Coefficient of Determination

The coefficient of determination, almost always written as R-squared (R²), is a statistic that measures how well a regression model fits the data. It tells you the proportion of the variation in an outcome that the model explains, and it's one of the most widely quoted figures in statistics and finance. This guide explains what the coefficient of determination is, how to interpret it, its limitations, and why it matters — in plain language. It connects to correlation and adjusted R-squared, and is a core topic in quantitative qualifications like the FRM.

What is the coefficient of determination?

The coefficient of determination measures the proportion of the variation in the dependent variable (the outcome) that is explained by the independent variable(s) in a model. It's expressed on a scale from 0 to 1 (or 0% to 100%). An R² of 0.70, for example, means that 70% of the variation in the outcome is explained by the model, leaving 30% unexplained — due to other factors or random noise. In short, it answers the question: "how much of what's happening does my model actually account for?"

How to interpret R-squared

Reading R² is about where it sits on the 0-to-1 scale:

• Close to 1: the model explains most of the variation — a strong fit, where the independent variables account for nearly all of the movement in the outcome.
• Close to 0: the model explains very little of the variation — a poor fit, where the variables tell you almost nothing about the outcome.
• In between: a partial fit — the model explains some, but not all, of what's going on.

It's worth noting that R² is the square of the correlation coefficient in a simple (single-variable) regression — which is where the "R" comes from. What counts as a "good" R² depends heavily on context: in some fields a value of 0.9 is expected, while in finance — where outcomes are noisy and hard to predict — a much lower R² can still be meaningful.

The limitations of R-squared

R² is useful but easy to misuse, and it has important limitations:

• It doesn't prove causation. A high R² shows the variables move together in the model, but not that one causes the other.
• It always rises with more variables. Adding any variable — even a useless one — never decreases R², which can make a bloated model look better than it is. This is exactly why adjusted R-squared, which penalises unnecessary variables, is used to compare models.
• A high R² isn't always good, and a low one isn't always bad. A very high R² can signal overfitting, while a low R² may still be useful for inherently unpredictable outcomes.
• It says nothing about whether the model is appropriate. A model can have a decent R² yet still be misspecified.

Why it matters in finance

The coefficient of determination is used throughout finance to judge how well models explain things — how much of a stock's movement is explained by the market, how well a factor model fits returns, or how much of a forecast's variation a set of drivers accounts for. It gives a quick, standardised sense of explanatory power. But its limitations mean it should never be used in isolation: a sensible analyst looks at R² alongside adjusted R², the significance of individual variables, and whether the model makes economic sense.

Why it matters for finance professionals

For anyone building or interpreting regression models, the coefficient of determination is essential to understand — both what it tells you and what it doesn't. Knowing how to read R², and being alert to its limitations, is fundamental to sound quantitative analysis and a regularly examined topic in professional qualifications.

Frequently asked questions

What is the coefficient of determination?

A statistic, written R-squared (R²), measuring the proportion of the variation in an outcome that a regression model explains, on a scale from 0 to 1. An R² of 0.70 means the model explains 70% of the variation.

How do you interpret R-squared?

Closer to 1 means the model explains most of the variation (a strong fit); closer to 0 means it explains little (a poor fit). What counts as "good" depends on context — finance often accepts lower values than other fields.

Does a high R-squared mean the model is good?

Not necessarily. R² always rises when you add variables, can signal overfitting if very high, and doesn't prove causation or that the model is appropriate. It should be read alongside other measures.

What's the difference between R-squared and adjusted R-squared?

R² never decreases when you add variables; adjusted R-squared penalises unnecessary variables, so it can fall if an added variable doesn't genuinely improve the model — making it better for comparing models.

Build your quant skills with Learnsignal

The coefficient of determination is a cornerstone of regression analysis. Learnsignal's tutor-led courses, including the FRM, develop the statistical understanding that topics like this build on — with clear teaching that makes the methods genuinely usable.