Business Cycle Forecasting: The First-Rate Results of Robert F. Dieli

This article is probably the most exhaustive and challenging piece I have written.  It was worth the effort because understanding the business cycle is crucial to making great investment decisions.  To get the full benefit, I urge readers to spend some time reading the background links and watching the videos.

I am going to follow up with another piece describing how I use this information for investment decisions.  For now, let us all focus on the method, understanding how and why it has worked so well throughout history.

Background

In May of 2011 I embarked on a search for the best recession forecasting methods.  I had been a long-time fan of the ECRI approach.  They were still very positive on the economy at the time, and my quest was not driven by their conclusions.  I was uncomfortable with the methodology and the lack of transparency.  I had many reader suggestions, and I reviewed them all.  The criteria were stringent -- "Jeff's Acid Test."  The easy winner of this competition was Robert F. Dieli's "Mr. Model."  (This article described the competition and the results).

The main conclusion from Bob's work was that there was no imminent recession.  This ran counter to some other well-publicized and popular forecasts.  Some readers complained in the comments that the history of the forecast included some imperfections.  Others disagreed with the methods.  The subject was too difficult for simple responses to these questions.  I promised to follow up in more detail, but I wanted to do so in a convincing fashion.

A Year Later

A year later, some key elements of my rationale should be even more convincing:

1. Bob was right -- once again, as so many times before.  And he did it in real time, not on a back-tested basis.
2. Imperfections in real-time forecasting are acceptable -- even desirable.  When I see a perfect forecast, it always means that the model has been tweaked and changed to fit all of the past data.
3. Simple is good.  Methods that over-specify the number of variables and numerical trigger points also imply excessive back-fitting and poor predictability.
4. Theory is important.  The model should make sense.

Most recession forecasting models fail because they emphasize weakness.  This is backwards.  A recession begins at a business cycle peak, something that I explain more carefully here.  A recession starts with excessive strength.  Seen any of that lately?

Dr. Dieli explains this quite clearly in this chart.

 

Your intuition about the business cycle would be better if you completely forgot the "R" word and took Bob's lead:  Substitute "business cycle peak." 

The key driver of Bob's forecast is what he calls the "Aggregate Spread."  By reviewing results over decades we can see that this method actually provides a warning of about nine months.  The image below describes the composition of the spread, using example data from August.

The most recent aggregate spread is shown below. Just as it did last year, it provides strong evidence that the US economy is not nearing a recession.

And Now -- The Show

Get some popcorn and your favorite beverage and settle back to watch the show.  I recently met with Bob Dieli to discuss economic forecasting and to create some videos.  The result is an eight-part series in which we discuss each of the recessions of the latter 20th Century.  [Thanks to Derek Miller for helping in the production of the videos and producing the key summaries.]

In this first video, Bob and I discuss National Bureau of Economic Research and why their definitions of a recession are important. The nonpartisan NBER looks at both the peaks and troughs of the business cycle to conclude when past recessions have happened, effectively making "autopsies, rather than forecasts" - as Bob says. Therefore, it is important for the Mr. Model to use the same criteria when it forecasts for recessions, providing a clearer picture than other models.

 

 

 

In part two, Bob and I take a close look at the recession of 1957. In doing so, they describe exactly how Mr. Model works. The model signals 9 months ahead that the business cycle will be heading towards a peak or trough when it crosses the 200 basis points (shown as a red line on the chart). 

 

 

 

Bob and I illustrate the ways in which policymakers can and do impact the business cycle and how this interacts with Mr. Model. In the run up to 1960, tightening by the Federal Reserve as well as fiscal cuts by the Eisenhower Administration led to an economic downturn. In 1967, when the Fed again tightened the yield curve, the model signalled a recession. Shortly thereafter the Fed eased up, thereby avoiding a recession. At the end of the day, the NBER never called a recession in '67.

 

 

 

Mr. Model had nearly spotless performance in predicting the recessions of the 1970's. Contrary to popular belief, the 1973 recession had less to do with OPEC and more to do with other government policies that laid the foundation for an economic downturn. 

 

 

 

Mr. Model shows the result of Fed Chairman Volker's monetary policy, which inverted the yield curve and brought the Fed funds rate to 20%. The result was a short 6-month recession, then a short recovery which was stifled by other policies. Interestingly enough, the recovery never took Mr. Model past 200 basis points - meaning a new peak could not have been established for the "second" recession.

 

 

 

After the "double dip" recession of the 80's, the recovery brought the business cycle to record highs. This led to the third-longest period of economic expansion into the summer of 1990. A combination of tightening monetary policy and changing policies regarding the first war in Iraq were both responsible in part for the downturn. In 2000, Mr. Model signaled a recession in an election year - something that was sure to happen regardless of who was elected. However, in both instances the model predicted short and shallow recessions unlike the seriousness of the early 80s. 

 

 

 

In the most recent recession, Mr. Model's results were decidedly different than they had been for any previous recession. The model alerted that the 200 basis point line had been crossed in 2007 but did not decline sharply. This is in part because tightening by the Fed did not effect the yield curve as they had in past events. Quick reactions by the Bush and Obama administrations also helped to prevent a dramatic decline in Mr. Model's basis points.

 

 

 

In this final video, Bob and I focus heavily on the 2007-2009 recession. The model appears to show a false positive as it crosses the 200 basis point line in 2006, but continues sideways for some time before the recession was officially called. In a sense, this suggests severe instability rather than the dramatic declines of the past. In any case, we had ample warning that a recession was coming. It did not take us by surprise.

 

 

 

Conclusion

If you have studied the evidence, you will see that recessions usually involve the Fed!

You might also have noticed that business cycle peaks do not typically come from a problem of "stall speed" but one of excess stimulation.

Market observers are completely mistaken:

• Wrong indicators;
• Wrong interpretation (weakeness versus strength);
• Wrong sources;
• Wrong point of the business cycle; and finally
• Wrong stocks.

These will be the subjects of the next installment.

Comments

[Samuel Crowe]

Thank you both for the model and explanation…very well done.

Would there now be a concern that the fed funds rate is so low that it plays no role in the model? And that a recession warning would likely initiate long before the fed funds would have a mathematical significance?

If inflation is an expansion of the money supply, and price inflation the symptom, would not the economy’s absorption rate (monetary expansion minus excess reserves) give heads up, since money supply contraction is often the coup-de-gras?

I would think, to get an honest state of the economy, the true rate of unemployment would be the number of unemployed looking for work plus the number employed by government or any job where any level of government funding is required to sustain that position. It appears that the proportion government jobs or government credited jobs have become a larger number in the employment statistics, where government employment really shouldn’t count at all.

Just some thoughts…but regardless I plan on keeping track of the model, and again thank you both for the work and the blog.

Samuel

[Ken MacNeal]

2008 was a watershed where old economics stopped working and "Balance Sheet" economics started. In a world that became overly indebted now we have deleveraging as a key driver. That is why even zero interest rates do not have people borrowing. The level of interest rates both long and short are totally manipulated and distorted. It used to be that the Fed would manipulate short rates but long rates were market driven. That is not the case any longer. Unfortunately this also makes models that depend on rates not work as they once did or at the very least one could say they are now suspect. I use the OECD CLI rather that the Conference Board CLI because it has less emphasis on interest rate fluctuations.

It would be great to see a chart of each side of the model separately to see how they forecasted separately.

Thanks for the thoughtful work.

[wkevinw]

Just a big thanks! Outstanding work, as usual.

Kevin

[oldprof]

Hi Bob,

I invited Bob Dieli to respond, and here is what he said:

"The model establishes no equivalence between the levels or rates of change of the variables. The model does note that when the variables are combined in the way described on the table the result informs us as the chances of us seeing a cycle event in the forecast period (the next nine months). Whether that is a peak or a trough depends on where we are in the cycle when the observation is noted.

The normal condition of the economy is for the inflation rate to be less than the unemployment rate. This is why the Real Spread is negative just about any time the economy is not in recession. The difference between the two series is telling us something about how well the real side of the economy is working. The "meaningful" part of the operation is the sign of the answer. The size does not really matter that much, as experience has shown that the economy is as prone to recession when the real spread is only slightly positive (as it has been around most cycle peaks) and when is extremely positive, as it was prior to the 1973 and 1981 recessions.

Please drop me a line directly, via the contact page on my web site www.nospinforecast.com and I will send you some additional charts."

Thanks for the interesting question.

Jeff

[Bob]

Great stuff as usual Prof. But there is one point that just is not clear yet especially to a layman such as myself. How does the model justify a treatment in which an increase of say 100 basis points in unemployment is precisely equivalent to a 100 BP decrease in inflation? Or to put it another way how can we subtract unemployment from inflation and have a meaningful number as a result? Is there a hidden rate factor X that just happens to have the value of 1 in units of (percent unemployment/percent inflation)?

[oldprof]

RB -- Good eye! The government did not have the "long bond" during one stretch, so the long yield was sometimes the 20-year and sometimes the 30-year. It is not significant for the results.

Jeff

[RB]

Jeff- another question. Did Bob Dieli always use the 20 year bond for this spread measure? I don't see monthly data for the late 80s/early 90s period.

[RB]

Very nice job. What is striking is that this is not a mechanical model, but one that involves judgement. The other indicator that you discuss in your weekly reports, the Superindex, strikes me as having a more mechanical approach, combining various recession indicators. the superindex also had smoother transitions for the 2008/2009 recession. It also seems to involve some fitting for determining trigger thresholds based on past recessions. I wonder if you have any thoughts on comparing the two approaches.

Mr. AggregateSpread makes me feel better though about sitting out of the market from mid-2006 until 2009. It wasn't easy watching the market go up in 2007.

[Joe]

Jeff: thanks for your very helpful blog, a real eye-opener and refreshingly absent of ideology.
My comment for this recession-forecasting model post is that it seems to be helpful for the timing, but not for the severity of a recession. The recession(s) 1980-82 appears much much worse than the recent Great Recession, which appears to be just as light if a bit longer than the 1990 and 2001 recession. But in reality it was the worst of them all.

[Proteus]

Great post, Jeff. Very well written and clearly a lot of thought and effort was put into this.

And congratulations, you're hit the big time - this story is on the home page of Business Insider.

[SI]

Good segments, but would have preferred a consolidated 20 min discussion, notwithstanding the universal tendency to pander to short attention spans.

[wei]

A much simpler indicator is to just use the yield curve. If it is inverted, then a recession is coming. If it's steep, then a recession is far away. It's MORE accurate and MORE simple. and you don't need to pick an arbitrary level of 200bps to indicate recessions, which in my opinion is just massaging the data to fit the conclusion you're looking for.

But then again, we're a victim of selection and survivership bias. we're just selecting the data series that proves out a given pattern. Using this methodology, you can find "forecasting" tools for almost any pattern. That is why forward looking analysis is important. For example, the Fed can keep us in expansion phase by just keeping the short rate at zero forever! but we all know that this doesn't work for a lot of reasons.

The fed has gotten very aggressive at keeping the yield curve steep, so i suspect that the "inverted yield curve" indicator alone is going to be a poor recession indicator. You can already see this in the last two recessions, where the contribution from the yield curve component of the Aggregate Index is less pronounced.