It's certainly not getting any easier to evaluate the state of the
business cycle as 2012 winds down. But if there's any chance of
accurately deciding what's likely to come next, it's essential to start
the analysis with a broad review of the data. Was the economy
deteriorating in recent months? Or was it holding up fairly well? The
numbers suggest the latter, as shown in today's update of the indicators
in The Capital Spectator Economic Trend Index (CS-ETI):
[Related -Buffett's Market Indicator Flashes Red, Prepare To Sell]
Plugging the numbers into a diffusion index tells us that recession
risk is still minimal, according to the latest data, as the next chart
below shows. In other words, the majority of indicators are still
trending positive through October, albeit based on 11 of the 14
indicators in CS-ETI for last month. Yes, the incoming numbers may
change the profile from light to dark fairly quickly, although October's
final profile is all but assured to remain relatively encouraging.
November and beyond are where the primary mystery begins. If the tide
turns for the worse, CS-ETI will lose altitude, and perhaps quickly. But
based on what we know at the moment, the bias toward growth is, or at
least was, substantial.
[Related -PBoC joins other major central banks with unconventional monetary policy action]
The quality and depth of the growth is another matter—a topic that
CS-ETI doesn't address. Rather, the focus here is on quantifying the
trend via a wide spectrum of economic and financial indicators in search
of one of the more elusive reads in macro: the business cycle, the
mother of all latent variables in economics.
Translating CS-ETI's time series into probabilities across time via a probit model
also suggests that recession risk is negligible as of October. But the
various risks noted above aren't necessarily reflected in the current
probability estimate. Given the precarious state of affairs at the
moment, it's not clear how much persistence to assume for the trend in
the months to come.
Looking ahead is always a murky affair and perhaps more so these
days. But with eyes wide open, let's consider what the data's implying
for the near-term future via a sophisticated econometric technique
that's applied in a relatively straightforward way. In particular, I've
generated forecasts for each of CS-ETI's indicators, independently of
one another, using an ARIMA model. I then aggregate the results to estimate CS-ETI for the next several months.1
The process starts by filling in the handful of missing numbers for
October and then estimating each of the data sets for November and
December. It's safe to assume a fair amount of error for any one of
these forecasts, although aggregating the individual estimates can
minimize the risk a bit if some of the errors cancel each other out.
As usual, the further out we look, the higher the potential for error
generally. That said, the basic message is that CS-ETI isn't set to
tumble, or so the estimated data for the next few months suggest. A
similar set of forecasts for CS-ETI using a vector autoregression technique dispenses a similar set of estimates.
We can't take these forecasts as gospel, of course, but neither we
should we dismiss them since the ARIMA estimates for CS-ETI have been
encouraging recently. For instance, the chart above shows that the ARIMA estimates of September 28
turned out to be fairly accurate for anticipating August and
September's profile—at a time when there was still a fair amount of
uncertainty about how those month profiles would fare, particularly for
So, what's the risk with all of this? The big one is that all the
trouble swirling about isn't reflected in the current data and the
potential for a negative shock is growing. For example, the news that investment spending in corporate America continues to fall is a dark sign.
The upbeat estimates for CS-ETI through the end of the year, in other
words, may be victimized by new events that aren't yet reflected in the
latest economic reports. That's always a risk factor, of course,
although the potential for negative surprises is probably higher than
average at the moment.
Even so, it's premature to assume that the business cycle for the
U.S. is doomed. Yes, the numbers above look counterintuitive compared
with how we may "feel" about the economy at the moment, or how the
outlook appears by focusing on a relative handful of reports. If we are,
in fact, at a turning point that unleashes a new recession there'll be
clear signs of the change via a broad set of the numbers, and soon. Yes,
you could assume the worst now and start making decisions accordingly.
But that's a risk factor too, and a rather large and costly one when
considered across time.
It's inevitable that calling major turning points in the business is
destined for inaccuracy in real time. There are too many factors working
against us to expect a high degree of accuracy at any one moment. The
question is how we'd prefer to be wrong as a general proposition? Is it
preferable to make premature recession calls? Or are we better served in
trying to recognize those times when cycle has turned down after the
fact—as early as possible?
My research tells me that the latter is the way to go. Why? Many
reasons, including the compelling statistical and econometric evidence
that the odds of success are considerably higher for accurately calling
the start of new downturns shortly after they've begun vs. trying to
anticipate these events before they've started and/or based on minimal
evidence for making such assumptions. That doesn't mean we can be
nonchalant about rising risks that could push us over the edge. But
history teaches that the vast majority of error in business cycle
analysis is bound up trying to assume too much about what may happen in
the months ahead. That's certainly been a problem during the last
several years, which is overflowing with examples of premature warnings
of a new recession that, so far, never arrived.***
1. The ARIMA forecasts are calculated in R software, using Professor Rob Hyndman's "forecast" package, which optimizes the model's parameters based on each data set's historical record. ^