Who here approaches trading from a purely statistical standpoint. By that I mean, screw charts and visual patterns. I'm talking measures of dispersion, variance, standard deviation, probability distributions, confidence intervals, p-values, z-scores, goodness-of -fit tests, correlation analysis, significance tests, etc....
I want to approach it this way. Is there a good book on how to do these things? I'm pretty good in math but how the hell to apply it on stocks?
I too am interested in the way individuals use this to look at the market. After completing many statistics classes in previous semesters, I want to see its applicable approach to trading. I'm not sure it is an approach I want to do, but I am just curious as to how some do it.
I would say that I am a statistical trader â I donât use patterns or charts per se. I use code to find âstatistically relevant environments and eventsâ and go from there. Simply put, it is relatively straight forward to define a medium term trend and exploit it for a profit â no charts or patterns required. Run the code over a large number of stocks and place the buy/sell signals. As for the stats: I have worked for a government statistics organisation and would suggest that traditional methods â t-test/ANOVAâs/means/distributions etc would not be the best way to go. What you need to do is analyse a data set aka trade log from the backtest, and determine the risk adjusted efficiency of the system compared to the performance â thatâs where the stat analysis comes in. Use such metrics to compare the performance of various systems and see if what you have produced is statistically relevant, or if changes to systems are significantly different. EG, there may be no statistical difference between two different exit strategies. pneuma
I use statistics extensively in all my systems, but as a tool, not the underlying concept. In my opinion, there has to be a valid premise behind any trading systems; otherwise you're just data mining. You could strike gold - i just consider that unlikely.
Ive tried doing the same thing but found it quite difficult. 1. The market isn't normally distributed so you need to use non parametric stats. 2. The only decent long term numeric datasets are of the EOD high low open close variety and a lot of the devil for markets really is in the detail. 3. I've used frequency analysis to look at runs [wave heights] to try and assist setting trade target, but it hasn't really been any better than visually looking at the chart. Same applies to stops. If anyone has had any meaningful success I would be interested. Cheers
This is a thread which throws up the expected failures of knowledge and understanding and consequently the misconceptions in this thread topic, and it is an area I mention very little. I have always worked with the metrics of the markets (eg CL,YM) established and known to myself. OHLC is insufficient and insufficient for any testing or extrapolation purposes. What is it that you see? What is that never gets discussed on ET? Price and? Price and time. The vertical axis is price and the lateral axis is time. Price by time. When? Just when are upmoves and downmoves occuring in the daily menu? What is their order and magnitude on the time axis? Certainly it is crucial in day trading to know the metrics of the market you are using, if you want to extract most of what the market daily offers. The reason all of this doesn't matter is because you do not keep time data and cannot otherwise access precise time data from back data. And you wouldn't have the time to set it up and tabulate it anyhow.
Just using math without any underlying fundamental premise is, mathematically speaking, a near surefire way to fail. I use stats extensively in the creation of my trading systems, but they all have an underlying fundamental premise to them.