Goldman's NFP Estimate: -25,000, And An Extended Discussion On How They Get There
It appears that the upcoming NFP report due this Friday will be a doozy. With estimates all over the place, it seems that the actual number will be anything but what is reported due to at least four fudge factors that will likely end up stretching reality extensively in either direction. Knowing this administration's propensity to hand out blue pills with reckless abandon, it is our guess that Goldman is probably underestimating the "real" number, although we were proven wrong the last three times we distrusted the Hatzius/Tilton crack economic team.
The fundamentals of the labor market appear to be improving gradually. Layoffs are still drifting down, signs of life have emerged on the hiring front, and surveys of employment conditions in January look slightly better.
However, Friday’s employment report—particularly the payroll count for January—will be influenced by several special factors that may obscure the underlying trend: 1) Census hiring (a small positive), 2) generally colder than usual winter weather (negative), 3) the impact of annual “benchmark” revisions on the current rate of employment change (likely a modest negative), and 4) the interaction of a very bad employment year in 2009 with January seasonal adjustments (extremely uncertain, but potentially a large positive).
Our preliminary forecast, released last week, is for a decline of 25,000 in nonfarm payrolls and a slight increase in the unemployment rate to 10.1%. This forecast includes all but factor 4) above, and was formed prior to yesterday’s positive employment indicators; it remains subject to change based on information to be released Wednesday and Thursday morning.
Trying to forecast the key numbers in the monthly employment report is hard enough under ordinary circumstances. In the case of payrolls, the median forecast has missed the actual reported figure by an average of nearly 60,000 over the past two years—big in an absolute sense, though more impressive when considered as a share (about 0.05%) of the more than 130 million employed persons in the United States. But the January 2010 report poses the potential for much bigger misses because of an unusual confluence of special factors that collectively create a great deal of uncertainty about the outcome.
Leaving these potential distortions aside for a moment, the fundamentals of the labor market do appear to be improving. Layoffs continue to drift down, with the four-week moving average of new jobless claims down to about 456,000 as of the end of January, about 10,000 lower than its average in December and about 200,000 lower than its peak in March 2009. (Although weekly claims have been higher in the last two weeks, at least part of this increase appears to have been the result of processing delays for claims filed in earlier weeks—hence our reference to the four-week averages.) Announced layoffs also keep falling, with the December Challenger, Gray, and Christmas survey reporting the lowest monthly total since the month the recession began.
On the hiring side, the most recent news has also looked somewhat better. The most striking improvement has been in the Conference Board’s monthly survey of online help-wanted advertising, which showed a seasonally adjusted month-on-month increase of more than 10% in January. Although this series has a very limited history, it appears to track the basic path of hiring fairly well, improving from 2005-2007, plunging in late 2008, and then remaining moribund throughout 2009. In contrast, the Conference Board’s survey of offline job advertising has been stagnant for the better part of a year—though given the structural shift to online media, this actually may be consistent with a gradual improvement in hiring activity as well. Still, this improvement in advertising may take a few weeks or months to translate into meaningful growth in payrolls; the total number of people receiving jobless benefits (including extended benefits) appears to have been roughly unchanged since the December employment report, and the Conference Board’s latest consumer confidence survey revealed only a marginal improvement in household perceptions of job availability.
Survey data related to employment have been slightly better in January as well. The employment index in the Institute for Supply Management’s monthly manufacturing survey rose to 53.3 in January, its highest level in nearly four years. (Its non-manufacturing counterpart will be released tomorrow.) This echoed improvements in the employment component of most regional business surveys.
Taken as a whole, the plethora of data on growth and employment suggest flattish to modestly positive payrolls, with the aforementioned online job-advertising survey the only indicator hinting at a substantially better outcome.
Now to the special factors. An unusual set of crosswinds will buffet the actual and reported January figures. In increasing order of potential size:
1. Census hiring (worth perhaps +10k to +20k). The Bureau of the Census will hire hundreds of thousands of short-term employees this year to aid its data collection and processing efforts (see “The 2010 Census Jobs Effect: Easy Come, Easy Go,” US Daily, December 8). Most will be brought on in the springtime—in the past several Censuses, the first substantial increases came in February or March, with the strongest month either April or May—but we may see the beginnings of Census hiring in January. We think this effect is unlikely to be much more than an incremental gain of 20,000 jobs, however.
2. Weather (-40k to -80k). Weather turned sharply colder in the middle of December and was colder than usual for the first half of January as well. As we have shown in prior research, deviations from normal seasonal temperatures can have a substantial impact on a number of economic indicators, including employment (see “What’s With the Weather?” GS US Economics Analyst 07/02, January 12, 2007). Some commentators suggested last month’s disappointing employment report was largely the result of bad weather. We disagree. Although cold weather in the payroll survey week may have had a modest negative impact, the East Coast blizzard occurred after the payroll survey, and in general the weather has been colder—relative to seasonal norms—in the weeks leading up to the January survey. Therefore, we see weather as a factor likely to be holding back January figures rather than helping them.
3. Annual “benchmark” revision (-20k to -40k). With this report, the Labor Department will institute its annual “benchmark” revision to payrolls. This involves adjusting the March 2009 level of payroll employment to reflect a near-comprehensive count of unemployment insurance records up to that point. Last October, the Labor Department estimated that it would have to revise down the level of March 2009 payrolls by 824,000 (the exact adjustment will be revealed Friday). This will result in a downward revision to monthly changes since the previous benchmark in March 2008 (if the adjustment turned out to be 824,000, then average monthly change between March 2008 and March 2009 would be revised down by about 69,000). The Labor Department will also adjust monthly changes since March 2009 to reflect its assessment of what the comprehensive data imply for more recent employment growth; based on the pattern of past years, these revisions seem likely to be downward as well, but significantly smaller in magnitude. (As an example, in the benchmark revision of early 2007, the Labor Department revised up the previous March payroll level by about three-quarters of a million jobs, but increased the average monthly change after that point by only about 20,000 jobs.) An educated guess is a negative revision to recent monthly changes in the -20k to -40k range, which would implicitly carry through to the January report.
4. Seasonal wildcard (0 to +several hundred thousand). Believe it or not, without seasonal adjustment, payrolls always fall in January—or at least, they have for the more than 70 years of data that we examined. On a not-seasonally adjusted basis, payrolls have fallen by 2.6 million to 3.6 million every January in the past decade. A variety of factors contribute to this decline, including post-holiday season layoffs in retail, seasonal downturn in construction, and year-end purging of previously departed individuals from payroll records. This seasonal pattern is important for the following reason: After a year of exceptionally large job losses and very weak hiring, there may have been fewer seasonal employees to lay off this January. This would result in a less dramatic decline in not-seasonally adjusted payrolls, and therefore a potentially large seasonally adjusted increase.
Looking back through the payroll data over the last 50 years, in years where employment fell at least 1%, the next January’s payroll report “surprised” by an average of +0.4% relative to the underlying trend …or the equivalent of more than 500,000 jobs given the size of payroll employment today! We hasten to add we highly doubt this effect—if it recurs at all—will be this large, for several reasons: a) we don’t see the opposite effect for good years, b) we don’t see any effect in the household survey data, c) there is some evidence that the payroll impact may have become weaker in recent decades, and d) the adoption of concurrent seasonal adjustment procedures several years ago may have attenuated it still further. As we’ve stated in previous reports, we are reluctant to second-guess the Labor Department’s complex seasonal adjustment processes, so this factor is not incorporated into our current payroll forecast, but we view it as a potentially large source of upside risk. (However, our forecast of -25,000 does attempt to account for the Census, weather, and benchmark effects listed above.)
Where does all this leave us? A little confused, perhaps, and eager to see more information, including a) the employment index in tomorrow’s non-manufacturing ISM report, b) the January ADP employment report, c) data on total jobless benefit recipients for the payroll survey week, and d) the Monster.com online job advertising survey, all of which will be released in the next two days. As always, our employment estimates are subject to modification depending on how these late-breaking reports look. Whether or not we make adjustments, however, we see substantially more uncertainty in this report than the typical monthly payroll figure.
Regardless of what the Labor Department reports on Friday, it will be important to look carefully at the details of the full report before passing judgment on the strength or weakness of the labor market. A weak number that comes disproportionately in highly seasonal industries such as construction or retail, or seems to be explained by an extrapolation of large downward revisions from the payroll benchmark in March 2009, should be discounted. Conversely, a strong payroll gain that is contradicted by the household survey or (less likely) that contains a large gain in government employment should likewise be viewed as suspect. We will be inclined to treat either a very strong or a very weak employment report—particularly the payroll portion—with a greater than usual degree of skepticism this month.