Congrats on the book Rob! I just received my copy. We also got a shoutout in the acknowledgements section, thanks! So I see you left it on us to figure out how to integrate fast mean reversion with a daily system using DO. I'll just use my usual strategy, wait to see how you implement it in pysystemtrade. Jokes aside, skimming through, looks like a great read, thank you!
Yes, thanks for the new book Rob! I think I'll at least integrate the skew rule and the 70\30 volatility estimation into my system, also need to finally recalculate the weights of everything based on handcrafting, I'm still using optimized weights from pysystemtrade, it'll require digging deeper into manual costs and turnover estimations from the first book to know which fast rules to drop for which instrument.. Got too lazy just sitting an looking at P&L, need to do something productive for a change
Been going through your new book Rob - a lot of good stuff. Your safer fast mean reversion system looks very smooth indeed. I see in the book you said it is difficult to attribute performance for the limit order backtest, but I had an idea to decouple a strategy's forecasting performance from its execution performance. If you take the fast mean reversion strategy and assume you always enter into the ideal positions at the mid price, and then take the difference between the results of that backtest and the original backtest assuming using limit orders, that difference should be your smarter execution performance? I would assume for most of the strategies in your book this measured execution performance would be negative since you are transacting using taker orders, but in the case of the fast mean reversion strategies, this should be a positive amount? For these strategies, would you say the bulk of the performance comes from this positive execution performance or the forecasting performance?
Yes that's a good idea, will add it to my (long!) list of things to look at when I get time. Probably not a priority for me, as I don't think I will actually trade fast MR myself. Rob
Does anyone have continuous EOD OHLCV history for Random Length Lumber future since early 1970s? For backtesting I was using RT data from Tradingview (ticker CME:LBS1!), but yesterday they not just delisted, but completely deleted this ticker and it's sub-tickers for monthly contracts!
I've been going thru Rob's new book, got to strategy 22 and I've now started to implement stuff from the beginning. Has anybody implemented basic stats Rob calculated in strategy 1? I've tried, both using my data, and using the data Rob provides with the book in the AFTS repo and I'm getting quite different results (I would expect some differences, as Rob mentions in the code, data might have changed). I think the main culprit is computing percentage returns - on the bottom of the page 30, Rob mentions that we should divide by capital required from the previous day, but if I use that (instead of just dividing by the capital for the same day), I get very different results. I also see that the chapter1.py code Rob shared also doesn't shift the capital by 1. Results of Rob's code (chapter1.py) when capital is shifted, for SP500 micro (both using MONTH frequency): Code: 'ann_mean': 0.0866116798374384, 'ann_std': 0.1492957335923111, 'sharpe_ratio': 0.5801349961811567 Same with capital not shifted: Code: 'ann_mean': 0.04983102897124981, 'ann_std': 0.1611072019146673, 'sharpe_ratio': 0.309303546824949 Latter corresponds to the values in the book much more closely. I'm sure I'm missing something obvious here.
For those who use the NIFTY futures on the Singapore exchange (SGX): I received an important message from IB about these futures (partial quote):