Does it work? I backtested many similar strategies. Worked great looking back but not in going forward. What is obvious is, if the strategy is obvious, it probably won't work because too many already are using it.
Tough to find great backtests. Tough to beat a simple BTC buy and hold. Does it works ? I’ll tell you that in one year. Currently market is shit and I have very little to no exposure. Seriously I’ve tried a bunch of stuff and nothing really beats the index. Indicators… Linear regression… Mean / stdDev… Tried on 5min, hourly and daily data. But just running something live is interesting.
A few things make the out of sample / live trading more likely to work: a) Longer timeframes: daily bars (or even weekly bars), long lookback periods eg 52+ week breakouts, wide stops b) Large universe, use all stocks in S&P, or even "All US stocks" c) Low number of parameters, simple system. An example strategy that works for trend following stocks on ASX market: * Weekly bars, "All stocks" universe for Australian stocks. Include delisted stocks in the backtest so there's no survivorship bias. * Minimum of 10 positions (diversification) * Only stocks under $10 (better trending stocks) * Entry: When weekly close breaks 52 week high, buy on next open * Exit: when weekly close drops below a trailing stop of 2.7 ATRs below the trailing close, or when weekly close is below the 36 week EMA. * If too many entries - rank by lowest volatility. Low vol trends are less likely to be stopped out. That's 18% return over many years, beating the index return (8-9%), but people don't want to run it because they think they can get 100% trading manually.. or don't want to handle the down months and volatility. Even some years will be down, but that's made up for by the great bull market years. A few tweaks like an index check, or rate of change check can improve slightly better as well. But you have to be careful not to overfit.
Thank you. I appreciate your suggestions and will definitely look into it and give it a try. Merry Christmas.
Low vol also have better risk adjusted returns. There is a documented anomaly which contradicts the fact that riskier assets should provide greater returns. I would rank by sharpe by default. That’s something I also replicate with cryptocurrencies within the backtests.
Let’s see in 1 month. I didn’t trade the current system the whole time. Lost a bunch on mean reversion with CFTX.
A scheme of the cross-validation procedure. Data are split into two sets. Cross-validation is performed on the first one (training dataset), to estimate the expected performance of a strategy. The model is then optimized on the entire training dataset, and validated on the second one (validation dataset).