, and is used to measure economic growth. There are three successive releases of GDP growth: Advanced, preliminary, and final. The advanced estimate is usually released one month after the current quarter’s end.

Real gross domestic product -- the output of goods and services produced by labor and property located in the United States -- increased at an annual rate of 1.0 percent in the first quarter of 2008 (that is, from the fourth quarter to the first quarter), according to final estimates released by the Bureau of Economic Analysis.“

Again, this statement must be taken with a grain of salt. Dennis J. Fixler and Bruce T. Grimm in 2002 find the following relationships between successive GDP revisions:

On average, the revision from the advanced release (1st) to the preliminary release (2nd) is 0.55%. That means, if the BEA estimate of Q2 GDP growth is reported at 2.5% (released this week), then the preliminary estimate (the second estimate) may be 3.05% or 1.95%. Note: the quarterly growth estimates are annualized; the quarterly GDP growth is converted to an annual growth rate.

On average, the revision from the preliminary release to the final release is 0.28%. The data are revised further annually, and then every 5 years with up to 1.1% change in the initial release.

One should expect that the initial release is trustworthy in the long term, but only in the range of +/- 1%. Compared to the average annual growth rate between 1983 and 2000, 3.6%, 1% is a significant revision.

Why is the data revised?

1. In the short run, the BLS and the BEA do not have all of the data when the reports are released. The BLS continues to gather payroll surveys after the end of the survey period (the 12th of the month), and the BEA gathers data that is released later than the advanced estimate. For example, trade data (used to estimate GDP) is released two months later than many statistics, and so the BEA has to wait until August to get the June trade data (still part of Q2), which is after the Advanced estimate of Q2 GDP has been released.

2. Seasonal factors change. Most data is adjusted for seasonal variations. For example, stores always hire new employees over the Christmas months; the BLS removes these seasonal factors so that the data can be compared across months. Seasonal factors change over time, and the data must be revised. According to the BEA, seasonal adjustments are the largest contributors to GDP revisions.

3. In the long run, the BEA and BLS may change their methodologies for estimating employment and GDP. Those changes are incorporated as revisions to earlier series.

In the end, what you see is what you get.

I take the initial release as an indication of the current trend (improvement or reduction), but wait until the second, or even third release, to attach any economic meaning to the number. Policy makers certainly wait until the final release numbers when making decisions on their policy moves.

Unfortunately, most revised data (except for GDP) is not reported in the widespread media; It is up to you to monitor the revisions.

If you would like, I can monitor the revised data for you and include it in my blog. Just comment below or float me an email if this is something that you want to see on a regular basis.

Please leave comments or suggestions. Best, Nontruths.

No comments:

Post a Comment