Sunday, March 15, 2009

Predicting Trending and Non-Trending Markets: A Direction for Research

My recent post suggested that the intraday NYSE advance-decline line ($ADD) offers a useful perspective on whether the day is shaping up as a mixed, range day or a one-sided trending day. Because an awareness of the emerging day structure is key to trading tactics--whether you'll trade or fade strength or weakness--few research questions are as important to the active trader. Nonetheless, I've seen no solid research on this topic in the writings I've encountered.

Let's frame the research challenge more broadly: If we consider indicators A, B, C...etc. at points of time X, Y, Z...etc. during the morning hours, which indicators most accurately predict trending and non-trending markets earliest during the market day?

Say, for example, that indicator C at time X is the best gauge of whether or not we'll have a trending market, significantly correlating with price movement from time X to the market close. It would then make sense for a daytrader to sit out the period from the market open to time X, waiting for the noise to sort itself out before an educated estimate could be made relative to the issue of day structure.

Armed with this information, a trader could then establish tactics for the trading day and use his/her feel for markets to aid in the execution of those tactics.

Here's a simple example: If we go back to late September, 2008 (which is as far as my intraday data set for the indicator goes), we find that the opening value of the advance-decline indicator ($ADD) correlates with the final, closing value by about .30. That means that 9% of the variance in the closing value of $ADD is accounted for by its opening value.

If, however, we look at the value of $ADD after the first 15 minutes of trade, that correlation with the closing advance-decline value jumps to .63. That means that 39% of the variance in the closing value of $ADD is accounted for--a significant jump. Indeed, if the first 15 minutes in $ADD are positive, the average closing value of $ADD is +675. If the first 15 minutes are negative, the average closing value of $ADD is -1063.

These findings are suggestive and illustrative only. Crucial questions remain: How do the indicators correlate with an actual price-based measure of trending/non-trending? Will a combination of indicators prove more predictive (or predict more early) than a single indicator? Are the indicator values related to trending/non-trending in a linear or non-linear manner?

Good posts offer fresh answers to tough trading questions. The best posts, however, offer fresh questions and directions for trading. Personally, I think this is the best post I've written in quite a while.


Joshua said...

I'll second that nomination. Good Stuff! Maybe combining a more anticipatory approach (mean reversion between trend days & range days) with this would boost the odds?

Is there a higher correlation when the opening TICK is skewed in the same direction as the ADD as well?


PancerBali said...

It’s so nice site. We love to see more on this site. Keep on updating… MonkAreYou Bali *sfsgd

Jorge said...

Dr. Steenbarger,

Your recent series of posts is (not surprisingly) certainly excellent. As for the predictive value of a combination of indicators vs. a single one, you may want to get in touch with Tim Knight of ProphetCharts/ThinkorSwim regarding their ThinkAI, which seems to combine both probability of trend/range and expected volatility (รก la pivot points), giving an initial short-term "prediction" 5 minutes after the opening and then expanding its length of projection as the session progresses.

While I am no fan of indicators, preferring simply volume at price and S/R, I think that this idea of developing/adapting indicators to diagnose what kind of day to get ready for as opposed to trying to predict a certain price or turning point is something that has true value and that many of us would be well advised to research further.

Thank you for the challenging questions. Best trading,


PS: Can't wait to see what the "become your own trading coach" blog evolves into once your new book is published (finally, breathe, breathe)

Brett Steenbarger, Ph.D. said...

Hi Josh,

Thanks; my research does show better results when indicators are combined with each other and with price level info.


Brett Steenbarger, Ph.D. said...

Thanks, Jorge; I agree: using indicators to identify market structure is different than trying to trade from indicators alone. I'm also not a big fan of the latter--


John said...

Dr. Steenbarger,

About a year ago I listened to Dr. John Clayburg ( explain something he calls the Directional Day Filter (DDF), and claim a 75% predictive power for the daily direction. Quoting from a paper in his website: "For example, the author introduces the use of a simple early morning filter which, with 75% accuracy, defines each trading day in such a way as to make interpretation and use of indicators much more reliable. By using this technique to enter the market each day, traders can significantly increase the accuracy of their trading."

Get In Get Out said...


Did you ever do a study on the exact time period after the market opens when we can 100% predict if the market is trend or range bound for the day? In addition, Do you have any research on lets say, a trader sees by 12:30 the market is trending, is there resarch on the odds/probability of picking a stock in a sector that is in trend with the trending market and maximing intra-day profit by buying or sellling any stock that is trading in trend with the sector and overall market trend?

dgoverde said...


I would agree with your assessment. Great post, and for the exact reason you noted. Keep up the good work.