#10
|
|||
|
|||
Re: Stock Market Data
[ QUOTE ]
I'm interested in what kind of models you tested the data and whether or not your performed some sort of optimization when running your backtest. I've ran a few tests on old data and I find it's best to split the historical data into 2 sets. The first half of the data you test your system, you then 'forward' test it on the more recent half of the data. [/ QUOTE ] [ QUOTE ] Where can I get future data? I'd be willing to pay well for it. [/ QUOTE ] Yeahyeah... what I'm refering to is doing what is stated above. Use 15 years of data starting 20 years back, say, to select and optimize, then use the most recent 5 years of data to see if your models worked. Mine never did. Mind you, I did this about 20 years ago and nothing I did was very sophisticated. Mostly no more than you can get today from Yahoo or whomever over the web. Tried various simplistic line-fitting things, picking stocks going up with minimal variance or maximum gain or some combination of the two, spectral analysis looking for cyclicals, simple "top N percentile in M categories" sorta things. Most of my attempts, in fact, seemed to produce an opposite result...if I picked stocks based on what my approaches suggested would rock, I would have very handily gotten a far-below-market return. So it was back to actually studying the companies and annual reports and whatnot. That, and just throwing darts. |
|
|