Direct answer
R Squared (R²) is a statistical metric that describes how closely data follow a chosen model, most commonly a linear relationship. It is a “fit” measure: it summarizes how much of the variation in a target variable your model accounts for, given the specific way you defined the inputs and fitted the model.
In forex research, people sometimes calculate R² to judge whether a proposed linear relationship between two quantities appears to track historical movements. R² itself does not indicate future price direction, does not control for trading costs, and cannot ensure that the relationship will remain stable.
Definition and simple model
In a basic linear model, you relate a dependent variable (often written y) to one or more independent variables (often written x). For a single-feature linear regression, the model takes a form like:
- y ≈ a + b·x
R² measures the fraction of total variation in y explained by the fitted model. Conceptually:
- If your model explains little of y’s variability, R² is low.
- If your model explains most of y’s variability, R² is high.
A key limitation is that R² is only meaningful relative to the exact modeling setup: which variables you used, whether you used levels or returns, how you handled missing values, and whether you fitted on all available data or used a training/validation split.
How it can be used in forex (and what to be careful about)
A common forex use is to compare how strongly historical changes in one quantity relate to changes in another under a linear assumption, by computing R² for that relationship.
Example with clear assumptions (no live prices):
- Assume you choose a y series and an x series that are aligned by time.
- Assume you fit a linear regression y ≈ a + b·x using historical observations.
- Assume you compute R² from the fitted line.
Interpreting the result:
- A higher in-sample R² means the fitted line tracks the chosen y values better on the data used to fit.
- A lower R² means the relationship is weaker under that linear form.
Material limitations and failure modes:
- Overfitting to historical noise: A relationship can look strong in one period and collapse later.
- Regime changes: Market behavior can shift, changing the underlying relationship.
- Different costs and execution effects: Even if y is “explained” by x historically, real trading outcomes depend on spreads, slippage, and latency.
- Non-stationarity: Statistical properties may change over time, making “explained variation” unstable.
- Model form dependence: R² reflects the fit of the specific model family (e.g., linear). A different functional form may fit differently.
Because of these issues, historical R² alone is not a reliable guarantee of future usefulness.
Limitations and risks, plus how to verify
R² is a descriptive statistic. It does not confirm causality, does not validate that a relationship will generalize, and does not translate directly into trading performance.
Verification checklist you can independently apply:
- Use out-of-sample evaluation: Compute R² on data not used for fitting.
- Test sensitivity to inputs: Recompute R² when changing preprocessing (e.g., returns vs. levels) and alignment choices.
- Check stability across time windows: Compare R² across different historical segments to see whether it holds.
- Consider whether linear assumptions fit the question: If the relationship is clearly nonlinear, a linear R² may mislead.
Next question to explore: When you compute R² for your forex-related variables, what exactly are your x and y definitions, your time alignment rule, and your evaluation method (in-sample vs. out-of-sample)? Those choices often determine what R² is actually measuring.