The model
Yi = b0 + b1X1i + b2X2i + … + bkXki + εi
Uses in finance: explaining returns with several factors, testing whether a variable matters after controlling for others, and forecasting.
Interpretation: b1 is the expected change in Y for a one-unit increase in X1 holding the other independent variables constant. That’s why a coefficient can change sign or size when a new variable is added.
Assumptions
- The relationship is linear in the coefficients.
- The independent variables are not random or are uncorrelated with the error, and no exact linear relationship exists among them.
- The expected value of the error is zero.
- The error variance is constant (homoskedasticity).
- Errors are uncorrelated across observations.
- Errors are normally distributed (for tests in small samples).
Residual plots (residuals against fitted values or against each X) are the first check. The next lessons cover what happens when assumptions fail.
Testing coefficients
t = (b̂j − Bj) ÷ s(b̂j), df = n − k − 1
Reject H0: bj = 0 if |t| exceeds the critical value, or the p-value is below the significance level. A coefficient of 0.85 with standard error 0.25 gives t = 3.40.
A confidence interval for a coefficient: b̂j ± tcritical × s(b̂j).
Predicting
Plug forecast values of each X into the estimated equation. If the model uses X1 = 4%, X2 = 1.5 and coefficients b0 = 0.5, b1 = 1.1, b2 = −0.3: Ŷ = 0.5 + 1.1 × 4 − 0.3 × 1.5 = 4.45.
Two sources of error: the model’s coefficients are estimates, and the error term itself. Forecasts are less reliable when Xs are outside the range used to estimate the model.
The tool shows a single X for clarity; the same logic, minimizing squared residuals, applies with many.
Exam focus
Interpret coefficients as partial effects, compute t-statistics and degrees of freedom, list the assumptions and make a prediction from estimated coefficients.