Jeez yes, the markets change continually, what is somebody searching for, "the music of the spheres" or the "unified field theory of the stock market"? I did test a system on the 1932 Dow Jones Industrial data once and it worked. I thought that was interesting. It did make me feel like I had something fundamentally sound. A backtest on 30 years of data would make me feel like I had a lot of backtest data, not like I had something that would work. Max
You need to test on unseen data and find out how well it performs. But good first step! Good luck on your testing....it is very difficult to find systems that perform well over a variety of market conditions.
For any EOD system where you test SPX, if your system enters or exits anywhere but the close, you will have better results than reproducible in real life. Why? Because the open that shows on SPX from most data vendors is not representative of related trading vehicles (ES, SPY) because most data vendors include the evening session as part of the open. E.g. ES closes 2 points from it's low. Your system says to buy 1 tick below close. Next day ES opens down 7 points (5-point gap below yesterday's close). A system using SPX data from a vendor that includes overnight session just made 5 points more than was possible in real-life. Test your system on SPY or ES and you'll get more realistic results. Having said that, your sample size is pretty small. Also, not knowing any parameters and how you optimized or what data sets you used, you need to test on out-of-sample data before you can draw any realistic conclusions. Last suggesion- go backwards in time and test the 800 bars before your test period. How did it do? And don't use SPX.
agree with all critics. too few trades. too short time. wrong instrument. do not test on something you cannot trade. index data distorts the picutre heavily - and always against you. nevertheless said that, thirty years of data are useless since there has been no tradeable instrument at that time. too long data can add to an illusion as well as too short data. using the future since mid eigthies will do the job. peace
I figured that he would get as much data as possible. I know the ES wasn't around 30 years ago, but I still think he should get as much data as possible. More data will reflect if his system is robust enough for a variety of market conditions.
in essence I think you are perfectly right and most people use too limited data. I think there is no final truth to this. it depends on your style. if you are willing to switch strategies several times a year, why not use shorter periods, as long as you are not doing the fit and run and refit again game. I see it as two sides of a singel coin. inefficiencies come and go, some last longer some shorter. if you limit yourself to those that already existed 30 years, you might loose opportunity. on the other hand the shorter the data the more likely you fall into a statistical trap. peace
1. Hardest thing is when to switch. 2. I think most of the short term system traders are doing fit and run. I've asked a question to some fit and run system traders about when they'd switch. There answer is always when market conditions change. Well...They end up losing a lot before they know conditions have changed. Next thing you know they don't exist any more. Just commenting... Good trade! Trend