Saturday, November 10, 2007

Making Decisions From Current Stock Market Data

In recent posts, I have taken a look at the relationship between volume and stock market movement. I've also examined market direction and its relationship to the NYSE TICK. It's been a bit confusing for some readers, because I reported on contemporaneous relationships among these variables: how the current volume or TICK is associated with current price movement. This is quite different from taking a predictive look: examining past values and their association with future ones.

And yet I find that the contemporaneous look is truer to my own decision-making process in trading than the predictive one, particularly with respect to short-term (intraday) trading. In this final post in my series, I'd like to explain that decision-making process.

We often think of decision making as an event: something that happens at a single point in time. We observe X and predict Y. We see a pattern, and we decide to buy or sell. There are certainly decisions that arise in such a manner, but I think there's generally much more to making up our minds.

Consider members of a jury. If they approach their task properly, they enter the courtroom with an open mind. They hear testimony from both sides, listen to cross-examinations, and weigh evidence. As each new witness or piece of evidence is introduced, the jury members revise their estimates of the guilt or innocence of the defendant. Their reasoning occurs over time, reflecting a weighting process that considers new data in the light of past information. A statistician would say that the jury members are engaged in an informal Bayesian process.

Much of our decision making in life follows the informal Bayesian model. Consider:

* We're in our car trying to get home by a new route. As we pass by streets and buildings, we revise our judgments as to whether we're going the right way or not. At some point, landmarks might be so unfamiliar that we decide we're getting lost and we turn around. Alternatively, we may see familiar streets and decide we're headed in the right direction.

* A physician listens to the complaints of a patient and then makes a physical examination. Along the way, the doctor revises his or her judgments about the patient's illness, discarding hypotheses and narrowing down to an eventual diagnosis.

* We gather economic reports to gauge the health of the economy. With each major piece of data, we revise our estimates of whether or not we're headed for recession, inflation, Fed easing, etc.

In other words, we behave like informal scientists. We accumulate tentative hypotheses about the world and then test those against our ongoing observations. Those hypotheses are either strengthened or revised based upon the fresh information gathered.

All of which brings us to stock market volume and the NYSE TICK. My job, as I'm trading, is to use current readings of volume or buying/selling sentiment to estimate the eventual distribution for the day. If volume in the first X minutes of the day is running significantly above average, I make an initial inference that institutions are active in the market and that the day will see above-average volatility. With each new five-minute reading of volume (compared to the average five-minute volume for that time of day), I revise my ideas about institutional participation and volatility to come.

Similarly, I view the NYSE TICK, not as a set of discrete readings, but as a distribution of values over time. If, in the opening minutes of trade, I note a negative distribution of the Adjusted TICK values, I infer that selling sentiment is outweighing buying sentiment. I then look to each bounce and drop in TICK to see if that distribution is changing, adding to negative or positive sentiment. On Thursday, the TICK distribution was negative but not falling. Then, with an upward shift (which is often heralded by an upside breakout in the TICK values), I quickly revised my estimate that the market would close near its lows. Indeed, my revision suggested we could see significant short-covering. With each subsequent positive TICK reading, my own estimate of the market's bounce potential gained confidence.

I believe this informal Bayesian process gets at the heart of what it means to be a discretionary trader. It's the impulsive or inexperienced trader who makes decisions at a single point in time based on a single observation. The skilled discretionary trader--even at very short time frames--weighs evidence as it comes in and revises expectations accordingly. For the scalper, that evidence might come from a depth of market ladder; for a short-term trader it might come from volume transacted at market bid vs. offer; for a longer-term trader, it might come from accumulation of money flows. Like the wandering driver in a new city, we look for landmarks and decide over time if we're headed in the right direction.

Getting lost in new places is inevitable. The good trader, like the good driver, is one who keeps an open mind to new information and can quickly change direction if needed. Such openness requires a non-defensive stance vis a vis one's ideas--a readiness to acknowledge being wrong--and an ability to be in the present and (like the jury member) impartially process fresh information.

RELEVANT POST:

How I Trade
.

4 comments:

Brandon Wilhite said...

Brett,

Thanks for the great series of posts.

This one ties it all together nicely. I had thought there had to be a discretionary element involved with the types of studies you were looking at.

It seems to me what you're using these for is to keep you in tune with what the market seems to be doing at the moment, and then trading accordingly. I've often tinkered around with the idea of creating a program that would calculate these types of probabilities on the fly and then trade them, but such an undertaking is fraught difficulties.

In a sense, I believe that this is actually what the algorithmic trader is doing, but because of more stringent requirements the decision making process is at a much coarser grain than can be achieved by the discretionary trader. In other words, the algorithmic trader would need a multitude of backtested trading systems to go with the real-time probability calculator.

One question I have is: Do you change your position-sizing based upon the informal Bayesian reasoning you've done?

BW

Anatrader said...

Brett

QUOTE
Bayesian inference usually relies on degrees of belief, or subjective probabilities, in the induction process and does not necessarily claim to provide an objective method of induction. Nonetheless, some Bayesian statisticians believe probabilities can have an objective value and therefore Bayesian inference can provide an objective method of induction
UNQUOTE

I am learning new jargon each trading day, thanks to your posts. You also liken analysis to a jury making decisions.

Just as I am taught to use primary and secondary technical tools of analysis to find the probabilty in my favour to enter a trade.

Brett Steenbarger, Ph.D. said...

Hi Brandon,

Thanks much for the note and support. I agree that it would be a worthy task to quantify those updates of odds as new data come through. I do adjust position sizes for volatility, so that each trade is a similar bet with similar risk/reward characteristics.

Brett

Brett Steenbarger, Ph.D. said...

Hi AnaTrader,

Yes, designating tools as primary and secondary can very much help with the ongoing reasoning process. Thanks for the note and all your support.

Brett