What Is R-Squared in Investing? How Much of a Portfolio the Benchmark Explains

R-squared measures how much of a portfolio’s return movement is explained by movements in its benchmark. It runs from zero to one, and it is the number that decides whether the other benchmark-relative statistics, beta, alpha, tracking error and the ratios built on them, are describing something real or are fitting noise.

What it measures

Imagine plotting a portfolio’s periodic returns against the benchmark’s returns for the same periods, one dot per period. A straight line fitted through those dots is a simple regression. It produces two familiar outputs, a slope and an intercept, plus one that gets less attention.

The slope is beta: how far the portfolio tended to move for each unit the benchmark moved. The intercept is the return not explained by that relationship, which is the basis of alpha. And R-squared is the share of the portfolio’s total return variation that the fitted line accounts for.

Stated in words rather than symbols: R-squared is the fraction of the portfolio’s ups and downs that lines up with the benchmark’s ups and downs. If it is 0.85, then eighty five percent of the variation in the portfolio’s returns moved in step with the benchmark, and fifteen percent came from something the benchmark does not capture.

There is a direct link to correlation. R-squared, in the single-benchmark case, is exactly the correlation between portfolio and benchmark returns, squared. That squaring is why the two numbers feel different. A correlation of 0.7 sounds like a strong relationship, but it corresponds to an R-squared of 0.49, meaning less than half the movement is explained. Correlation flatters, R-squared does not.

Because it is a squared quantity, R-squared is always between zero and one and carries no sign. It cannot tell you whether the portfolio moves with or against the benchmark. That direction lives in beta and in the correlation itself.

How to read it

Read it before beta and alpha, not after. This is the practical rule that most justifies knowing the statistic. Beta and alpha are outputs of a fitted line, and R-squared tells you how well that line fits. When R-squared is high, the beta estimate is describing a genuine relationship and alpha is a meaningful residual. When R-squared is low, the line is being fitted through a cloud of dots that barely form a pattern, and both beta and alpha carry wide uncertainty. Quoting a precise alpha off a poorly fitting regression is the single most common misuse of this family of numbers.

A high value is a statement about behaviour, not about quality. An R-squared close to one says the portfolio’s movement has been benchmark movement. It says nothing about whether that was good. A portfolio can track a benchmark closely and lag it consistently, or track it closely and lead it. Level of explanation and level of outcome are different axes.

A low value asks a question about the benchmark. If a portfolio shows low R-squared against the index it is measured against, there are two possibilities. Either the portfolio genuinely does something the index does not, or the benchmark is simply the wrong comparator. A mid-cap-tilted portfolio measured against a large-cap index will show a lower R-squared for reasons that have nothing to do with the manager. Choosing a fair comparator, as discussed in benchmark selection for portfolios, is upstream of every benchmark-relative statistic.

Read it alongside tracking error, because they answer different questions. R-squared asks what share of movement the benchmark explains. Tracking error asks how large the unexplained differences have been in absolute terms. A portfolio can have high R-squared and meaningful tracking error at the same time, if it moves with the benchmark but amplifies it. The combination of the two is more informative than either.

Watch it through time, not once. R-squared computed over a fixed window blends whatever conditions fell inside that window. Rolling the calculation over a moving window often shows the relationship strengthening in stressed periods, when broad market moves dominate everything, and weakening in calm periods, when company-specific factors have room to matter. That pattern is the same mechanism that drives correlations upward in a crisis, described in the correlation matrix.

Mind the frequency. Daily, weekly and monthly returns can produce noticeably different R-squared values for the same portfolio and benchmark. Higher frequency picks up short-term noise and timing mismatches. Lower frequency smooths them but leaves fewer observations. Neither is canonical, so comparisons need matching frequency and matching window.

What it does not tell you

It does not measure performance. R-squared has no view on whether the portfolio made or lost money, or whether it beat the benchmark. It only reports how much of the variation was shared. A portfolio with an R-squared of 0.95 could have led or lagged its benchmark substantially over the same period.

It does not prove causation or holdings overlap. A high R-squared does not mean the portfolio holds the benchmark’s constituents. Two very different holdings lists can move together because they share common drivers. Equally, a portfolio holding many index members can show a lower R-squared if its weights differ sharply from index weights.

It only sees linear relationships. Like correlation, R-squared measures how well a straight line fits. A portfolio with an asymmetric relationship to the benchmark, participating in rallies differently from falls, is poorly summarised by one line and one R-squared. That asymmetry is exactly what upside and downside capture ratios are built to expose.

It is an estimate, with error attached. Computed from a finite sample, R-squared moves around. A short window with few observations can produce a value that looks precise and is not. The number of periods used should always accompany the figure.

It does not tell you which benchmark is right. R-squared can be computed against any series. Running a portfolio against several candidate benchmarks and picking the one that produces the most convenient result is an easy and quiet form of self-deception. The benchmark should be chosen for its relevance to the mandate first, then measured.

It does not decompose the unexplained part. A portfolio with an R-squared of 0.6 has forty percent of its variation coming from somewhere else. R-squared does not say where. That could be sector tilts, size tilts, style exposures, single-stock outcomes, or cash holdings. Answering it requires a multi-factor view rather than a single-benchmark one, which is the domain of factor exposure analysis.

It can be raised or lowered without any change in skill. Holding more cash lowers beta and can change R-squared. Adding index-like positions raises it. Neither adjustment reflects insight, yet both move the statistic.

Putting it in its place

R-squared is best treated as a gatekeeper rather than a headline. Before reading any benchmark-relative number, check whether the benchmark explains enough of the portfolio’s movement for that number to mean something. If it does, beta, alpha, tracking error and the ratios built on them can be read with reasonable confidence. If it does not, the honest conclusion is that the portfolio is doing something the benchmark does not describe, and the right response is to find a better comparator or to lean on measures that do not depend on one.

Related reading

• Portfolio Metrics Explained: the hub for portfolio-level statistics.
• What Is Beta in Investing?: the sensitivity estimate that R-squared validates.
• What Is Tracking Error?: the size of the benchmark-relative differences.
• Upside and Downside Capture Ratio: the asymmetry a single R-squared cannot see.
• Correlation Matrix in Portfolios: correlation before it is squared.

This article is educational. Altys Labs is not a registered research analyst or investment adviser, and nothing here is investment advice or a recommendation to buy, sell, or hold any security.