Why R-Squared Alone Cannot Rank Trading Predictors

Summary

The document asks how to compare two simple regression-based trading models when the model with higher R-squared has weaker backtest results. It describes one predictor with a higher in-sample fit measure and average returns, and another with lower fit but better returns. The author also asks whether a particular R-squared threshold should be considered adequate.

No answer or performance evidence beyond these stated comparisons is provided. The question recognizes that R-squared is not a sufficient strategy-quality measure: explanatory fit does not directly establish profitability, robustness, or tradability. Model assessment would also need to consider out-of-sample returns, risk, transaction costs, and stability, none of which are reported here. The document does not establish which model is preferable or a universal target R-squared.

Key ideas

• A higher regression R-squared does not guarantee better trading results.
• The question contrasts a better-fitting model with a model reporting stronger backtest returns.
• The document gives no universal R-squared threshold or recommendation for choosing between them.
• Risk, costs, robustness, and out-of-sample evidence are not discussed in the source.

Tags

# R squared of a good Trading strategy


# R squared of a good Trading strategy












What would you consider a decent R square value of a good trading strategy.

I know R squared is not a good metric for judging trading strategies but still at an initial stage how do you decide to keep a predictor variable if r squared of the model is good but trading strategy results are poor.

More specifically , I have 2 trading models with 2 different predictive variables one have a R square of 5% and average backtesting return. The second have a R square of 1% and good backtesting returns.(I have assumed the simplest regression model for both with no parameter fitting other than the coefficients themselves.)

So, my question is

- which model is better?

- Is R squared of 5% decent value or should it be atleast 20-30%.

Please provide any other inputs regarding the metrics if you can.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.