Having a stop-loss is a prudent way to improve a strategy. In practice, you want to limit the loss in case the unexpected happened. On the other hand 4% improvement is in the range of statistical error unless the number of trades in the backtest is in high thousands.
Data mining is a database term I used when I was an IT DBA building data warehouses on large scale IBM DB2 systems before I retired. We tend to over use the term and associate it with price data alone. Actual data mining uses data in many tables (and columns in them). IMO the type of data mining we are closer to using in back testing is filtering. One technique I use with filtering is extracting optimization data. The filter process screens out profitable settings that do not meet my goals. The process works like this: - First I perform a back test and produce the optimization in Tradestation. - Then each row in the optimization is loaded into excel where I do a 3D graph of the key input parameters. - Next all the settings (rows) are loaded into an access database. - Using SQL I screen out all the settings that do not fit the specs in my plan. - Then I compare the screened out data to the 3D graph to ID the best settings to test - Then I forward test the back test settings that made it through the process. - Last the performance during the forward test has to be inline with the plan and backtesting. This is the type of data mining any trader can do. To do this your plan should identify what statistical data from a backtest optimization is acceptable to trade with.
"3D graph of the key input parameters"? Is this taking the input parameters, two at a time (for the axes X & Y), and plotting them against an optimization output (e.g. net p&l, or win/loss ratio, or profit factor, etc) (for the axis Z)? Thanks.
Yes, the way you described the process it is the way I do it. I normally take my 3 key input parameters 2 at a time against profit factor, etc⦠Then I try to fit them on my screens all at once and take notes. There are many interesting relationship you can dig out of these 3D graphs that tell you how you strategy performs against price data. One field I like to test in swing trading, for example, is using the average bars in losers. I look for fewest bars.
Good morning! It's a bright and sunny day out there. Too bad they're all dark and cloudy for you huh?
Mizhael, Hi I am a bit late to this thread, but I am thinking about stops myself right now. 4% is noise. BUT does it limit your max hit on any day? If so, as it's not effecting your performance overall - put it in. BUT is it really helping? I have found it difficult to find a working stop in a fully automated strategy. For both long-term models and intra-day. I know a lot of chaps on this forum would disagree. SO I wonder if any one who uses stops and is fully automated, would mind sharing a chart? I would really like to see the equity curve of some model, plotted next to the equity curve of the same model with stops. That would help prove something to me and I would be most grateful