When would you say an algo strategy , on a liquid instrument, on daily timeframe, would become interesting?
R-squared = 0.1?
0.2?
0.3?
0.4?
0.5 is probably good enough..
R-squared = 0.1?
0.2?
0.3?
0.4?
0.5 is probably good enough..
R-squared of profitable daily algo strategy
- Thread starter damien.dualsigma
- Start date
There isn't a numerical value for R-squared that's "good enough": a strategy with a higher R-squared isn't necessarily better, per se, than one with a lower one, as R-squared doesn't in itself represent the reliability of the strategy, nor even whether you've chosen the right regression. So the question you're apparently asking is a category error, really ... sorry.
R-squared out of sample then?
Or what would you say is a suitable metric and its threshold value out of sample ?
You mean "suitable" to judge whether something's "interesting" (i.e. "worthy of further investigation"), presumably?
For myself, I need nothing more complicated than a Profit Factor of 1.5+ combined with a maximum peak-to-trough drawdown of 5%, monitored over a few hundred consecutive trades, to find it "interesting" ... but someone like @kevinkdog will doubtless be able to give you a far better answer than I can.
In summary, that would pretty much mean a max loss of 5% for an average running gain of 50% of traded value. Definitely interesting for anybody.
May I ask where do you test these metrics? Tradestation or which other setup?
May I ask where do you test these metrics? Tradestation or which other setup?
LOL ?? Economists?
I believe you are thinking of the wrong R-Squared . Check it out.
Anyway, From the top or your wisdom, what sharpe and sortino are you looking for?
I believe you are thinking of the wrong R-Squared . Check it out.
Anyway, From the top or your wisdom, what sharpe and sortino are you looking for?
R-squared is a statistical measure of how close the data are to the fitted regression line. It is also known as the coefficient of determination, or the coefficient of multiple determination for multiple regression. 0% indicates that the model explains none of the variability of the response data around its mean.
- R^2 is a good tool for explaining variance, but not very useful for evaluation of the strategy. You can easily have a strategy that has good predictive power but mostly loses money and vice versa.
- That’s tricky in itself - a PM with a large number of strategies (eg myself) might be happy to add a strategy with a low sharpe and high t-stat. For a stand-alone strategy, it becomes a matter of your required risk metrics, statistical significance and verification window. Once you add anything where PNL is not normally distributed, you have another layer of things to consider.