Is It Time To Drag Out The Recession Forecasting Models?

by: Robert Brusca


Here's why so many recession forecasting models are useless.

What actually has predictive value?

I discuss how to connect with processes that have predictive value.

The unexpected is happening. Will more of the unexpected happen? Brexit was not supposed to happen. We are told that while recession probability is higher, it is still very low. Well, so what? Have you looked at any of the recession-predicting models recently?

This is the NY Fed (FRBNY) index using the yield curve in a recession prediction mode (here). The chart shows several features common to all recession prediction models that I would like to shine a bright light on. When we chart these indices vs. business cycle bars they look 'pretty good' on the surface. But a closer look is warranted if we plan to use them for anything at all.

First, this particular index rarely reaches a probability value above 0.6. This makes me wonder what the real signal is. Several rules are suggested, but that is not the point.

Second, even those recessions where the probability is so high we don't need to invoke a rule do not give that signal until very late in the cycle in many cases. Does better late than never work?

Third, if we were to form a rule apart from the raw probability statement, there would still be some significant signals that would start out looking like recessions but the would fail to trigger the rule. False positives take away from usefulness of these models.

Fourth, the model gives inconsistent lead times. Some lead times are too long to be useful. (if you saw 'The Big Short' you got a dose of how painful it can to be right but to be 'righter than the market' or too-early.) 'Too early' can turn out to be another form of 'wrong.'

Fifth, some lead times are too short to be useful. Yogi Berra said, "…it is tough making predictions, especially about the future." But some of these predictions have such short lead times that by the time they emit 'true signals' they aren't even about the future anymore transforming them into a class of 'forecasting devices' known as 'now-casts'. When a forecast is contemporaneous with available data it is behind in real time since all data (except market data like from the yield curve) lag. If it takes the signal several months to develop we become economic historians, not forecasters.

There is another probability of recession model from the St Louis Fed that hits higher probability levels than the FRBNY model and has tight, crisp lags, but it still has two of the problems listed above (1) no real lead time and (2) the signal goes from zero-to-sixty way too fast (i.e. no warning for the warning). Like an air-raid horn that goes off as the bomb explodes.

The applicable point for TODAY is that while some want us to be 'reassured' about the current low probability of recession, if recession can spring up suddenly with no advance notice how useful is a gauge that tells us that the probability of recession is low? Surprise! Recession!

The NBER and recession-dating

Basically all recession prediction models have this problem. The NBER, the official recession dating institution, usually does not make the call on recession until we are six months into it (and that is done by convening a meeting of the recession cycle dating committee which uses human judgment). Historically the committee has taken from 6 to 21 months to call a downturn a recession. A model that predicts recession with greater timeliness than that may have some usefulness. But don't kid yourself into thinking that there is a secret key out there that someone will find. Or that actually calling it a recession late is going to help your investment performance all that much since, by late in the cycle, much of the damage will have been done. The NBER dating committee tells us (here) that a business cycle downturn needs to have three characteristics to be a recession (1) the disruption must be deep enough, (2) last long enough and (3) affect a broad enough swath of the economy for it to be deemed a recession. You can imagine how hard it is to anticipate all three metrics in advance.

Shaving with Occam's razor

The Federal Reserve Bank of Philadelphia (FRBP) has one of the longest economic surveys on the manufacturing sector. It reports a series of diffusion indicators (like the ISM, only using an up minus down presentation instead of the ISM model). In the ISM model 'neutral' is at a value of '50' in the Philadelphia model neutral is at 0. But that is just a matter of presentation not substance.

Amazingly the Philadelphia survey which also has a survey module on expectations for six-months ahead can, with application of simple filter diagnoses every single recession since 1969 (save 1982 which is a special case) and does it with an early and reliable signal. I know what you are thinking this is just too good to be true. But it isn't. It is true. And it is simple.

Predictive results from FRBP survey…drum roll please

The FRBP Future survey evaluated with a filter that the signal for recession is any negative future index value smaller than -6.4. This produces the following results: 2007 recession (signal lags recession start by 2-Mo), 2001 recession (signal leads by 2-Mo), 1990 recession (signal leads by 13-Mo), 1981 recession (No Signal), 1980 recession signal (leads by 13-Mo) 1973 recession, (signal leads by 3-Mo) and 1969 recession (signal leads by 3-Mo). And since the survey is released topically (the June survey is released late in June) there is very little data lag.

The Philadelphia MFG Future index misses the 1981 recession partly because the 1981 downturn comes right on the heels of the 1980 recession. There are only 11-months between the end of the 1980 recession and the start of the 1981 recession. In fact the Philadelphia MFG future index did not post a decline of any magnitude in the 1982 recession as firms were optimistic all the way through the recession. However, the Future index could be supplemented with a rule about weakness in the current index as every salient macro indicator was collapsing in 1982. In any event this is the only real flaw in the Philadelphia story which is so simple and all the more impressive because of that simplicity.

Why this works: I suspect that the Philadelphia signal works because the region is in the center of the country and is well plugged into the economy at large. The filter (X<-6.4) serves to make sure that manufacturing is expected to fall significantly and posits a significant enough drop in manufacturing to assure that will affect the economy deeply and broadly.

Other signaling devices: Occam's razor- available by mail order as well as internet

I will explain this measure after we look at its properties. The performance of this times series (which you note is inverted) shows that it has a very consistent 1, 2 3 wave pattern early in the cycle, this is a most remarkable consistency. It also serves to keep the 'one' wave from being seeing as a signal and the 'two' wave from looking like the start of an early recession. This index peaks late in the business cycle and recession comes up to 22-months after it has reached its cycle low point (high point as plotted). The median time to recession is 14.4 months; the average is 13 months. The range is from 9 to 22 months. The signal is always leading. The power of this index comes from its amazing regularity. Of course, in this cycle there is no 1, 2, 3 wave just an inexorable rise to its lowest value since the early 1970s. The index is off peak for only one month in a row so far. So the recession clock is ticking. The signal for this index is less precise than for the Philadelphia index but it does put you on notice nice and early which is quite helpful. It gives you plenty of time to look for other conforming signs of recession and to prepare your investment strategy UNLIKE the other statistical methods talked of earlier where the signal for recession all but pops out of the head of Zeus.

Now for the kicker: The series I have plotted inversely to create this signal is jobless claims…simple jobless claims. Once again let me warn you not to cut yourself on Occam's razor.

While many economists have been talking of how strong the economy is and how unlikely a recession is (in part) because 'jobless claims are so low' the exact opposite is true if history is any guide. Lord, forgive them they know not what they do. Economists seem unable sometimes to think outside the box especially if it's a black-box. (Caveat emptor: I have a PhD in economics.)

Epilogue to a research note

One beef I have about modern economics is how economists seem to love to make something more complex than it needs to be. There are many examples of this that I hope to be writing about in the coming months. But for now this is my pet peeve. Jobless claims are not a measure of economic strength to stay. They are ephemeral. When they get this low they become a liability, like a loaded spring in a mouse trap (caveat cheese-eater). The unraveling will be palpable when it comes. Non-farm jobs already are off peak. The global economy is under pressure. Do you doubt that claims have seen their cycle low? It's possible that they have not, but it's not a big deal. If they have not we will simply reset the clock. But for now it's a good bet that the low is behind us. I am just reporting here some statistical facts and 'righting' some misperceived platitudes. You decide if they work for you or not. What I find most impressive is that they are so darn simple. And the arguments about how or what works are right here out in the open, not buried is some statistical tangle that only three econometricians in the world claim to understand. These not based on misleading derivative ideas. The clock is ticking. Even the three blind mice can hear it. It's an equal opportunity signal.

Disclosure: I/we have no positions in any stocks mentioned, and no plans to initiate any positions within the next 72 hours.

I wrote this article myself, and it expresses my own opinions. I am not receiving compensation for it (other than from Seeking Alpha). I have no business relationship with any company whose stock is mentioned in this article.

About this article:

Author payment: $35 + $0.01/page view. Authors of PRO articles receive a minimum guaranteed payment of $150-500.
Want to share your opinion on this article? Add a comment.
Disagree with this article? .
To report a factual error in this article, click here