Was wondering if any of you do any probability testing on price series before running actual trade testing on the price data. For instance stationarity or dependency models, or any other probability tests. I've been reading more and more about this and am wondering how much more of a grasp on the vehicle your trading it gives you. For instance, would you be able to make decisions about the strategy you're designing based on the results you find? Or is it just semi-redudant... based on the premise that: if there are other people who have designed systems and are making money then I should be, too. I guess that could also be called the semi-lazy way, too. Any thoughts?

I do this alot. But, the type of analysis is not that complicated. IMHO, it has given me some interesting insights... nitro

hey nitro, thanks for replying... what programs and what type of testing do you use. and what kind of results have you found. i know you're primarily a scalper so i'm guessing you're applying this to those systmes your developing. BTW how're they doing? regards, onelot

1 -lot, I mostly use Mathematica (http://www.wolfram.com) for this analysis. My style very much reflects what I found. For example, I ran a Hurst Exponent analysis on ES data and found that it is less than (statistically significantly) .5. Therefore, I tend to take my profits fast, and look to reenter on the continuation. Although most see the scalper side of me, I like to think that I have a wide view of the markets. I am constantly trying to expand my horizons. For example, I may start intra-day "swing trading" the ES to capture more of the move than my scalps give. In addition, I am thinking of putting on longer time frame position trades on in certain European markets. If these types of trades did not adversely affect my margin, I would have done it already. The analysis that I have performed helps me on defining my "style" in those cases to be consistent with the time frame. I am doing well, thanks for asking... nitro

I do lots of time series analysis. Most of it is original work, but some of it comes from ideas outside of trading (like weather forecasting and immune system reponse). Here's a book that's coming out next month that I think will be of great value if you're interested in this kind of work. http://www.riskbook.com/var.htm

acrary and nitro, I'm glad both of you took the time to respond to my post. I had a feeling you (acrary) would be into this stuff and i'm suprised to find out that you are too nitro... but it makes sense, and I have more respect. I'm just getting my feet wet learning about this stuff and was wondering if you guys might be able to point me toward any more resources software/books. Value at Risk looks to be good and I just finished The Mathematics of Technical Analysis: Applying Statistics to Trading Stocks, Options and Futures by Sherry. I'll have to take a look at Mathematica, also Matlab. I'm a little worried as my background is definitely not as analytical/mathematical as I would like it to be, but the more I trade and the more I learn about trading the more it takes me in that direction. I have a couple of questions that maybe you guys could help me with. The first is: should every time frame that your're looking to test on be considered a seperate data series. I know that yes will probably be the answer but it ties into the next question. Which is: should every variable in the individual prices (OHLC) be tested as individual data series or is there a way of grouping them... especially if you use all the inputs for trading decisions. Or I suppose the most thoruough would be to do all sorts of combinations. I can see how this could get tricky. thanks a lot, onelot

People have suggested all kinds of ways to do this. One that I found interesting was to multiply the O, H, L, C by sin and cos terms, then adding them together. I don't remember the exact details, but it helped in Harmonic Analysis. Respect? 1-lot, serious, I put very little faith in the stuff, except it is a very rough place to begin for the _trader_, since traders usually place directional bets, and that stuff (VAR) is meant for directionless trading. I understand if feel you have to take your respect back now. nitro

How does one model or extrapolate a potential squeeze in the market ? Congestion/inside range ? I trade off of order flow in an attempt to determine whose nuts are in a vise.

Here's some stuff to start with: http://archives.math.utk.edu/topics/probability.html The Analysis of Time Series an Introduction by C. Chatfield Statistical Data Analysis Handbook by Fancis Wall Probability and Statistics by Athanasios Papoulis I treat all data for a symbol as part of the same time series regardless of the period I'm analyzing. The beginners in the field used to look at the data and try and find something with a positive expectation. Then they did a walkforward test to show that it persisted. I've looked at that method and found it to be pretty useless because all they're doing is mapping to the old character of the data and hoping that it continued into the future. What I'm looking for is opportunities that show better results than 70% of random possible combinations. What I've found is these opportunities tend to persist year after year. This also gives me a objective method for stopping the use of a method before I lose any money with it. Below is a gif for one daytrading model that has had 20+ years of profitability in the SP market. The overall profits are correllated with the daily range so that larger profits happened during periods of higher volatility. There's probably as many ways to view the data as there are chess moves. The more bars you look at, the more complex they become. I look at hundreds of different patterns to find relationships that may be interesting. It's only through banging out the analysis that I find something that beats randomness at a acceptable level. Here's some examples of grouping: current bar close to it's intra bar close (high+low)/2, inner range of the current bar (open and close versus high and low), high to range of current bar, etc. You can see this can go on and on. I temporarily exhausted my creativity in the hundreds of combinations but I know I'm only scratching the surface. Throw in a indicator such as macd with the patterns and you've added hundreds more combinations. If you need starting ideas, look at all the candlestick patterns.

I'd start by trying to find evidence of squeeze events happening in a market. Say 10 years of events in the coffee market. Once identified, work backwards to see if anything in the data was predictive. I've found that throwing in all the data that might be useful and using a Genetic Algorithm package is a good starting point for finding out what data may be useful. Once done, it's grind away at the data to find correllations that are predictive. After you complete all that, create a model with the predictive associations to see how it compares when mixed in with all the other data. If it makes enough money to be worthwhile and is much better than random at predicting the events, you're ready to put some money to work.