FRM Exam Part I · Regression with Multiple Explanatory Variables

Dummy Variable Regression and Model Specification

Updated 11 October 2026 · Fact-checked

A dummy variable is a 0/1 indicator that places an observation in a category. In multiple regression, its coefficient is the average shift in the dependent variable versus the omitted base category, holding other variables fixed. Use k − 1 dummies for k categories to avoid the dummy variable trap.

Understand Dummy Variables and Model Specification

A regression usually uses numbers such as size, leverage or GDP growth. Many drivers of risk are categories instead: a bank is or is not a G-SIB, a month is or is not January, a year is or is not a crisis year. A dummy variable (indicator variable) turns a category into a number. It equals 1 if the observation is in the category and 0 otherwise.

Suppose you model a fund's monthly return as Y = β0 + β1 × D + ε, where D = 1 for crisis months and 0 otherwise. When D = 0, the expected return is β0. When D = 1, it is β0 + β1. So β0 is the average for the base (omitted) category, and β1 is the difference in average Y between crisis and non-crisis months. A dummy shifts the intercept. It does not change the slopes.

If you have k categories (for example four quarters), include only k − 1 dummies when the model also has an intercept. If you include all k, the dummies add up to the intercept column of ones. This is perfect multicollinearity, known as the dummy variable trap, and OLS cannot be computed. Each dummy coefficient is then read against the omitted category, so the choice of base changes the coefficients but not the fitted values.

An interaction term multiplies a dummy by a continuous variable, for example D × X. It lets the slope differ between groups. In Y = β0 + β1 X + β2 D + β3 (D × X) + ε, the slope on X is β1 when D = 0 and β1 + β3 when D = 1. The intercept is β0 when D = 0 and β0 + β2 when D = 1. A t-test on β3 tests whether the slopes differ. An F-test on β2 and β3 together tests whether the two groups follow the same line.

Model specification is about choosing the right variables and form. Omitting a relevant variable that is correlated with an included one biases the included coefficients (omitted variable bias). Adding irrelevant variables does not bias estimates, but it raises their standard errors and lowers adjusted R². Adding variables always raises R², so use adjusted R², t-tests, F-tests and information criteria (AIC, BIC, which penalise extra parameters) to compare models. Too many variables risks overfitting: a good in-sample fit that fails out of sample.

Key formulas to remember

Dummy regression (intercept shift): Y = β0 + β1 X + β2 D + ε; Base group (D = 0): intercept β0. Dummy group (D = 1): intercept β0 + β2. Slope on X is β1 for both.
Interaction model: Y = β0 + β1 X + β2 D + β3 (D × X) + ε; Slope is β1 when D = 0 and β1 + β3 when D = 1. Intercept is β0 when D = 0 and β0 + β2 when D = 1.
Number of dummies: k categories → k − 1 dummies (with an intercept); Using k dummies plus an intercept gives perfect multicollinearity (dummy variable trap).
Test of a single coefficient: t = (β̂ − hypothesized value) ÷ SE(β̂); Degrees of freedom are n − k − 1, where k is the number of slope coefficients.
Joint test (restricted vs unrestricted): F = [(SSR_restricted − SSR_unrestricted) ÷ q] ÷ [SSR_unrestricted ÷ (n − k − 1)]; q is the number of restrictions, k the number of slope coefficients in the unrestricted model.
Adjusted R²: Adj R² = 1 − (1 − R²) × (n − 1) ÷ (n − k − 1); Can fall when a variable adds little. Unlike R², it penalises extra regressors.
Information criteria: AIC = ln(SSR ÷ n) + 2k ÷ n; BIC = ln(SSR ÷ n) + k ln(n) ÷ n; Lower is better. BIC penalises extra parameters more heavily than AIC when n is 8 or more.

How to solve Dummy Variables and Model Specification questions

Use this routine for any question on dummies, interactions or specification.

1Identify the base category: the one with no dummy, or with D = 0.
2Write the fitted equation for each group by substituting D = 0 and D = 1.
3Read the dummy coefficient as the intercept difference from the base group, holding the other variables fixed.
4If there is an interaction term, read its coefficient as the difference in slope between groups. Add it to the base slope for the dummy group.
5For significance, compute t = coefficient ÷ standard error and compare with the critical value. Use an F-test if several coefficients are tested together.
6For trap questions, count the categories and dummies. With an intercept, k categories need k − 1 dummies.
7For model choice, check bias first (omitted variable), then efficiency (irrelevant variable), then compare adjusted R², AIC or BIC.
8Check units and the sign, and state the answer in the context of the question.

Quickest way: Plug in 0 and 1

When to use it: Use it for any interpretation or prediction question that gives a fitted equation with a dummy.

Write the equation and set D = 0 to get the base line.
Set D = 1 and collect terms. The change in intercept is the dummy coefficient; the change in slope is the interaction coefficient.
Insert the given X value and compute the prediction.
For a dummy-trap question, just count: dummies equal to the number of categories plus an intercept means invalid.
For model-choice questions, remember that R² always rises with more variables, while adjusted R² and BIC can fall.

Common mistakes in Dummy Variables and Model Specification

Including a dummy for every category along with an intercept.
It feels complete to give each category its own variable.
Fix: Drop one category as the base, or drop the intercept. With an intercept, use k − 1 dummies.
Interpreting a dummy coefficient as the average level of that group.
Students forget that the intercept holds the base group's level.
Fix: The coefficient is the difference from the base group. The group's level is β0 plus the coefficient, with other variables held fixed.
Saying a dummy changes the slope.
Intercept shifts and slope changes are confused.
Fix: A plain dummy shifts only the intercept. A slope change needs an interaction term D × X.
Choosing the model with the highest R².
R² looks like a measure of quality.
Fix: R² never falls when you add a variable. Compare adjusted R², AIC or BIC, and check that the signs and significance make sense.
Believing that omitting a relevant variable and adding an irrelevant one have the same effect.
Both are seen as specification errors.
Fix: Omitting a relevant variable correlated with included ones biases the coefficients. Adding an irrelevant one leaves OLS unbiased but increases variance.
Using the interaction coefficient alone as the slope for the dummy group.
The base slope is forgotten.
Fix: The dummy group's slope is β1 + β3. The coefficient β3 is only the difference.

Worked examples

Example 1

A regression of monthly excess fund return (%) on market excess return (X) and a crisis dummy D (1 in crisis months, 0 otherwise) gives: Y = 0.40 + 0.90 X − 1.50 D. What is the predicted return in a crisis month when X = 4%?

Show the solution

Set D = 1: Y = 0.40 + 0.90 X − 1.50 = −1.10 + 0.90 X.
Insert X = 4: 0.90 × 4 = 3.60.
Y = −1.10 + 3.60 = 2.50.
Check: in a non-crisis month with X = 4, Y = 0.40 + 3.60 = 4.00. The crisis month is 1.50 lower, matching the dummy coefficient.

Answer: Predicted excess return is 2.50%.

Example 2

A model of bond spread (bps) is: Spread = 80 + 20 × Lev + 30 × D + 10 × (D × Lev), where Lev is leverage ratio and D = 1 for financial issuers, 0 for non-financial issuers. What is the predicted spread for a financial issuer with Lev = 3?

Show the solution

Set D = 1: Spread = 80 + 30 + (20 + 10) × Lev = 110 + 30 × Lev.
Insert Lev = 3: 30 × 3 = 90.
Spread = 110 + 90 = 200.
Check using the original equation: 80 + 20×3 + 30 + 10×3 = 80 + 60 + 30 + 30 = 200.

Answer: The predicted spread is 200 bps.

Exam tips

For dummy questions, always substitute D = 0 and D = 1 and compare the two lines. It is quick and catches most traps.
Watch for the phrase 'holding other variables constant'. The dummy coefficient is a ceteris paribus difference from the base group.
When a question mentions four quarters or twelve months with an intercept, count the dummies. The valid number is k − 1.
Know which direction each specification error pushes: omitted relevant variable gives bias, irrelevant variable gives higher variance.
For model selection, recall that R² always rises with added variables, while adjusted R², AIC and BIC can penalise them.

Practice questions from Regression with Multiple Explanatory Variables

Dummy Variables and Model Specification in other exams

The same ground in other exams, if you are preparing for more than one or want another angle on it.

CFA Level IIDummy Variables in Multiple Regression

Dummy Variables and Model Specification: frequently asked questions

What is the dummy variable trap?

It is perfect multicollinearity caused by including a dummy for every category along with an intercept. The dummies sum to the column of ones, so OLS cannot separate them from the intercept. Drop one dummy to fix it.

How do I interpret a dummy variable coefficient?

It is the expected difference in the dependent variable between the dummy group and the omitted base group, holding the other regressors fixed. A coefficient of −1.5 means the group averages 1.5 units lower than the base.

What does an interaction term add to a regression?

It lets the slope on a variable differ between groups. The coefficient on D × X is the difference in slope between the D = 1 and D = 0 groups. Test it with a t-test.

How does omitted variable bias differ from including an irrelevant variable?

Omitting a relevant variable that is correlated with the included regressors biases their coefficients and makes OLS inconsistent. Including an irrelevant variable keeps estimates unbiased but raises standard errors and lowers precision.