David R. Henderson  

The Best Question I Heard Yesterday

PRINT
Incentives Matter: Prison Rape... Unchecked and Unbalanced

I sat in on Jeff Hummel's graduate course in Monetary Theory at San Jose State University last night. In the first part of the lecture, Jeff gave a quick tour through macroeconomics: the various schools of thought plus what various schools had said about the Great Depression. He said that the G.D. was sui generis, a statistical outlier. One of the students, Sherwin de la Cruz [he gave me permission to use his name], asked:

If you regard the Great Depression as a statistical outlier, what's the point in studying it?

Comments and Sharing


CATEGORIES: Macroeconomics



COMMENTS (17 to date)
david writes:

You can't just leave us hanging like that! :) What was Jeff Hummel's reply?

fundamentalist writes:

In statistical process control, the outliers are the data points we need to study the most. Data that stays within 3 standard deviations of the long run mean are considered random. Anything that jumps the 3 sd line is a special cause and needs to be investigated.

I hope finance people aren't ignoring it because it is an outlier.

David R. Henderson writes:

david writes:
You can't just leave us hanging like that! :) What was Jeff Hummel's reply?

I found Sherwin's question so profound that I wrote it down right away. I figured it was a neat question no matter what the response. But, if I recall correctly, Jeff responded that it was a good question and that the reason to study even a statistical outlier is to understand what led to it so that we are less likely to do it again. That, at least, is what I recall.

Best,
David

John Thacker writes:
I hope finance people aren't ignoring it because it is an outlier.

As Arnold Kling mentioned before, the popular finance valuation of "Value at Risk" (VaR) specifically excludes outliers when determining how much money is at risk. It only considers events up to the 90th percentile loss.

Mark Bahner writes:

"In statistical process control, the outliers are the data points we need to study the most."

Yes, exactly. The "signal-to-noise" ratio is much higher for the Depression than for a mild recession.

David R. Henderson writes:

John Thacker and Mark Bahner,
Good answers to a good question.
David

mark writes:

Correct answer: to make sure it stays that way.

Barkley Rosser writes:

A good question and an absolutely pathetic answer.

The real issue is "what is the underlying probability distribution"? Large amounts of macro theory that now looks utterly stupid has assumed Gaussian normal distributions, on which these sorts of remarks about numbers of standard deviations are based. In such a distribution, all you need to know is the mean and the standard deviation.

However, all financial returns series and many others, exhibit "fat tails," or more formally leptokurtosis, positive fourth moments (the variance is the second moment). This means that such "extreme events" are more frequent than would be the case if the distribution was normal or Guassian. Such matters are discussed quite eloquently in the book by Nassim Taleb, The Black Swan, who critiques sharply the sorts of idiotic VAR analyses that contributed to so much to the recent crashes.

I hope that Hummel addressed this. If not, he should not be teaching.

E. Barandiaran writes:

Barkley Rosser,
As Herbert Gintis has said in many of his hundreds of book reviews, many critiques of economics are well justified but as long as they are not followed by new ideas on how to overcome its deficiencies they will hardly make a difference in the analysis of particular issues. I started to read The Black Swan but I stopped at the introduction because Taleb acknowledged that the book was only a critique. If I remember correctly Taleb said that he was going to look for new ideas. Do you know how much Taleb has advanced in the past three years?

Justin Rietz writes:

(disclaimer: I am in Hummel's class)

I believe Prof. Hummel said it was an outlier in regards to the 19th and 20th century. I don't believe he was thinking of any specific statistical distribution, just that the GD was a magnitude larger than any other depression we have experienced in recent history (unemployment at 25% and a 30% drop in production).

Of course, if we get into fat tails vs. Gaussian distributions, localized recessions in specific countries, etc., it may not look like such an outlier.

John Thacker writes:

Properly used, VaR isn't an absolutely terrible idea. I'm just not convinced at all that human tendencies allow it to be properly used. It focuses all the attention on the common risks and none on the tails. It's very dangerous when used as a sole metric of risk.

Another way of putting it is that using VaR alone invites people to fall for the martingale betting strategy. (If I lose, double my bet. Keep doing so until I win.) The martingale betting strategy has been invented time and again in different forms by apparently smart people (such as LTCM). People have to be discouraged from employing it, but using VaR as a primary metric encourages people to shove risks into tail event. The martingale betting strategy is zero risk from a VaR perspective, if you make enough bets.

Barkley Rosser writes:

David Henderson has asked me offlist to apologize to Hummel. I would say that my remark about him "not teaching" should be viewed as over-the-top rhetoric. The hard fact is that very few macro teachers talk about such matters, so he is probably not any worse than most, and may well be better. For the record, I most certainly do not think he should be removed from his job.

That said, I do not retract any of the rest of it. The substance of the complaint stands, and in this regard my strong words were really directed at the great mass of macro teaching and its apparent reversion to form after we have just had another so-called "outlier" event. It really was not directed at him personally, and I gather from Henderson's enthusiasm that he is probably a very good lecturer in general.

Justin Rietz,

It is certainly true that whatever one calls it or whatever statistical distribution one assumes, the GD was the most extreme, downward macro shock we have seen so far since the Industrial Revolution and the emergence of "modern" business cycles. In that regard, it is reasonable to call it an "outlier."

I would also say that his answer as reported by Henderson was not unreasonable in the sense that just because it was an outlier does not mean it should not be studied, and indeed should still be studied so as to avoid such things happening again.

The whole discussion of "fat tails" is to emphasize that it may not be all that much of an outlier. As John Thacker notes, ignoring outliers can be very stupid, even deadly in some circumstances. I am regularly appalled by hearing of researchers tossing out "outlier" data points from their samples, which are sometimes the most important ones of all, although such a practice is justified if the outlier is such because there is a problem with the data (and the datum in question thus may simply be inaccurate).

I wish Professor Hummel the best and it may be that he simply does not address the issue of the nature of the statistical distributions, although, as I said, the models still being used and almost worshipped in the basements of most of the world's central banks all assume Gaussian normal distributions, along with the even sillier assumption of rational expectations (which itself is formulated in terms of a Gaussian normal).

Barandian,

Taleb pushes several different lines, from "nothing can be done" to the use of various ad hoc techniques. One of those, used by a lot of investment operations, is to redo all the standard stuff using the mode as the base rather than the mean. This helps a bit if one is dealing with a badly skewed distribution, especially one like a Levy for which there is neither a finite mean nor a finite variance, but that is strictly ad hoc. I do not think broader macro outcomes are as far off from the Gaussian as one gets with a Levy distribution or some of its other ugly cousins.

David, you will note that I have clarified my position, but that I have not apologized per se, as I do not think that is called for. I think you were overreacting. Hey, this is a blog, and on this one I see people calling each other all kinds of nasty names from time to time without ever apologizing (it has certainly happened to me on more than one occasion, and I have not been on the receiving end of many apologies here).

David R. Henderson writes:

Dear Barkley,
You're right that it's not an apology, but you did do something approaching what I asked for. Thanks. I appreciate it. I think I sometimes take attacks on my friends with more difficulty than I take attacks on myself, although the latter are sometimes hard too. :-) In the case of Jeff, I think that if you saw how methodically he prepares for class, how well he pulls it off, and how even-handed this radical anarchist is at dealing with the various schools of thought, you would appreciate my upset more.
For the record, though, if you are unfairly attacked, Barkley, I think you are owed an apology. As is anyone who is attacked unfairly.
Best,
David

Barkley Rosser writes:

David,

Fair enough. I am sure Hummel is a good guy and an excellent lecturer. Again, I was reacting more to widely spread views and practices than to what he said in particular.

I also sympathize in that I tend to be much more vociferous about defending the honor of others. I am used to being attacked, and, frankly, do not give much of a damn most of the time. I tend to view people calling me names as having lost the argument and having nothing useful to say.

Sean writes:

Just for the record, the concept of an outlier is not predicated on a Gaussian curve. Every type of distribution may have outliers present.

The true failing is assuming outliers away for mathematical convenience.

I read that the hole in the ozone was present in the data for many years, but it wasn't "discovered" until they stopped deleting the outliers.

Then again in business, I recommend that everyone delete and ignore their outliers since I will get a competitive advantage by studying and learning from them.

Of course your mileage may vary,

Hi, Barkley.

You wrote in two of your preceding comments:

Hey, this is a blog, and on this one I see people calling each other all kinds of nasty names from time to time without ever apologizing (it has certainly happened to me on more than one occasion, and I have not been on the receiving end of many apologies here).

and
I am used to being attacked, and, frankly, do not give much of a damn most of the time. I tend to view people calling me names as having lost the argument and having nothing useful to say.

I have checked my records for the last two years and I can find no instances at all of your having been on the receiving end of any name-calling on EconLog. EconLog has a very strictly enforced policy of no name-calling or ad hominem attacks. Your claim that you "see people calling each other all kinds of nasty names" on this blog is not accurate. I'm sure an instance or two of name-calling gets by us in the middle of the night now and then temporarily, but most people who engage in name-calling are banned and their comments removed immediately. I've privately offered you my personal email address, in case you feel you have been personally attacked.

Even if you are inured to being attacked or called names on other blogs, please do not propagate your experience to EconLog. It's your own personal matter if you don't give a damn if you are attacked elsewhere, but we do give a damn if people attack each other here.

If you have been attacked personally on EconLog, we'll be very happy to apologize and to address the matter.

Mark Bahner writes:

Oh, c'mon...even one http triggers a spam block?! Well, out with the http then...

A good question and an absolutely pathetic answer.

The real issue is 'what is the underlying probability distribution'?

But the "underlying probability distribution" is affected by the actions taken during the Depression.

When I had my courses in introductory economics and my one course in Money and Banking in college, I was told that the Fed never messes with the required reserve ratio because it's "too powerful." It was described as "the atomic bomb of Fed policy."

So I was extremely surprised to learn that the Fed actually RAISED the required reserve ratio in 1936/1937, right in the middle of the Depression (which has a tremendously restricting effect on the money supply):

journals.cambridge.org/action/displayAbstract;jsessionid=F1E47414A772E0AE0AF60462A4566457.tomcat1?fromPage=online&aid=449537

We need to study the Depression so that all of the incredibly stupid things done during the Depression can be avoided in the future.

P.S. As an engineer, if I'm taking an economics course, and the professor starts explaining the Great Depression by, "...all financial returns series and many others, exhibit 'fat tails,' or more formally leptokurtosis, positive fourth moments (the variance is the second moment)"...

...I start checking to see whether the deadline has passed for dropping a course. ;-)

[Mark: The issue with the initial holdup wasn't the number of urls. It's that the particular website--journals.cambridge.org--is unfortunately listed by a spam-blocking filter. Legitimate, high quality websites get reported as spammers sometimes. I usually quickly notice and okay those comments. Perhaps you might be patient for a few minutes.--Econlib Ed.]

Comments for this entry have been closed
Return to top