Can We Intelligently Estimate Recession Risk?

Media chatter about economic recessions tends to come in three flavors. One is the rarely seen diligent species that considers how the macro trend is evolving currently, and may evolve over the next month or so, based on a diversified set of metrics. Then there’s the long-run forecast that makes a valiant effort to divine the future a year or more in advance. The third alternative, which is by far the most popular flavor, is to zero in on one or two indicators, usually as a prelude to declaring that the end is nigh… any day now. But only one of the three is relatively reliable while the other two can safely be labeled as lighted-hearted diversions bordering on entertainment.

On the matter of the long-range forecasts, Bloomberg columnist Barry Ritholtz wisely summarizes why we should pay no mind to these generally hollow labors. “Saying a recession might occur within the next four years is a statement that contains almost no information.”

During the 20th century, there were 20 recessions, or one every five years on average.  In other words, if you predict a recession within the next four years you will be accurate on average about 80 percent of the time.  It’s only slightly more useless than your local weather forecaster predicting that temperatures will fall this winter and rise next summer. This prediction is, of course, absolutely true and of no value whatsoever.

The odds of success aren’t any better by looking for the next downturn by cherry-picking indicators. But that doesn’t stop any one from trying. The latest twist is looking for macro insight via Donald Trump’s presidential campaign. But as Ben Casselman at FiveThirtyEight correctly advises, it’s high time for certain pundits and economists to “Stop Saying A Trump Win Would Cause A Recession.” Why? The basis for such forecasts is wafer thin at best, falling somewhere between rank speculation and the macro equivalent of a Rorschach test.

Some cherry-picking aficionados know better and instead focus on valid indicators, albeit one at a time. A popular theme of late is to assume the worst based on the ISM Manufacturing Index. This benchmark’s weak readings in recent history translate into a high risk of a US recession, according to former hedge fund manager Raoul Pal. The one-time co-manager of the GLG Global Macro Fund in London tells MarketWatch that a recession will strike at some point in the next 12 months, reasoning that the soft manufacturing sector, by way of the ISM data, is the proverbial canary in the coal mine.

Maybe, but maybe not. Indeed, Pal made a similar prediction last December. Nearly a year later, the next recession is still somewhere just over the horizon.

To be fair, the ISM index (unlike a Trump-based analysis) is a legitimate variable for analyzing the business cycle. But no indicator is infallible. In fact, if you look at a reasonable set of macro and financial benchmarks that, in the aggregate, explain US economic activity, there’s usually some corner that’s stumbling. But confidently declaring that recession risk is elevated requires a higher bar.

How can we reliably evaluate the business cycle? The first step is to monitor a broad, representative data set and consider what the numbers show in the aggregate about recent/current conditions. If you must forecast, do so carefully, i.e., with a robust methodology for estimating future values for the next two to three months at most. In fact, that’s exactly the framework for The Capital Spectator’s monthly US economic profiles, including yesterday’s update.

How has this model fared? Quite good, actually. Consider, for instance, what the numbers reflected back in mid-March, when it was tempting to assume that the economy was inescapably slipping over to the dark side. Yet the hard data told us that recession risk was low and the near-term projection hinted at the likelihood that a rebound was coming, as this March 17 analysis reported. In the months ahead, the case for expecting a moderately firmer trend strengthened—a reality that’s now obvious.

True, economic growth is still wobbly and the potential for trouble continues to lurk. At some point, the recovery will end, perhaps sooner than expected. History, however, offers two key lessons for monitoring business cycle risk. First, a truly high-confidence estimate of the recession start date is inevitably a backward-looking effort. The best we can hope for is a relatively short lag between a recession’s start date and when we can reliably declare that the turning point has (recently) arrived. Second, robust efforts at modeling the future on this front should be limited, extending out 2-3 months at most.

Meantime, more is generally better, which is why the weekly US Business Cycle Risk Report (US-BCRR) reviews a mix of proprietary models and macro indexes published by the Federal Reserve banks. As a baseline, US-BCRR aggregates a spectrum of macro benchmarks via the Composite Recession Probability Index. On that note, here’s how this benchmark stacks up at the moment, based on data through Oct. 20:

crpi-probit

Recession risk, in other words, is low at the moment. To be precise, the probability is virtually nil that NBER—the official arbiter of US recessions—will declare that a new downturn has recently started. But that’s not written in stone… for tomorrow and beyond.

Yes, another recession is coming. When? No one really knows. But this much is clear: trying to estimate recession risk far into the future, or by looking at one or two indicators, is prone to false signals. Fortunately, there’s a better way: use a diversified set of macro and market benchmarks. There’s still no guarantee, of course. But if you absolutely, positively need a dependable measure of business-cycle conditions, the usual routine won’t cut it.

 

 

 

 

5 thoughts on “Can We Intelligently Estimate Recession Risk?

  1. Nick de Peyster

    Can you show more data points on that graph? 100% accuracy on a sample size of two recessions is better than 50% or 0%, but it’s not that big a sample size.

    Nick de Peyster
    undervaluestocks.info

  2. James Picerno Post author

    Nick,
    CRPI’s history dates to 1998 due to the limits for some of the underlying components. For a longer perspective on one of the benchmarks used in CRPI, see the ETI/EMI graph here:
    http://www.capitalspectator.com/us-business-cycle-risk-report-17-march-2016/
    Keep in mind that we’re looking at history via revised data. How do the vintage results compare? They’re encouraging, at least for ETI during the 2000-2012 period, as discussed in my book, Nowcasting The Business Cycle, charts 12.4 and 12.5
    –JP

  3. Pingback: Estimating Recession Risk - TradingGods.net

  4. Bob Dieli

    Since there has been a recession in the first term of every Republican president since the turn of the twentieth century, one might not be far off in asserting that President Trump might suffer a similar fate.

    Since 1955 my indicator, the Enhanced Aggregate Spread, has called all the cycle peaks and troughs, except those associated with the 1960 recession, from about a year away from their actual dates.

  5. James Picerno Post author

    Bob,
    I agree that a Trump presidency implies elevated risk for the economy (and beyond) on a number of fronts. But from an econometric perspective, projecting a recession on this basis is speculation.
    The Treasury yield spread, and related metrics such as your Enhanced Aggregate Spread, are more reliable and arguably second to none among indicators for the simple reason that there’s data history here and it looks encouraging as a predictor. The question is whether yield spreads and the like will be reliable going forward? Given the unusual state of monetary policy these days, it’s reasonable to wonder if the next time will be different.
    –JP

Comments are closed.