Fed "Workhorse" Model Says Odds of Recession in Next Year Only 3.56%; What are the Real Odds?

Of all the ridiculous opinions as to why the US is not about to enter a recession, the Fed's "Workhorse" Model is at the top of the list.

Torsten Sløk, chief international economist at Deutsche Bank, is taking the optimistic route by drawing attention to a certain economic model that currently puts the chance of an imminent contraction in the single digits. The Federal Reserve's so-called probit model looks at the difference between 10-year and three-month U.S. Treasury rates to gauge the probability of a U.S. recession over the next 12 months.

"The Fed has a workhorse recession model where [the] yield curve today is a predictor of future recessions, and running the Fed’s probit model with today’s values for 10-year and 3-month rates shows that there is currently a 4 percent probability of a recession over the next 12 months," Sløk said in an e-mail.

The Model

Prepare to have your eyes gloss over because here is the model.



Highlights in yellow are mine. Note that two constants are estimated using data from January 1959 to December 2005. Not only that, the constants were fitted to match what happened. Lovely.

Practical Considerations

Those hand-picked constants happened to work in prior recessions with varying degrees of success as noted in a New York Fed paper appropriately called The Yield Curve as a Leading Indicator: Some Practical Issues.

The report failed to mention the most practical of practical issues: It's damn hard for the 3-month to invert with 10-year treasuries when the Fed has artificially held short-term yields close to zero.

Of course there is a practical reason for the Fed not pointing out that practicality. The article was written in 2006 before short-term yields went to zero.

You might have thought chief international economists would have stopped to consider such practical issues, but you would be wrong.

One might also have thought such issues would have crossed the minds of the New York Fed, but please banish that thought as well.

The New York Fed research still promotes this ridiculous model on its page The Yield Curve as a Leading Indicator.

You can download the current data to play with in Excel. You can also download the current 12-month look-ahead probability chart.



Yield Curve and Recessions

Despite the obvious uselessness of the indicator under current conditions, others are on the same bandwagon.

Pater Tenebrarum at the acting man took them to task back in 2014 with his post Yield Curve and Recessions.

One popular theme gets reprinted in variations over and over again. Here is a recent example from Business Insider, which breathlessly informs us of the infallibility of the yield curve as a forecasting tool: “This Market Measure Has A Perfect Track Record For Predicting US Recessions” the headline informs us – and we dimly remember having seen variants of this article on the same site at least three times by now:

At a breakfast earlier today, LPL Financial's Jeffrey Kleintop noted that the yield curve inverted just prior to every U.S. recession in the past 50 years. "That is seven out of seven times — a perfect forecasting track record," he reiterated.

The yield curve is inverted when short-term interest rates (e.g. the 3-year Treasury) are higher than long-term interest rates (e.g. the 10-year Treasury yield). "The yield curve inversion usually takes place about 12 months before the start of the recession, but the lead time ranges from about 5 to 16 months," wrote Kleintop in a recent note. "The peak in the stock market comes around the time of the yield curve inversion, ahead of the recession and accompanying downturn in corporate profits."

This is it! The holy grail of forecasting, Jeffrey Kleintop has discovered it. You'll never have to worry about actual earnings reports, a massive bubble in junk debt, the sluggishness of the economy, new record levels in sentiment measures and margin debt, record low mutual fund cash reserves, the pace of money supply growth, or anything else again. Just watch the yield curve!

Holy Grail

No need to sell now. The holy grail has been perfected. When the yield curve inverts, wait another six months for stocks to peak then sell.

Tenebrarum points out the amazing success of that method for Japan.



The model last worked in1991. Since then, the yield-curve method has had a perfect track record of failure. In at least 6 recessions since, counting one in 2014 the above chart does not show, the model has failed.

Recession Odds

I think there is a 25% chance or better the US already went into recession as of December 2015 or January 2016. A couple of bad jobs reports will seal the deal, and it may not even take that. But that's just a guess. I have no economic formula for economists to go gaga over.

John Hussman does have a nice discussion in his latest post Wicked Skew: When Extreme Losses are Standard Outcomes.

On the economic front, I continue to believe that a U.S. recession is not only a risk, but is now the most probable outcome. As I noted last week, among confirming indicators that generally emerge fairly early once a recession has taken hold, we would be particularly attentive to the following: a sudden drop in consumer confidence about 20 points below its 12-month average (which would currently equate to a drop to the 75 level on the Conference Board measure), a decline in aggregate hours worked below its level 3-months prior, a year-over-year increase of about 20% in new claims for unemployment (which would currently equate to a level of about 340,000 weekly new claims), and slowing growth in real personal income.

Last week, new claims for unemployment jumped to 293,000, a level we’ve observed only once since last April. Even at this early point (given that employment measures are among the most lagging economic indicators), we already observe a pickup in claims from last year’s lows. To put this increase into perspective, the chart below shows points where the ISM Purchasing Managers Index was below 50, the S&P 500 was below its level of 6 months prior, and the 4-week average of initial unemployment claims was at least 5% above its 12-month low. While a year-over-year increase in unemployment claims closer to 20% would be a more reliable confirmation of recession, it’s clear that even here, the current combination of evidence is more associated with recession than not.

One Model is Wrong

Seven out of seven times we have been in these conditions, the economy was close to or in recession.

That's one possible model. It contrasts with the Fed's model which says the 12-month look ahead odds are only 3.56%.

Take your pick.

Some economists take the Fed's model.

Why? Because the Fed put a formula to it. That makes it official even though the model has no scientific basis under current conditions.

Simply put, some economists refuse to think.

Reflections on Economic Modeling

Constants α and β are on the verge of massive revisions.  After the next recession (which we may already be in), the Fed will see fit to dream up a revised formula that will take into account conditions at zero bound.

That revision will work for the last two recessions but it may not work for the next three.

Mike "Mish" Shedlock