What it means
Regression fits a relationship between an outcome and one or more predictors. For a fund, an analyst might regress monthly returns on monthly benchmark returns, and R-squared summarises how much of the sample outcome's variation the fitted model accounts for.
The usual ordinary-least-squares calculation compares squared residuals with total squared deviations from the sample mean, so a model that substantially reduces squared errors relative to predicting the mean has a higher R-squared. Penn State's statistics course shows R-squared as regression sum of squares divided by total sum of squares, equivalently one minus error sum of squares divided by total sum of squares for its standard setup.
The baseline and assumptions matter. An R-squared of 0.80 in an appropriate in-sample model means 80% of the observed variation is accounted for relative to that baseline, not that 80% of every price move was caused by the benchmark.
Investopedia describes the measure using a stock and an index, but uses the word influence too strongly. Statistical association alone cannot prove that index moves cause the stock's moves, because shared market conditions may move both.
R-squared and beta also answer different questions: beta describes the estimated slope or sensitivity to the benchmark, while R-squared says how well that linear fit accounts for variation, so a fund can have a high beta with modest fit. A high value does not tell whether a fund produced a positive return, because a fund that closely tracks a falling index can have a strong fit and still lose money.
For an index fund, a high historical R-squared may support the observation that returns moved with its benchmark. Differences in fees, holdings or timing can still matter, and the specific benchmark must match the fund's mandate.
The value changes with the sample period, so a relationship estimated during calm years may weaken in a crisis. Report the frequency, date window and benchmark when comparing figures from two providers.
An unusually large observation can also change a fitted slope and R-squared, so inspect a scatterplot instead of relying only on one summary number. A low linear R-squared does not rule out a curved relationship, since the predictors may need a different form or important factors may be missing.
A high R-squared does not guarantee that the chosen form fits every part of the data well, and some models fitted without an intercept or evaluated on new data can yield values outside the familiar zero-to-one range. Compare specified models on the same outcome and period, then check residuals, because the statistic is not a trading instruction.
In practice
Real-world examples.
Example
A fund's monthly returns have R-squared of 0.90 against its stated benchmark over a chosen three-year sample.
Example
A stock has a high beta but low R-squared against an index, indicating large estimated sensitivity yet much unexplained variation.
Example
Changing the benchmark from global equities to a sector index changes the fit, so the analyst reports which index was used.
Formula
Calculation
For ordinary least squares with an intercept, R-squared = 1 - SSE/SST, where SSE is the sum of squared residuals and SST is total squared deviation from the outcome's sample mean. If SST is 100 and SSE is 25, R-squared is 1 - 25/100 = 0.75 or 75%. This is an in-sample fit measure, not a 75% chance of a forecast being right.
The same figures can be read the other way: the regression sum of squares is 100 - 25 = 75, and 75/100 gives the same 0.75. If a second benchmark left SSE at 60, its R-squared would be 1 - 60/100 = 0.40, so the first benchmark describes this fund's sample returns much better.Case study
Seen in the real world.
Fictional example: Amira compares a global equity fund with a broad index using monthly returns. A regression reports R-squared 0.82 and beta 1.1 over four years. She records the date range and calculates that the model still leaves a portion of return variation unexplained in that sample. She then inspects the return scatterplot and sees one unusual crisis month. Rerunning the analysis without hiding that point changes the statistics.
Amira reviews holdings, fees and the fund mandate before deciding whether the index is a fair benchmark; she does not present 0.82 as proof of future performance or causation. Her note to the investment committee lists the benchmark, the four-year window and the monthly frequency next to the 0.82 figure. The committee can then compare it with a provider's number on the same basis. Amira and the fund are invented for illustration.
Watch out
Common mistakes.
- Interpreting R-squared as a probability of making money or forecasting correctly.
- Calling a high fit proof that an index caused the fund's returns.
- Comparing two R-squared figures without matching variables, sample periods and model specification.
Questions
People also ask.
Is R-squared the same as beta?
No. Beta estimates a slope or sensitivity; R-squared summarizes in-sample fit against a baseline.
Can a high value mean the investment is good?
No. A fund can closely track a declining benchmark and still lose money.
Is it always between zero and one?
That range describes the standard in-sample least-squares model with an intercept; other fitting or evaluation setups can differ.
From the founder's library
Take it further with the book.
Build your financial confidence beyond this definition. Shihan's full-length guide, Accounting Fundamentals, takes the same plain-English approach and turns it into a complete, practical playbook for non-finance managers, business owners and students - with chapter-end quiz answers and presentation slides included.
25% off with code MMHQ25, applied at checkout. Priced in USD - checkout may show the equivalent in your local currency.
View the book and save 25%Related