Here's a pnl chart I thought I'd share. The first segment was the in sample backtest which the strategy was developed using(25 months). The middle was an out of sample backtest results tested after the majority of the curve fitting (2 months). The last is live trading for the past 7 months lining up with the backtest win rates/expected pnl. I've made lots of small changes since starting live trading but it still the same strategy. The total sample is currently just under 1,000 trades combining the backtested trades and the live trades. In my opinion backtesting works well as long as you can model slippage. And imo the most best way to gauge risk/profitability is by comparing the drawdown to total pnl.
ok thanks...so these 1000 trades are like over 100 equities, so like only 10 trades per equity over a 4 year period??
I have no idea how many trades per individual stock. I'd bet there are a lot of symbols it has only traded once. I guess its more relevant in futures trading when people are looking at only a few markets, which behave very differently. Basically the algo watches the nearly the whole stock market, if the set up occurs in any symbol, it takes the trade. Having a decent sample size on every equity I trade doesn't really make sense. All stocks behave similarly for the most part so just worry about the set ups and not what you're trading.
Yes I agree....thanks for your points. Since I trade mostly futures, I am coming from that world I am curious which software you used to backtest and research all these stocks together in a portfolio. Since you are talking about all stock market, you are talking about testing 2000-4000 stocks together. I use multicharts and MC surely can't do anything like this
Anyone can post a historical strategy that makes money in the past. The real question should be: How did your modeling results compare to actual trades...???
Best back and forward is profit factor of around 2.5 for 7 years tick data in betterrenko. Forward has been going for 6 months.
The software is made in house. The whole market really isn't that big if you're ignoring illiquid and low range symbols. Usually 1,000 symbols at the most. The third segment was live trading (starting about 7 months ago until now) compared to the first which was backtest(24ish months starting in 2011). So it's pretty clear that live results are fairly close to historical results. You do see some discrepancies when backtesting the live period. There's some slippage and the occasional missed trade but for the most part it's quite accurate. It's also fairly predictable. With limit orders, you have to be aware of how you model fills. If you are trying to trade a strategy with 2,000 shares don't expect to trade stocks with 20c spreads with 100 shares on the bid and offer.