Set out which investment questions multiple linear regression can answer and the steps involved in running one

Analysts reach for multiple regression for three jobs: to find which variables are related to an outcome, to forecast it, and to test whether an existing theory, such as a factor pricing model, fits the data. The process runs in order. Frame the question, choose the dependent and independent variables, then ask whether the dependent variable is continuous; if it is a yes-or-no outcome, logistic regression fits better than ordinary least squares. Estimate the model, then examine the residuals to check the assumptions before trusting any test, and only then use the model. The trap is jumping straight from estimation to conclusions, or reading a tested theory as a forecast.