LINQCRED: A Hedge Fund Manager
Yesterday we revealed our disciplined approach to interpreting information. It is actually quite simple, but we already have some complaints. Too complicated. Too many steps.
Investors and traders who want to gain a real edge must be willing to do a little work. This is a good test. If you are not willing to employ a careful process of critical thinking when evaluating information, you should just buy an index fund and move on. Those who are willing to do the work can beat the market by several points each year.
Learning by Example
The best way to learn — and really learn right — is to study examples until you get them all right. Earnings season is a great time to do this. There is plenty of anecdotal evidence and interpretation. We read and analyze them all, but this series of articles will try to provide some illustrations from varying perspectives.
Background: Observers and Data Analysis
One of the most important aspects in interpreting information is understanding the expertise of the observer. At "A Dash" we like to look outside the investment world to help readers see the point. Then we come back to an investment application.
The Numbers Guy has a great article on streaks in baseball. This is a worthy topic for another day, but we were struck by the following comment on one of the most astute baseball analysts, Tim McCarver:
Streakiness and momentum aren’t the only tenets of conventional
baseball wisdom to collide with math. Fox analyst Tim McCarver
expressed surprise during the Red Sox-Indians series upon learning that
multirun innings are more likely when the leadoff batter hits a home
run than when he walks. His assumption was that a runner on base
affects the pitcher and batters psychologically, leading to a rally and
multiple runs, while a leadoff home run makes everyone start with a
clean slate. But the numbers show that it’s easier to get one more run
with the bases empty, than getting two runs when starting with a runner
on first base. Mike Kellermann, a Harvard graduate student, rounded up
the reaction from blogs and showed that historical numbers side with
the homer. Meanwhile, sports-statistics company Stats Inc. told me that
in the 2007 regular season leadoff home runs led to multirun innings
28.4% of the time, compared to 27.2% of the time for leadoff walks.
The point: Impressions, even by the best experts, are inferior to analysis of the data.
Example: A Good Hedge Fund Manager
Erin Burnett, who is asking opinions about recession of anyone in front of the microphone, interviewed a good hedge fund manager in one of CNBC’s new segments yesterday morning. There is no point in naming the manager, a very sharp and engaging man who embarked on a trading career right out of college, and who now has his own fund. We know many people just like him, and believe his perception and approach to be quite typical.
The key question and answer were as follows:
EB: You can sit here
and argue persuasively for a recession or against a recession —-RS: A lot of the
evidence that argues against the recession is backward-looking and a lot of the
evidence that argues in favor of it is forward looking. Until we get more data it’s going to be hard
for this to sort itself out. …Friday I
was particularly troubled by a note from JP Morgan … which downgraded Ford and GM credit
because of rising delinquencies in the prime auto market. As you start to see evidence that this is
expanding beyond the sub-prime homebuilders you have to become more and more in
the camp of recession. I’m not
optimistic that years and years of financial excess can be wiped away with a
couple of small funds and a couple of moves here and there…..we are headed
toward more difficulties.
Implementing the Method
After reading the information carefully, we readily see that it involves interpretation and analysis of information disseminated in the market last week. The novelty, step "N" in the method (LINQCRED), depends upon the interpretation of the data, not the data itself.
Turning to Qualifications, we have no idea about the success of this manager in making global macro calls of this sort. He is someone who got an interview on TV.
Turning to Competence is a crucial point. The interview subject seems to have begun with a viewpoint about the markets and market history. How does one know the extent of "financial excess" or how long it might take for this to be resolved. We have had a period of below trend growth in GDP, intended by the Fed? How much will be enough?
Most importantly, has this observer really studied what types of data are backward-looking and what data provide a leading indicator? Do rating agencies look forward or do they look back on past results? [This is rhetorical question. Rating agencies are obviously deciding on recent performance. That is past data. Is there evidence that it is predictive?]
Conclusion
This observer in our example has a viewpoint and reads the news carefully. In a known and planned period of reduction in economic growth, to fight inflation expectations, we see such data every day. If one begins the day by looking for it, it will be there.
Our example observer spends his day trading and running his fund. He has not done an analysis to determine which indicators lead and which do not. It is an opinion, and one that lacks face validity.
Compare this approach — typical among traders and hedge fund managers – with those who specialize in taking data from all sources to make forecasts — like the ECRI. There is a difference between those who study indicators to find those that lead, and observers who begin with an opinion.
Following the LINQCRED method carefully should warn the investor about embracing this conclusion — even if CNBC decided to feature the comments.
“But the numbers show that it’s easier to get one more run with the bases empty, than getting two runs when starting with a runner on first base. ”
I think that both he and you are confusing causation with correlation!
Multirun innings probably occur more frequently in innings that start with leadoff homeruns because the quality of pitching relative to hitting is poorer, on average, in those games. The causal factor is most likely this mismatch of talent, and NOT the situation.
Bill –
Actually, I did not offer an opinion about causation. I want people to understand that unsystematic observations can be very misleading, so I am trying to stick to that point. My own interpretation is that it does not make much difference, which I find surprising in the opposite direction from McCarver.
McCarver has a causal model. I am not sure about the bloggers cited by the Numbers Guy. Your suggestion for a statistical control could be tested easily enough. I think there is some sabermetric research suggesting that good pitchers have more control over walks than they do over home runs.
Thanks for an interesting comment. I wish I had more time to do baseball research!
Jeff
The last Fantasy Baseball league I participated in had a category for WHIP – (Walks + Hits) divided by Innings Pitched. Combined with SO per Inning, you can learn a lot about pitching performance.
That said, I’m not a fan, I just participated for the love of modeling. I’m a geek.
I would be interested in the denominators for the 28.4% and 27.2% percentages, and a two-tailed hypothesis test that they were different (and at what confidence level).
Bill – My guess is that the number of cases used is large enough go establish “statistical significance” if one wanted to do that. As to substantive significance – I just think that McCarver expected a big difference and one percent or so either way is not what he thought.
Meanwhile, there are some sabermetric studies that show that once a ball is put in play, the skill of the pitcher has nothing to do with whether or not it is a hit. Pitchers are good because of high strikeouts and low walks, creating a dominance ratio. It would be fun to see whether your suggestion about overall pitcher skill affected this. We would need a good measure for “skill.”
Jeff
I admit I’m rusty here, but you see about 44,000 innings pitched in a season (30 teams, 2 per game, 162 games, 9 innings per team). Taking a SWAG at how many start with either a leadoff tater or walk, let’s say one out of every four IP, maybe a fan can step in with a stat or reality check of that assumption. Assume further that each situation is equally likely, and the average for both is 27.8%. Regardless of the number of IP with either situation, if my hypothesis is that they are equally likely to create multi-run innings given that they have already happened, the null situation is that the weighted average probability should lend the case.
So with SQRT[0.278*(1-0.278)/12,000] = 0.004 being one standard deviation, both the observed stats are about 1.5 Z from the combined mean, which is not significant. ASS U ME ing a normal distribution, or close enough for “government” work.
If we have fewer than 12,000 IP with the combined situations, the STDEV would be higher, meaning that the difference of either percentage from the combined mean would have to be larger in order to be credible.
In my professional opinion, 27.2% is indistinguishable from 28.4%. Unfortunately, I have dealt a lot with actuaries who mistake precision for accuracy – they would carry those percentages to 6 significant digits, and then management would want to vary rates based on the 4.4% difference in frequencies (28.4/27.2), etc.
If we saw the same difference over multiple seasons, that would be an additional measure of credibility, in addition to the ability to combine multiple seasons (ASS U ME ing that they were comparable, perhaps not a robust assumption) to lower the STDEV of the binomial approximation.
However, if the percentages were widely different for several seasons (in the low 20’s, low 30’s, different years have different situations higher than in other years, etc) I would have to say that not only are the odds indistinguishable, but meaningless.
A good question would be if the odds of getting one or more runs in an inning, given that the first batter either got a home run or a walk, are statistically different from the odds of getting one or more runs in an inning given that no batters had been faced yet in that inning. Think about it.
While you claim your responses to your analytic framework were primarily of the ‘too many notes’ variety, my null hypothesis is that when you use a baseball stat analogy for a financial point on a blog you will get a statistically insignificant number of baseball comments. Which is false, ergo proven!
I’m confused on what your conclusion really means. I work in the hedge fund industry and there is definately traders and professionals who have an opinion to begin with and those who lead the markets with their actions.
– Richard
Hedge Fund Managers Blog
http://richard-wilson.blogspot.com/2007/10/hedge-fund-managers-pedigree.html