Best Recession Forecaster: Robert F. Dieli

Since last May I have been reviewing the record of those who forecast the business cycle.  I developed a stringent list of requirements, "Jeff's Acid Test," and I frequently invited nominations.  Here were the stated requirements:

• Openness -- with the potential for peer review
• Small number of input variables.  Most people do not understand that "small is good."  If you have a lot of variables, it is easy to do back-fitting on a few cases.  Beware.
• Real-time performance.  This means that you do not go back in history doing any data-mining.  You create an indicator and live with it through time.

I received a number of suggestions in the comments and by email.  I very much appreciate the help from readers.  One result is that I am monitoring a number of new and promising forecasting methods.  I plan a later article on the honorable mention winners.

Somewhat to my surprise, there was only one candidate who met all three criteria:

Robert F. Dieli and Mr. Model

Most people think they know about recession forecasting, but they are often responding to someone's last good call or who has the best PR team.  The Dieli method hits a winning trifecta -- it is based on sound intellectual premises, it has worked better than any other method in real time, and it is open for our review.  What more can we ask?

Background

This is Part 2 of a planned five-part series on recession forecasting (Part 1 is here).  I know that many will want to dig into the nuts and bolts of the Dieli method, and I will start that in the next segment.  There will be plenty of opportunity to for comparison and discussion of various methods.

For now, let us just think about the results in terms of the long-term track record, and learn a little more about how the model was developed.

Track Records

Let's start with a look at Mr. Model.

The key variable is the Aggregate Spread, depicted by the blue line.  It depends upon monthly data, and will be updated next week.  The trigger point is the 200 level.  Whenever the blue line crosses 200, it is a forecast of a "cycle event."  The forecast horizon is pretty close to nine months.  This means that when the line crosses 200 moving lower, it is a nine-month warning of a peak, AKA recession as defined by the NBER. When the line crosses moving higher, it provides a nine-month warning of a trough.

Doug Short does an excellent job with the two most popular candidates in his update article, The Great Leading Indicator Smackdown.  There are several excellent charts, and I recommend reading the entire article.  For our current purposes, I am selecting the one that best matches the Mr. Model forecasts.

I invite the reader to scroll from left to right, looking at the lead times for both the onset and the end of recessions.  (I understand that the ECRI uses both an acceleration term and other indicators to augment their calls, but we have to start somewhere.)

It would be nice to put all of these indicators on a single chart, but I think the strength of Mr. Model is apparent.

The Man behind the Model

As background for the record, I have known Bob Dieli only for a few months.  Since we both reside in the Chicago suburbs, it was convenient for us to meet for lunch after a joint appearance on a panel.  I appreciated his openness, honesty, and intellectual rigor.  I was even more impressed by the results of his method.  When I learned that he had not tinkered with it over the years, I really perked up.

This is a very unusual combination, helping to define someone as the "real deal."  Here is the interview I later conducted.

Q:  Bob, tell us a little bit about how and when you first conceived the ideas behind Mr. Model?

The origins go all the way back to graduate school at the University of Texas in the 1970s, when I first became interested in the business cycle.  I began to work on the model in its current form while I was doing economic research at the Continental Bank from 1978 to 1984 and at the Northern Trust from 1987 to 1994.  Both were financial institutions with major exposure to the risks associated with business cycle peaks and troughs.

Q:  You have had a number of high-profile jobs.  Did your economic research on this topic continue through all of these experiences?

Over the course of my corporate career I spent time in staff assignments in economic research and later in line assignments in credit risk management at the Continental during the crisis that led to its implosion.  In 1987 I became a fixed-income portfolio manager at the Northern, where my clients were mostly high net worth individuals.

Can you tell us a little more about how these positions helped you develop your skill as a forecaster?

The combination of those experiences gave me some important insight on the preparation and use of forecasts.  I learned that details of great interest to the forecast originator may not be very important to the forecast user.  The bottom line was that accurate and insightful forecasts, delivered in a timely and usable manner, were the most sought after by decision makers.

Q:  You feature a chart of economic performance and model signals, with an excellent real-time record.  Can you elaborate on that a bit?

My objective was to find a format that allowed the user to draw conclusions quickly.  This turned out to be charts with long historical tails that provide perspective and context at a glance.

Q:  You also write extensively on employment.  This actually represents a second approach for your economic analysis, I think.  Is it giving you a similar signal right now?

The employment figures are among the most informative statistics we have.  They are the best coincident indicators of economic conditions and, as such, give you a great place to start on figuring out where we are in the business cycle and what is likely to happen next.

Q:  How much lead time do you usually see between a signal from Mr. Model and an economic peak or trough?

The Aggregate Spread, which is the principal forecast statistic, operates with a constant nine month forward look.  Unlike other leading indicators which operate with a variable lead time, the Aggregate Spreads looks ahead nine months at all times.  Think of it in the same terms as the beam on the radar on your local weather channel.  The historical record has shown that when the Aggregate Spread gets to 200 Basis Points, from either direction, we have reason to think there will be a cycle event (either a peak or a trough) some time in the time period nine months ahead of the arrival of the Aggregate Spread at the 200 Basis Point boundary.

Q: Making those calls out nine months must lead to some interesting conversations with your clients.

Indeed they do.  For example, the signal for the peak of the 2001 recession that began in March of that year, came from model readings obtained in June of 2000.  Telling folks, in the middle of the tech boom, that the business cycle had not been repealed and that we would have a recession the following year was a tough sell.  Similarly, the indications that the recession of 2007 would end in the middle of 2009 began to emerge late in 2008 and in early 2009.  Trying to tell folks that the economy would turn up while it was in the midst of what looked like a free fall in the first quarter of 2009 was even more difficult.  But, because the model has the track record that it does, by the end of the first quarter of 2009 most of my readers were convinced that the worst of the recession was over and that a bottom would be forming.

Q: To make this clear, while you talk about markets, you are not making market predictions.  You are predicting the economy, right?

That is correct.  What I am out to do is anticipate the dates of business cycle turning points as determined by the National Bureau of Economic Research (NBER) with enough warning to allow effective planning.  The stock market, the fixed-income market, and the housing market, to name just three, all have their own cycles.  Sometimes those cycles match up closely with the NBER turning points, and other times they don’t.  But you can’t know that until you know the NBER dates.  The Aggregate Spread has an excellent record of showing, as much as a year ahead, when an NBER event is likely to take place.  Armed with that information, and the specifics of their industry, or market, informed decision makers can make appropriate plans.

Q: Could you tell us a little be about your firm and its clients?

I’d be glad to.  RDLB was started in 2002.  My clients consist of three main groups: money managers, companies that make things, and individual investors.  I send my monthly reports, which are available by subscription on my website, to all of these groups. I am available to all my subscribers for additional interpretation of the report contents. I also have a consulting practice in which I function as their economic research department.  The assignments are as varied as the firms themselves, which makes the work very interesting to me.  I also do public speaking before trade groups and gatherings arranged by and for my clients. 

Q: Why do you call your site "Nospinforecast"?

Because the forecasts and viewpoints expressed there are completely data driven.  I talk about what is on the charts.  I don’t rant and I don’t take sides.  I report the information and provide complete access to my forecasting methods.  While I do talk about possible future outcomes, I do so within the context of numbers themselves.  My objective is to provide my readers with information they can use to assess other views and forecasts as well as information they can use to effectively manage their business and financial affairs.  One of the reasons I do this is because of lessons I learned while working with the trading desks at both the Continental and the Northern.  The same piece of economic information might be reason for one desk, say short-term fixed income, to buy and another desk, say foreign exchange, to sell the instruments they traded.

Q:  Thanks, Bob, for your helpful and informative comments.

Thank you for the chance to talk about Mr. Model.

Conclusion -- Part 2

There is plenty more to discuss.  I will get into the workings of Mr. Model in Part 3.  We will revisit the comparison with the ECRI in part 4.  These are tentatively on the agenda for next week.

Meanwhile, everyone should note that a "cycle event"  -- aka recession -- is not expected for at least nine months.  Unlike those who ascribe 0% or 100% chances of events, I understand that bad things can happen.

Nevertheless, you should keep this in mind:

In the 50-year history of Mr. Model, when the indicator is at current levels, there has NEVER been a recession within nine months.  In fact, we are not even close to the nine-month signal.

Comments

[oldprof]

pennypack -- As usual, my blog agenda exceeds my bandwidth!

If I were more worried about a recession, this would rank higher. Meanwhile, two of the installments are close to completion, but will include some outside content.

Pretty soon....

Thanks for noticing!

Jeff

[pennypack]

Jeff, where are #3 to #5 of your series on recession forecasting? i can't seem to find them...

[oldprof]

macrotool -- This model has been a subject of discussion among those I have cited. It is based upon data several months old and subject to revision. Even with that, it is not better than the sources I have cited.

Thanks for the suggestion, and I welcome any others. I have held this "competition" open for more than a year. Most nominees have models totally based on back-testing.

Jeff

[macrotool]

a bit late on the topic but have you looked at the Chauvet recession model. I follow it closely and pretty good track record. IF so, can you tell me what you think of it.

[John Butters]

Hi,

I noticed the discussion after my last comment. I commented on ECRI's long, short and inflation leading indices here: http://bit.ly/xdCBYA

ECRI may have made minor changes to the components but if you look in the right place you can get a good idea of what is in them. Having pulled them apart, I concluded that they weren't gold dust after all -- which isn't a surprise when you think about it -- and made my own index using the components of the OECD and Conference Board indices that still work (e.g. not M2).

[John]

No idea. You have defined ECRI black box as an orange and the Diele's Aggregate Spread as an apple so there's no where to go with that comparison. However, with their black box, whatever it is, ECRI has called a recession for the first half of 2012. That's an output from their black box. Diele's Aggregate Spread with their transparent apple, says "recession?" no way, certainly not within the next 9 months. Where does this discussion go if we in fact experience a recession by July 1, 2012? ...and is Dieli's apple red, transparent, or perhaps actually black also. The Aggregate Spread is an apple output. What else do we know about what is inside the apple? Do we know enough to replicate the Aggregate Spread. If we can, does it really matter if we experience a recession by July 1st, 2012. This whole discussion is a little like debating Anthropogenic Global Warming, we just don't need to wait as long to get a real feel for the answer.

[John]

...also after spending some time with Robert Dieli's Aggregate Spread, I'm not seeing a very consistent relationship between crossing the 200 basis point threshold and recession onset. Please step through each instance from left to right to show just how this works. Thanks. I don't know that ECRI's performance is any better, but we should know soon enough. ECRI maintains we're still looking at recession onset in the first half of 2012.

[oldprof]

John -- I am well aware of the ECRI contentions and the circumstances of their "non call" or a recession in 2010 and the sudden switch in September of 2011. Since they do not reveal any of their "world indicators" or "long leader indicators" everyone is left to wonder what is inside the black box.

When a group makes the assertion that they are the best at something, you can expect it to be studied carefully. I think that Doug Short provides the most objective data analysis and charts on this topic, and he has featured the WLI for years. Most other people look at it in the same way.

All I am doing is comparing the Dieli record with that held out to be the best.

So to be fair, do you have any other suggestion for a comparison? If Mr. Model is an apple, what apple would you propose for a test?

Jeff

[John]

Jeff,

Why are you showing the WLI as a recession indicator. ECRI makes it very clear that is very definitely not the purpose of the WLI. You are comparing apples and oranges. Read here: http://www.ritholtz.com/blog/2010/08/weekly-leading-index-still-widely-misunderstood/

[John Butters]

Thank you for your kind reply. I look forward to learning more in due course.

Best wishes,

John

[oldprof]

John -- Since a recession is not imminent, this is on my list as important but not urgent. Stuff like the employment preview and assorted news events have interfered.

I have an ongoing series on recession forecasting with two or three more installments. I am also two installments behind on Europe.

To summarize -- it is a very important subject that I want to do effectively. I'll get there.

Meanwhile, I hope you come back to read for other content as well. You can also subscribe to our email updates, or follow me on twitter. You won't miss the article:)

Thanks for your continuing interest.

Jeff

[John Butters]

Very interested to know how this is calculated -- I have been coming back here since this piece was published to find out! Are you going to give us the details?

[oldprof]

Eric -- There are two errors. The one you note is what Bob refers to as a "high turn" and the other was in the late 60's where economic weakness occurred, but not with enough magnitude for the NBER to call a recession.

I plan to continue the explanation of how I am using this indicator along with what else to watch. An important question is whether conditions are enough different to alter the probabilities at each level.

More soon.....and thanks for the question.

Jeff

[Eric Kennedy]

This looks like a good indicator, yet it is worth noting it never got down to 200 for the 1960 recession. Did Bob say why?

[oldprof]

Fundmyfund -- The 2005 initial signal came from some Katrina-related effects.

It would be easy to take the indicator and add a couple of variables to eliminate this. One of the things I like about Bob's approach is that he stays with the method. He does provide discussion and color in real time. More in the next installment.

Jeff

[oldprof]

Fundmyfund -- Moving above 200 after a time below indicates that a trough is expected. The NBER uses the trough to indicate the end of a recession.

The indicator is not trying to describe current conditions, but to anticipate peaks and troughs in the cycle.

I hope this helps.

Jeff

[Fundmyfund]

Also did Robert have any comments about ~2005-2006 which seems to be the one main false positive? THanks

[Fundmyfund]

Hi Jeff, interesting stuff. I can see clearly the recession calls (i.e. when the line breaks 200 to the downside) but not sure what it tells us when we go back over the 200 reading. Simply that we are not expanding again? Or will expand in 9 months? Thanks

[gm]

Thanks Mr. Jeff. I wish I can work for smart people like you and Mr. Dieli. Thanks to internet, ordinary folks like me have access to best in the business through mouse click.

Regards.

[Proteus]

Jeff: Congratulations on having this as a featured story on Business Insider. You've hit the big time :)

It's a well written and persuasive post.