Interpreting Predictive Forecast R-Squared and Lookback Windows
Summary
The document considers an online indicator labeled as a predictive forecast with an R-squared value and a parenthetical period. It asks whether the period describes historical observations or a future forecast horizon, what relationship the statistic measures, and whether the value conveys price direction. The response suggests that the period likely refers to data used to calibrate a model, such as a rolling sample, but stresses that the label alone is insufficient to identify the model or its inputs.
R-squared is described as the share of variation in a dependent variable accounted for by independent variables within the observations used to fit a model. It therefore does not, by itself, establish how much of the next day’s price movement can be predicted. Nor does R-squared encode whether prices rise or fall: it is a fit statistic, and the direction would depend on the model’s coefficients or forecast output. The answer is tentative and recommends obtaining the model or data source before interpreting the particular indicator; it notes that returns are often modeled instead of raw prices.
Key ideas
- R-squared describes in-sample variation explained by a fitted model, not necessarily future predictive accuracy.
- A period in an indicator label may refer to the model’s calibration window, but the label alone cannot confirm this.
- R-squared does not reveal whether the predicted price movement is upward or downward.
- Interpretation requires knowing the dependent variable, predictors, fitting sample, and model specification.
- Returns are commonly used in regressions instead of price levels.
Tags
Full text
# Predictive Forecast (Close, 14) # Predictive Forecast (Close, 14) I've been following an asset wherein a "R-squared predictive forecast (close, 14)" is posted online each day. On some days, this figure is extremely high, like .92. - Exactly what is the significance of the "14?" Does it refer to the past 14 days? The next 14 days? Something else? - How do I interpret the R-squared predictive forecast? For example, does it mean that 92% of the variance in the next day's closing price can be explained by the current day's change between closing and opening values? Or maybe it means that the change between the next day's closing and opening values can be explained by the change between the current day's closing and opening values? - Also, how do we account for direction (price going up versus going down)? Let's look at an example: Let's say the current day's R-squared predictive forecast (close, 14) is .80. Couldn't this value of.8 describe a scenario wherein the current day's open was \$20 and the close was \$19 (a price decrease) as well as a scenario wherein the current day's open was $20 and the close was about \$21 (a price increase)? Thank you! ## Answer by KaiSqDist (score 0) https://quant.stackexchange.com/a/79007 Welcome to the forum. - Sounds more like the past 14 days are used in calibrating a model to obtain the R^2 value. - I don't think the R^2 ever refers to the next day. The definition of R^2 is the proportion of variance in the dependent variable that is explainable by the independent variable for the set of observed data that is used to calibrate the model. For example, this could be the historical time series used in calibrating a linear regression model. - Not sure what the model to obtain this R^2 predictive forecast looks like but usually the returns instead of the price is used in the regression. Maybe you can provide more details or even share the link to the time series you are observing?
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.