John P. Hussman, Ph.D.
Over the past few weeks, investors used to setting their economic expectations based on a « stream of anecdotes » approach have seen their economic views evolve roughly as follows:
« After a brief ‘scare’ during the third quarter, economic reports have come in better than expectations for weeks – a sign that the economy is on a gradual but predictable growth path; Purchasing managers reports out of China and Europe have firmed, and the U.S. Purchasing Managers Indices have advanced, albeit in the low 50’s, but confirming a favorable positive trend, and indicating that the U.S. is strong enough to pull the global economy back to a growth path, or at least sidestep any downturn; New unemployment claims have trended gradually lower, and combined with a surprisingly robust December payroll gain of 200,000 jobs, provides a convincing signal that job growth is on track to improve further. »
I can understand this view in the sense that the data points are correct – economic data has come in above expectations for several weeks, the Chinese, European and U.S. PMI’s have all ticked higher in the latest reports, new unemployment claims have declined, and December payrolls grew by 200,000.
Unfortunately, in all of these cases, the inference being drawn from these data points is not supported by the data set of economic evidence that is presently available, which is instead historically associated with a much more difficult outcome. Specifically, the data set continues to imply a nearly immediate global economic downturn. Lakshman Achuthan of the Economic Cycle Research Institute (ECRI) has noted if the U.S. gets through the second quarter of this year without falling into recession, « then, we’re wrong. » Frankly, I’ll be surprised if the U.S. gets through the first quarter without a downturn.
Three basic issues are at play. One is that analysts aren’t making distinctions between leading, coincident and lagging data. The second issue is that there is little effort to measure the predictive strength of a given economic data point (or set of data points) in explaining subsequent movements in the economy. The third is that analysts seem to be forming expectations report-by-report (what I call a « stream of anecdotes » approach) instead of taking those reports in context of the full ensemble of data that is available at each point in time.
Let’s examine the seemingly most « compelling » data point first – the fact that December payrolls grew by 200,000. Surely that sort of jobs number is inconsistent with an oncoming recession. Isn’t it? Well, examining the past 10 U.S. recessions, it turns out that payroll employment growth was positive in 8 of those 10 recessions in the very month that the recession began. These were not small numbers. The average payroll growth (scaled to the present labor force) translates to 200,000 new jobs in the month of the recession turn, and about 500,000 jobs during the preceding 3-month period. Indeed, of the 80% of these points that were positive, the average rate of payroll growth in the month of the turn was 0.20%, which presently translates to a payroll gain of 264,000 jobs.
Likewise, in 5 of the past 10 recessions, the ISM Purchasing Managers Index was greater than 50 just weeks before the recession began, and the new orders component of that index was greater than 50 in most cases, immediately prior to the recession.
Very simply, neither a strong monthly employment gain nor a slight uptick in the PMI are informative signals that recession risk has eased. Both the PMI and the level of payroll job growth are what one might call « weak learners. » It’s not that these figures aren’t useful – just that neither of them has a particularly good record by itself of signaling recessions. As it happens, a PMI below 54, coupled with year-over-year payroll growth below 1.3% is a stronger « learner » than either of the two data points individually (see the 2007 comment Expecting A Recession ). That combination – which is actually alternate Condition 4 of our Recession Warning Composite – remains in place at present, as are the other conditions in that Composite. Our more complex ensemble models also indicate strong recession risk.
The chart below provides a good picture of the behavior of non-farm payroll growth in the months before and after a recession begins, based on all U.S. postwar recessions. Notice in particular that in the month a recession starts, payroll job growth has not only been positive in 80% of cases, but has actually been higher, on average, than the three preceding months. Neither the level of job growth nor its short-term trend had any « leading » information content at all about the subsequent direction of the economy.
Notably however, the month following entry into a recession typically featured a sharp dropoff in job growth, with only 30% of those months featuring job gains, and employment losses that work out to about 150,000 jobs based on the present size of the job force. So while robust job creation is no evidence at all that a recession is not directly ahead, a significant negative print on jobs is a fairly useful confirmation of the turning point, provided that leading recession indicators are already in place.
I’ve discussed the « positive surprises » argument (see When Positive Surprises are Surprisingly Meaningless ) and the negative implications of the European ISM, despite last month’s uptick (see The Right Kind of Hope ) in other recent comments. Suffice it to say that broadly speaking, the recent « surprises » in the data reflect minor fluctuations within overall levels that remain fairly tepid, and more importantly, that remain clearly unfavorable as an ensemble.
How to spot a leading indicator
I want to begin this section with a simple statement – I do not hope for a recession. Rather, that is the expectation that the data forces on us. Frankly, much of my time in recent weeks has been devoted to analyzing data in the hope that a more compelling case could be made for avoiding a recession, since that would free us to be more constructive should market internals improve. But that’s not what the evidence indicates here, and the recent economic data hasn’t reversed that conclusion – not yet at least. I like to think I do a good job of showing you the same things that I am seeing. I don’t challenge rosy outlooks because I enjoy being defensive – I don’t. In fact, I can hardly wait for market conditions where risk is priced appropriately. It’s just that Wall Street’s simplistic cases that stocks are cheap and recession is « off the table » just don’t hold water when we examine the data.
I’ve written a lot in recent months about the distinction between leading, coincident and lagging indicators. One of the ways to distinguish these is to calculate a whole set of correlations between an indicator and what it is intended to predict, using various leads and lags. If a given indicator is correlated with whether or not the economy was in a recession say, 6 months later, we would say that the indicator has a certain amount of usefulness as a « leading » indicator. In contrast, if a given indicator is correlated with whether or not the economy was in a recession say, 6 months previously, we would say that the indicator has a certain amount of usefulness as a « lagging » (or « confirming ») indicator. The stronger the correlation at a given point, the more useful that indicator is as a leading, coincident, or lagging indicator. Importantly, it is the strength of the correlation, not simply where the correlation curve peaks, that defines the usefulness of the indicator. [Geek’s note – it’s more elegant to do this work in the frequency domain, but correlations work nicely for the purposes here].
The chart below is a little bit busy, but presents the correlation profile of a variety of widely followed indicators, as well as an ensemble of recession indicators we track (see Measuring the Probability of Recession in the September 5 comment). In the chart below, month zero represents the start of a recession.
Notice that about 9 months prior to a recession, the Conference Board Leading Economic Indicators, the ECRI Weekly Leading Index and the 6-month change in the S&P 500 often show some weak leading characteristics, but the correlation is too small to make inferences very reliable. Advancing to about 6 months prior to a recession, a few more indicators begin to show a weak correlation with the oncoming recession, including our own ensembles, as well as the average of Fed surveys (such as Philly Fed and the Empire Manufacturing survey) and the ISM Purchasing Managers Index. Still, at that point, the correlations are typically fairly weak. Though none of these indicators are particularly good at anticipating a recession even 6 months out, the Conference Board Index of Leading Economic Indicators (LEI) has historically had a slight edge looking two quarters ahead (the LEI makes a very interesting study on its own here, so more on that below).
Once a recession is within three months away, the strongest leading indicators are our own ensemble and the ECRI Weekly Leading Index (though I expect that an ensemble of ECRI’s other indicators, such as the long-leading and coincident measures would, in combination, give an even stronger overall signal than the WLI alone). The 6-month change in the S&P 500 approaches its strongest correlation with an oncoming recession with only a 1-3 month lead, suggesting that investors wishing to anticipate recession-linked stock market weakness would want to focus on indicators have even better leading characteristics than stocks themselves.
Once a recession hits, our recession ensemble, the ECRI Weekly Leading Index, and the average of multiple Fed surveys have the strongest likelihood of confirming the downturn in real-time. Immediately following entry into the recession, as noted earlier, payroll growth tends to turn negative. Though recessions tend to be preceded by sub-par employment growth over the preceding 3-12 month period, the 3-month growth rate of payrolls actually acts as a bit of a lagging indicator, reaching its highest correlation with a recession – not surprisingly – about 3 months after the recession starts.
New claims for unemployment have very slight short-leading usefulness, but new claims, the unemployment rate, and the slope of the yield curve (flattening) actually have much better lagging characteristics, so these should be used primarily to confirm an ongoing recession (particularly if the NBER hasn’t made an official determination yet), rather than to anticipate a downturn. The yield curve generally flattens significantly coming into a recession, but the change in the yield curve (not plotted) is also most useful as a lagging indicator. Consumer confidence has mixed characteristics, with weak leading characteristics and somewhat greater usefulness as a lagging indicator, but in any case is too much of a « weak learner » to be used in isolation.
At present, our own recession ensembles, as well as ECRI’s official views, remain firmly entrenched in the recession camp. This feels more than a little bit disconcerting, as the entire investment world appears to have the opposite view. My problem is that the data don’t support that rosy « U.S. leads the world off the recession track » scenario. Leading data leads. Lagging data lags. Weak data is weak data. To anticipate a sustained economic upturn here would require us to place greater weight on weak, lagging data than we presently place on strong leading data. It’s really that simple. If the evidence turns, we will shift our view – and frankly with some amount of relief. At present, though, we continue to expect a concerted economic downturn.
The LEI and monetary bias
One of the interesting aspects of present conditions is the apparent disconnect between the Conference Board’s index of leading economic indicators and the ensemble of other economic indicators that we follow (including ECRI’s indices). The LEI is a composite of 10 measures, including the average workweek, jobless claims, new consumer orders, capital equipment orders, vendor deliveries, building permits, consumer expectations, stock prices, the yield curve, and real M2 money supply.
What’s problematic here is that close to half of the weight in the index goes to the two monetary components – the yield curve, and real M2. I suspect that this is a legacy of inflationary business cycles where monetary tightening in response to inflation was the typical event preceding recessions, but it adds noise in the present environment, where the primary economic risks are related to leverage and credit strains. Remember that at present, monetary policy is way out on the « liquidity preference » curve, to an extent that is historically unprecedented (see Monetary Policy in 3D ). Normally, there is a general, if weak, linear relationship between monetary variables, interest rates and economic activity. But given the current scope of monetary policy, M2 velocity has collapsed (and moves as a perfect inverse of M2 itself), and interest rates are at the zero bound, so these variables are essentially detached from economic activity. So you’ve got two highly weighted variables in the index that have gone almost perfectly horizontal with respect to their effect on the economy. The crisis in Europe has triggered a flight of time deposits from European banks to U.S. banks, which shows up as a further boost to M2, which has driven much of the advance in the LEI.
Notably, the Conference Board announced last week that they will replace real M2 with a new « Leading Credit Index » component, among other changes, which will be reflected in the January 2012 release.
That change makes sense. If you’re going to put nearly half of your weight on monetary variables, it’s really only sensible if about half of your predictive power resides in those two variables, but in the case of the LEI, that’s not true at all. Below, I’ve weighted the present components of the LEI in proportion to their correlation with subsequent recessions (using a 6-month smoothed growth rate for the non-stationary ones such as stock prices, capital orders and so forth, and standardizing the values of each to have zero mean and unit variance prior to weighting). The chart also presents the simple average of the non-monetary components of the LEI, as well as the smoothed growth rate of the actual published index, similarly scaled.
Notice that unlike the typical behavior of the LEI in prior recessions, the LEI did not spike down to nearly the same extent as the nonmonetary components during the downturn that began in 2007, thanks to unprecedented monetary policy actions. Likewise, the LEI has held up much better in recent months than either its non-monetary components, or its accuracy-weighted components, also as a result of monetary policy that is outside of historical norms and stretched far along the zero bound.
A troublesome issue here is that once the non-monetary components of the LEI have turned negative to the extent we observe presently (again, on a 6-month smoothed basis), we find only one instance (a brief signal in the late-1960s) that was not associated with an actual recession. Below, the red bands denote official NBER-dated recessions. Downturns in the non-monetary components of the LEI are highlighted in blue.
Our own recession ensembles remained unfavorable last week, and the ECRI Weekly Leading Index deteriorated to -8.2, from -7.6 the previous week. The 3 month growth rate of non-farm payroll employment – despite last month’s employment gain – is among the lowest 13% of all historical observations. The 6-month change in the S&P 500 is among the lowest 20% of historical observations. The current value of ECRI’s Weekly Leading Index is among the lowest 9% of all historical observations. We don’t disregard the marginal improvement in various economic measures in recent weeks. It’s just that those marginal improvements are either too small or too statistically uninformative to be helpful in shifting the evidence.
In sum, the balance of leading evidence continues to indicate a very high likelihood of an oncoming recession. We respect the various marginal improvements in the data in recent months, which do take the probability to less than 100%, but that is a far cry from suggesting that recession risk is anywhere close to being « off the table. » Recession is not a certainty, but it remains the most probable outcome at present.
All of that said, significant new strength in stocks – particularly if broadly based – further contraction in new unemployment claims, well-defined (not just marginal) improvement in a broad sampling of Fed economic surveys, and a reversal in industrial commodity prices, among other factors, would provide a good basis to ease recession concerns. If that was coupled with confirmation by a reversal in ECRI’s measures, the evidence that weighs down our economic views would become dramatically lighter. But that’s what we need – evidence. Well correlated, strong, leading evidence.
Even if we allow for the possibility of improvement, my impression is that the potential outcomes for the market are very asymmetrical. Investors now expect pleasant, if gradual, economic progress, convinced by a stream of economic anecdotes in recent weeks. That hopeful expectation is already largely reflected in the overvalued, overbought, overbullish condition of the market. If the economy does in fact improve, we may observe further upside progress, but again – this is largely reflected in the advance that stocks have already enjoyed. The asymmetric risk is the potential for great disappointment if the economy does fall into a contraction – as well-correlated leading evidence continues to suggest.
As of last week, the Market Climate for stocks remains « hard negative » – characterized by conditions that cluster among other historical instances that usually resulted in « whipsaw » declines on the tail of overbought rallies. About 30% of these instances did resolve into further gains, and we aren’t frozen to a defensive stance. As noted above, there are certainly developments that would mute our economic concerns and even allow for a modestly constructive position despite what we continue to view as an overvalued market from a longer-term perspective. Presently, we would need at minimum a further improvement in market internals – primarily breadth and leadership. The situation would also be helped by clear strength in Fed surveys and a further retreat in new claims. Without this sort of broad-based improvement, the modest « positive surprises » we’re seeing are still too tightly centered in a range that really doesn’t change the picture at all. Strategic Growth and Strategic International remain well-hedged. Strategic Total Return continues to carry a duration of about 3 years in Treasuries, and we used the spike advance early last week to clip a few more profits in our precious metals shares, taking our exposure to a still-constructive but comfortable 12% of assets.
As a final note, given our pointed economic and market concerns, I’ve included a chart below showing the profile of past major market declines, mostly as a tool to display the significant variability of outcomes. The chart shows major U.S. market declines as a « stochastic » – the bull market high being 1.0, the bear market low being zero. Time is measured in days, with the bull market peak set at 100 days in each (denoted by the red arrow).
The main regularity you’ll notice is that the first 6-8 weeks or so off the top are uniformly bad, typically inflicting about one-fifth to one-third of the eventual peak-to-trough loss. That initial decline is then typically followed by a rebound of highly variable duration, lasting anywhere between 2-5 months and usually recovering half to two-thirds of the initial decline (denoted by the green arrow). As a result, by 3-6 months into a major market decline, the market is often not far from its original peak (a tendency I noted last May in Extreme Conditions and Typical Outcomes ). Unfortunately, that is not a rule that one would want to rely on, because when it has failed, it has often failed spectacularly.
Each decline has its own character, so there is no predictable point at which breakdowns occur. Notice that the bottom of the decline is also highly variable, so outside of quoting a broad range from 3 months to 3 years, with the average at a bit less than 18 months, major market declines don’t have a predictable duration. The upshot is that major declines are not diagonal and do not follow well-behaved patterns. It’s exactly that variability that makes it dangerous to « finesse » them excessively, and advisable to stick with the broad evidence, recognizing that there will be a lot of unpredictable short-term volatility. For our part, we remain defensive here.