Saturday, October 28, 2006

Can There Be An Objective Basis For Subjective Knowledge?

In my recent post, I cited Ayn Rand's assertion that philosophy is the most practical of disciplines. This is particularly true with respect to epistemology, that branch of philosophy that deals with knowledge and its acquisition. What we count as truth ultimately determines how we pursue truth, whether in markets, science, or politics.

Ms. Rand also stressed that contradictions cannot exist in reality. Where we find a contradiction, she advised, check your premises. At least one of them will be incorrect.

The positivism expressed by David Aronson's fine book Evidence-Base Technical Analysis yields just such a contradiction. Before I launch into the contradiction and a possible resolution, allow me to mention (on an unsolicited basis!) that Aronson's book is a substantial contribution to the literature on technical analysis. One need not agree with his strident formulation that the discretionary application of technical analysis does not draw upon "a legitimate body of knowledge but a collection of folklore resting on a flimsy foundation of anecdote and intuition" (p. 261) to benefit from the reading of his work.

Specifically, Aronson has accomplished four worthy ends:

1) He clearly explains the importance of testing trading ideas and illustrates how that is done;
2) He describes both the strengths and potential weaknesses of data mining approaches;
3) He tests specific technical trading patterns and demonstrates how difficult it is to obtain statistically significant findings (and how easy it is to generate illusory ones);
4) He reviews major theories and research findings in behavioral finance to help traders begin the process of finding more promising patterns.

For a book that deals with technical themes of logic and mathematics, his work is eminently readable and understandable. I would rate it alongside Kenneth Grant's "Trading Risk" as a must read for developing traders.

That having been said, I believe Aronson's positivist roots--leading him to equate knowledge with declarative statements known to be true--create a Randian contradiction. If all knowledge consists of verifiable statements about observables, then Wittgenstein is correct in his formulation: Whereof we cannot speak, thereof we must be silent. Subjective knowledge must be an oxymoron.

But here is the contradiction: It is common--certainly in my visits to proprietary trading firms, hedge funds, and investment banks--to find discretionary traders who have achieved a high level of trading success year after year, trading actively. Indeed, I wrote about just such an individual in my new book on trader performance. These are not mere anonymous figures on bulletin boards puffing up their performance stats. These are traders who have account statements and risk managers able to verify their superior performance. And yet they cannot verbalize specific rules or systems for their trading.

In short, they have knowledge, but it is not of the verbal, declarative kind.

The existence of such implicit learning has been known in cognitive neuroscience circles for decades. Philosopher Michael Polanyi offered an influential treatise on tacit forms of knowledge, and Arthur Reber began his groundbreaking studies in the 1960s, culminating in his 1993 text "Implicit Learning and Tacit Knowledge". More recently, Timothy Curran, in the "Handbook of Implicit Learning" summarized research that found different brain mechanisms mediating implicit learning and explicit, verbal knowing.

How does implicit learning occur? Through intensive repetition, in which individuals become sensitive to complex and noisy patterns. This is how young children learn to speak grammatical English before they can verbalize the rules of English grammar. It's also how we can identify a face that we could never adequately describe in words, and it's how we know when such a face is starting to display anger or sadness. Serial reaction time experiments show that subjects can learn complex statistical probabilities in sequences of data with enough repetition and feedback. Interestingly, they can anticipate events in those sequences, but cannot verbalize the complex patterns that they have internalized. (The research of Axel Cleeremans is particularly eloquent on this point).

Such subjective knowledge is not "devoid of information" as positivist philosophy would have it. There are, of course, intutions that prove to be invalid, but reducing all knowledge to testable hypotheses would probably eliminate most of the knowledge and understanding that lies behind great art, as well as most performance fields such as athletics. The deep knowing of musicians, chess players, and fighter pilots can hardly be reduced to sets of explicit propositions.

If we admit the possibility of such subjective knowledge, then it follows that the development of algorithmic systems with fully backtested rules is not the only way to achieve trading success. It may be possible to generate success by accelerating processes of implicit learning through the use of simulation/replay and intensive feedback. Ironically, the weakness of much technical analysis is not that it is subjective, but that it pretends to an objectivity that it cannot support.

Can there ever be an objective basis for subjective knowledge? I believe so. A trader's track record of profit/loss can be compared to random entries/exits (as well as buy and hold) to objectively determine whether or not that trader--over time--exhibits significant skill. Imagine a Monte Carlo simulation in which we create random entries and exits each day that a trader trades, with identical trading frequency and holding times. Suppose that such a simulation is conducted 10,000 times by computer. The resulting distribution of P/L would display the likelihood of achieving a given level of profitability by chance alone. If a trader's subjective trading methods consistently produce results at the very upper tail of that distribution, we can objectively infer that the subjective trader is skilled.

In other words, by treating each trader as a trading system, we can evaluate that trader's level of knowledge, regardless of whether the knowledge is subjective or objective. In the absence of such score-keeping, discretionary traders have no basis for a belief that they possess a true edge in the marketplace. One need not resort to positivism--or system-based trading--to be rigorously scientific. It is precisely because intuitions are fallible and human senses are so easily deceived that we need to distinguish truly superior outcomes from merely random ones.

9 comments:

Brandon Wilhite said...

Nice post! I couldn't agree more. There is so much we know that does not fit neatly into the reductionistic, and cartesian, framework of knowledge. It works really well for physics (to a point), but not well at all for many other valid areas of human endeavor. You give a good example of how we could argue this empirically.

Brett Steenbarger, Ph.D. said...

Thanks for the note. Much of these discussions become polarized as pro-technical analysis vs. anti-TA. What I've seen in working with successful traders is that their (stated) methods are very, very different. It makes me think that what they are doing and how they are conceptualizing what they're doing are two different things. That's certainly the case among psychotherapists: what they do that works actually has little to do with their adopted theories.

Brett

John Wheatcroft said...

Great post – some thoughts neither well thought out nor well articulated but for what it they are worth.

Isn’t - implicit learning through the use of simulation/replay and intensive feedback – the same as the development of algorithmic systems with fully backtested rules?

If not – why isn’t it – I “learned” the process long before I “described” the process through simulation/replay and intensive feedback – i.e. I tried to do the same thing over and over again until I found out which “same thing” led to success. Then I wrote it down or, actually, completed development of an algorithmic system with fully backtested rules.

The weakness of all technical analysis is that it reports “last weeks news” as if it is fresh and current. But as a result of observation over a period of time we know that price – once in an apparent pattern of rise or fall remains in that pattern for a period of time within certain parameters of certainty. While I don’t use MACD, ROC, SAR, Bollinger Bands or any other “technical indicator” in my trading I can, as a result of study and prior experience, generally predict what they would show even though they aren’t on my screens. Is that “implicit”?

I guess what I’m trying to say is that the “mechanics” of selection become unnecessary for most successful traders after awhile and indeed sometimes become impediments. For example if someone insists that just because a certain technical indicator has been in an oversold configuration for weeks and weeks that the market can’t continue to go higher yet it does and breaks records in the process – are they a good trader or bad trader?

I’d posit that perhaps they were a good or successful trader if they continued to trade and make profits during the period in question but a bad trader if they sat on the sidelines and waited for their definition of “inevitable” to occur.

Or we could say that they are they merely managing their risk during a spate of uncertainty and thus, because of their “implicit” learning, they have preserved their capital? Once more a question answered with the fateful two words – “it depends.”

Bottom line – I believe that your test would fail because looking at P&L sometimes (many times?) only exhibits the result of luck, both good and bad. We simply can’t “know” what a trader sees in his/her mind’s eye.

Brett Steenbarger, Ph.D. said...

Thanks John. Ideally, a test of trader skill would first identify traders with superior P/L and then track their performance in real time over a considerable period. This would be like in-sample and out-of-sample testing of trading systems. If the trader's fresh performance still lies at the upper extreme of a distribution that you'd obtain with random entries and exits, then you'd have a very solid basis for inferring skill. If the fresh performance fell sharply and rested firmly in the middle of the distribution of random outcomes, you'd be more likely to think that luck was a factor in the trader's initial success. Thanks for the opportunity to clarify.

Brett

Brandon Wilhite said...

What I would like to know is whether or not you have done any such studies? ...I would find such results very interesting from a philosophical standpoint as it would help to verify a certain thought expirement I constructed.

I think that there would be some overlap between implicit learning in a discretionary trading method and algorithmic trading, but they are not the same. I think that the main difference would be that a human discretionary trader would be able to learn much more quickly than an algorithm. The human trader also encapsulates much more information than an algorithm ever can. I trade with algorithms, but I don't change them or override them (even though I "know" sometimes that the signal is bad). Although as I think about it, I believe that the algorithms themselves are also "training" me as a result of this whole process. Anyone who's really interested in the posts you've given on this topic so far should go read Michael Polanyi...he's really an easy read.

Brett Steenbarger, Ph.D. said...

Hi,

Thanks for the note. Such an experiment would indeed be interesting, to compare how machines and people learn. I have not performed such research myself; nor am I aware that anyone has tackled the area.

Thanks for the Polanyi recommendation. His books on the Tacit Dimension and on Personal Knowledge are very worthwhile reading.

Brett

Globetrader said...

Brett,
I'm a discretionary trader. Meaning I'm not able to mechanize my trade selection and trade entry and exit rules. I really tried, but failed miserably in real trading, even if backtesting looked so promising.

But for a long time my %-Win, %-Breakeven and %-Loss trades remained relatively static.
So this June instead of defining hard trade entry and exit rules, I decided to trust myself to be able to achive about the same statistic results next month again.

It was a very difficult process for me, as that meant giving up control and relying on my intuition to provide me with a feeling that a certain setup seen on the screen would fit my tade selection criterias, even if I'm not able to verbalize them all.

It was exactly what I had to do to become a profitable trader.

It's nothing to be recommended to a novice trader, as you need a lot of screentime to train your intuition. For me it took 5 years looking at charts, which today still have features I had on my charts 5 years ago.

If you can compile statistical data about your trading over a long time and you see consistencies, you don't need to come up with a objective trade rule, you can replace it with a trust in yourself to achieve this result again. And you can then build on it and try identifying the areas in your trading which need improvement.

For me it was bringing down the average $-Loss / Trade proving market lore again right: Take care of your losses, the profits take care of themself.

Brett Steenbarger, Ph.D. said...

Hello Globetrader,

Thank you very much for the very insightful post. Your point about how statistical data on your trading reveals consistencies that are your own, internal trading rules is excellent. Congratulations on your development as a trader. You'd make a fine coach and mentor for others.

Brett

Sophokles said...

Hi! Thanks for a great read!

Globetrader said, "If you can compile statistical data about your trading over a long time and you see consistencies, you don't need to come up with a objective trade rule, you can replace it with a trust in yourself to achieve this result again. And you can then build on it and try identifying the areas in your trading which need improvement."

If we, let's say, compare this knowledge above with the strathegic thinking of a chessplayer. Would the reason for thinking mechanical be based on lack of experience or logic thinking? Since it's very unlikly that the same situation would accour in the exact same way again?

So, in order to get the experience required I would need to make mistakes? Even though some of those mistakes could be prevented by mechanical thinking.

Hope you follow my thoughts here.

From a swedish student of economics...