Although coincident data such as employment have exhibited steady improvement during the last several months, leading indicators continue to suggest that the economy is losing strength as we approach the second quarter of 2012. We have reached the point in time at which coincident data should begin to deteriorate and follow the leading data lower as discussed by fund manager John Hussman in his latest weekly commentary.
Leading indicators essentially place weight on the unobserved true state a few months into the future (allowing that state to be estimated today based on those observed indicators), while coincident and lagging indicators load on previous components. To see what this looks like, the following chart presents the load factors we estimate for dozens of widely followed economic variables, including one-month, 6-month and year-over-year employment gains, new unemployment claims, real consumption growth, ISM data, consumer confidence, quarterly and year-over-year GDP growth, stock returns, credit spreads, OECD leading indices, and a score of other measures. Notice that most of the variables load not on the first (most leading) economic state variable, but instead on the fourth or fifth component. That’s another way of saying that most observed economic variables actually lag the best leading indicators by several months. A good example is year-over-year growth in payroll employment, which trails year-over-year growth in real consumption with a consistent lag of about 5 months.
So what do the unobserved components look like today? In the chart below, the green line shows the average standardized value (mean zero, unit variance) of dozens of economic variables, and provides a very good summary of what can be observed directly. Note that this observable data has enjoyed a clear bounce in recent months. The blue line presents estimates of the unobserved economic states that drive the observable data. Importantly, the extracted signals lead the observed economic composite by several months. This is really simply a reflection of the underlying structure of the data – leading indicators lead, and lagging indicators lag (the full analysis uses data since 1950, but the lead of a few months is hard to distinguish visually on a long-term chart).
What strikes me about these estimates is short-lived spike in the implied economic signal between September and November. My thinking on the recent improvement in economic data was that it was primarily driven by the large intervention by the ECB near the end of last year. But even when we estimate the parameters of the model using half the data set, and then run true out-of-sample estimates of the economic signal through the present, we still get that burst of improvement in the September through November period. What’s interesting about that improvement was that it was not driven by any obvious shift in the observable data. Rather, the spike was driven by the failure of the data to deteriorate during that period to an extent that would have been expected, given the trajectory of the economic state (this is similar to shifting your expectation for a bird’s flight path not because it turned, but because it failed to turn as much as expected). In that context, we can see that the improvement in the observable data in recent months has faithfully followed the improvement in that underlying state, which actually happened months ago. Since then, however, the estimated state has deteriorated to a point that is now worse than it was last July.
This is important, because given the deterioration in the inferred economic condition between October and December, it follows that we would expect to see a clear deterioration in observable economic variables over the next 8-12 weeks. If the trajectory holds, the weakness is likely to emerge slowly and then accelerate. For example, the preliminary expectation would be for continued positive payroll growth in March (roughly 50,000-70,000 jobs) and a shift to net job losses in April. Equally important, to the extent that we observe economic variables coming in better than expected, the inferred underlying state of the economy is likely to improve sharply, as it did last September. This will take a few months of data, but it’s not going to require quarters and quarters of it. At present, we have to view the economic situation negatively, but on the optimistic side, we should also be able to abandon our own recession concerns if the observable data move off of the trajectory that we’ve imputed to-date. On this, leading measures such as consumption growth and OECD leading indices will be most important in driving our estimates of future prospects, while the employment data over the next few months will be useful in confirming the downturn we’ve seen in the inferred state estimates since November.
Notice how closely the observed composite has tracked the extracted signal of the leading data with a lag of several months in the second chart above. As Hussman notes, for this close correlation to continue, the coincident indicator should begin to turn lower sometime during the next several weeks, so it will be important to monitor economic data releases in March and April for the anticipated reversal.