In our view, it is very difficult to obtain useful views about economic direction using the standard "flow of anecdotes" approach that is the bread-and-butter of many analysts. The economic data reported daily are a mix of leading, coincident and lagging indicators, often noisy and subject to revision, and without any overall economic structure. Adjusting one's entire economic views following each report, as if each somehow adds significant information, is a recipe for confusion. Treating economic data as a flow of anecdotes, without putting any structure around them, is why the economic consensus has failed to ever anticipate an oncoming recession.
We use a variety of methods to gauge recession risk. The most straightforward is to form fairly low-order indicator sets like our Recession Warning Composite (see November 12, 2007, Expecting A Recession ), that have a long historical record of accurately distinguishing recessions. These indicator sets are comprised of what might be called "weak learners" - conditions that do not in themselves have infallible records of identifying recessions, but that provide very strong signals when observed in combination with other recession flags. They include fairly straightforward conditions such as whether or not the S&P 500 is below its level of 6 months earlier, whether credit spreads are wider than they were 6 months earlier, whether the Purchasing Manager's Index is in the low 50's or below, and so forth.
As of last week, a simple average of 20 of these binary recession indicators continued to show a preponderance of signals still in place - a condition that has never been observed except alongside a U.S. recession.
Moreover, we can select random subsets of these indicators across random periods of time, in order to make the model less sensitive to exactly how it is put together. That method typically produces more variation in the overall conclusion about the economy, so the confidence in that conclusion is particularly strong when multiple models agree. At present, we observe agreement across a broad ensemble of models, even restricting data to indicators available since 1950 (broader data since 1970 imply virtual certainty of recession). The uniformity of recessionary evidence we observe today has never been seen except during or just prior to other historical recessions.
…
In short, recent U.S. economic reports have improved modestly from the clearly negative momentum that we saw in late-summer. Unfortunately, the underlying recessionary pressures we observe are largely unchanged. When we take the present set of economic evidence in its entirety, we see very little evidence of a meaningful reduction in recession risks. Indeed, the evidence from the rest of the world, both developed and developing, reinforces the expectation that the global economy is approaching a fresh contraction.