One of the great challenges of investing is the distinction between hindsight and foresight. Hindsight treats each major advance, each market crash, each recession and each expansion as if their turning points were obvious, and extrapolates prevailing trends as if their continuation is equally obvious. Foresight is much messier, because it deals with unknowns and unobservables. It recognizes that major financial and economic events are often hidden from view when they are actually already in motion. Foresight requires the willingness to rely on data that tends to precede important outcomes (recessions, market crashes, durable long-term returns), even when those outcomes can't be observed in recent economic and market behavior that we can see and touch. Most importantly, hindsight creates the illusion that uncertainty is never very great, and risk management is never very challenging. Foresight demands a much greater appreciation for randomness, noise, uncertainty, risk management, and stress-testing.
Presently, there seems to be an unusually wide gap between hindsight and foresight, both in the financial markets and in the economy. In both cases, forward-looking evidence suggests weak outcomes, but recent trends encourage optimism and risk-taking. Rather than sugar-coat these uncertainties and minimize the messy divergences in the data, I think the best approach is to review the evidence, warts and all, including economic risks, market conditions, and the strengths and limitations of our own investment approach.
The economy: weak leading, lukewarm lagging
The most important news in the financial markets last week was undoubtedly the January employment report, which showed a 243,000 increase in non-farm payrolls, outstripping the 150,000 figure expected by a consensus of economists. Two questions immediately arise. What does this news do to change the likelihood of an oncoming economic recession? And what does this do to change the prospects for the returns and risks in the financial markets?
With regard to recession risks, the January employment report increases the divergence between leading evidence on one hand, where the broad set of data remains in a conformation that is almost exclusively associated with oncoming recession, and the more favorable, if lukewarm, signs from coincident indicators (e.g. employment, purchasing managers index, weekly unemployment claims) and lagging indicators (e.g. unemployment rate).
There is always some element of information when divergences and inconsistencies emerge in the data. But you can't extract that information very well by throwing all the data in a high-speed blender and just taking the average. Rather, inferences should be based on which indicators are relevant in which contexts. Specifically, we know that leading indicators lead, lagging indicators lag, and coincident indicators are coincident. Given that coincident indicators have improved in recent months, we can easily conclude that economic activity has also improved in recent months. But to make a forward-looking statement, we can't just extrapolate those improvements, because we know that coincident data doesn't extrapolate reliably at all. So we have to focus primarily on leading indicators instead.
And that's our dilemma here. It's undeniable that coincident measures have improved in recent months, but we have not seen a convincing turn in the leading data. So either the leading data will uncharacteristically lag the recent improvements, or what remains more likely, the coincident data will taper off and deteriorate. I'll reiterate that we aren't table-pounders for recession, and that we certainly don't hope for a recession (though we would welcome higher prospective investment returns that would be brought about by lower market valuations). Overall, an economic downturn remains the most likely prospect, and it's not at all clear that the latest employment report changes that risk. I think the best way to see why, as always, is to show you the same things that I'm looking at.
To begin, it's useful to understand how the Bureau of Labor Statistics calculated the 243,000 increase in employment that it reported for January. Total non-farm employment in the U.S., before seasonal adjustments, fell by 2,689,000 jobs in January. However, because it's typical for the economy to lose a large number of jobs after the holidays, largely in retail trade, construction, and manufacturing, the BLS estimated that the "normal" seasonal decline in employment should have been 2,932,000 jobs in January. The difference between the two numbers, of course, was 243,000 jobs, which was reported as an increase in employment. The fact that the size of the seasonal adjustment was more than 12 times the number of reported jobs, and more than 30 times the "beat" in economists' expectations, should provoke at least some hesitation in taking the number at face value.
Notably, the January 2011 and 2012 seasonal adjustment factors ( seasonally adjusted payrolls divided by unadjusted payrolls) have been the two largest factors used by the BLS since the 1960's, at 1.0166 and 1.0165, respectively. This compares with a January seasonal factor of 1.0155 a decade ago, and a factor of 1.0152 as recently as 2009. Now, a range of 0.0014 in the seasonal factors for January may not seem like much, until you consider that non-seasonally adjusted payrolls are presently about 130 million jobs, so variation in the seasonal adjustment factor alone amounts to a difference of 182,000 reported jobs. I'm not suggesting there's anything nefarious going on here, it's just that part of what we're seeing here is most likely a statistical artifact of the adjustment process.
Moreover, we've had a remarkably mild winter in the U.S, particularly in January, and it's clear that this has favorably affected both construction and retail activity. Ironically, however, nothing in the seasonal adjustment actually adjusts for this purely seasonal effect. If the mild winter weather reduced the "normal" number of January layoffs by just 3-4%, that would account for the entire amount by which the January employment number "beat" economists' expectations.
Our understanding is that most economic series are seasonally adjusted using the same algorithm from the Census Bureau, and indeed, we've been able to closely replicate the labor department's adjustments to various data series using that software [Geek's note: take the option to log-transform the data]. One concern we are aware of is that some data providers such as the ISM use exceptionally short windows (such as 5 years) to estimate their adjustment factors, which appears to invite a large amount of statistical noise in these factors due to the deep and unusual weakness of the 2008-2009 period.
As a side note, because the ISM incorporated the newly released seasonal factors from the Department of Commerce, we saw some significant downward revisions in the December ISM figures that made the January figures appear stronger. For example, the January figure for new orders was 57.6, the same as the original December figure. But since the December figure was revised down to 54.8, the January report appeared to be an improvement. Compared with the original December figures, both production and employment actually dropped. The original December PMI was 53.9, inching higher to 54.1 in January, primarily due to higher inventories. The upshot is that the composite signal from Purchasing Managers Indices and regional Fed surveys has improved modestly, but the overall picture remains lukewarm.
I certainly don't want to push that argument to the point of suggesting that recent reports are irrelevant, or that they don't reflect actual improvements. There is enough conformity across multiple pieces of economic data to conclude that the positive economic performance of late is not purely statistical noise. The real issue is the extent, durability, and "leadingness" of those improvements, where we continue to be adamant that lagging data (such as the unemployment rate) should not be expected to lead. Indeed, job growth has typically been reasonably positive in the 1, 3, 6 and 12 months prior to a recession. Job growth was positive in the month prior to 8 of the past 10 recessions, and in the 3 months prior to 9 of the past 10 recessions. In other words, we shouldn't expect weak job reports to lead recessions, though the year-over-year growth rate in payrolls invariably drops below 1.5% in the early months of a downturn (a level that we're still below).
In any event, a reasonable interpretation of the January employment report is that fewer jobs were lost in January than the BLS estimated that the economy should have lost on the basis of seasonal patterns. The economy is essentially bouncing around the flatline, and the main question is how much longer we can avoid a negative shock of any kind.
On a related note, we've seen a few suggestions that because the latest Purchasing Managers Index came in above 54 (the January figure was 54.1) and the S&P 500 is now above where it was 6 months ago, any concern about a recession is now invalidated as two of the four components of our basic Recession Warning Composite (see Expecting A Recession) are no longer active. Put simply, this is not how this particular "Aunt Minnie" works. At least one signal from the Recession Warning Composite has appeared either just before or during each of the past 8 recessions, without false signals (the PMI never hit that 54 level in 2010), but those signals are typically not "step" impulses that stay continuously active. Rather, the appearance of even one composite signal is, in and of itself, cause for some recession concern. But given the simplicity of the Recession Warning Composite, a much broader set of evidence is clearly preferable, much of which has been the subject of numerous recent weekly comments.
As it happens, I received identical criticism of my recession concerns in May 2008, when the S&P 500 briefly rose above its level of 6 months earlier, and credit spreads briefly retreated from their levels of 6 months earlier, leading to suggestions that even our own recession evidence had "turned." At the time, the Fed was easing, Congress had passed an economic "stimulus" in the form of tax rebates, economic reports were coming in tepid but ahead of expectations, and any concern about recession was viewed with disdain. The S&P 500 had advanced about 12% over a period of about 10 weeks, and was only about 8% below its 2007 peak, having recovered much of what was (in hindsight) the initial bear market selloff. This was the most recent example of the "exhaustion syndrome" that emerged again last week (see Warning: Goat Rodeo).
At the highs of that May 2008 advance, I observed "The reality is that as recessions develop (and I continue to believe the U.S. faces a much more significant downturn than we've observed to date), the data can take months to accumulate to a compelling verdict, and in the meantime, speculative pressures can remain alive" (see Poor Fundamentals with Borderline Market Action). A few weeks later, the surreal calm in the face of seemingly obvious risks prompted the title of my June 2, 2008 weekly comment - Wall Street Decides to Close its Ears and Hum, where I noted "investors appear to be viewing the recent period of weak but not terrible economic news as a signal that the worst is behind us and that clear conditions are ahead." Memorably, that was not the case.
Though I don't expect a 2008-type collapse here, I would view a 25% market decline as only run-of-the-mill. I don't view the probability of recession as 100%, but the leading evidence continues to indicate recession as the most likely probability. While we track a very broad set of data, a crude but useful rule of thumb is that the combination of a) an upturn in the OECD leading indicators (U.S. and total world), coupled with b) a turn to positive growth in the ECRI weekly leading index, has generally been a good sign that recession risk is receding. Those shifts can occur fairly quickly, but we don't observe them at present.
We aren't oblivious to the comfortable reports from indicators that typically lag the economy, but we also see disturbing recession risks in indicators that typically lead the economy. The problem is that even though investors know that lagging data lags, it deals with actual recent outcomes that can be "seen and touched." In contrast, even though investors know that leading data leads, it deals with unobserved future prospects that have not yet been realized. It's natural to focus attention of what can be seen and touched, even if it's not indicative of the future.
