Assumptions of OLS

All of these assumptions about the data should hold true to reap the benefits of the OLS regression techniques:

• Linearity: The true underlying relationship between X and Y is linear.
• Homoscedastic: The variance of residuals must be constant. The residual is the difference between the observed value and predictive value of the target.
• Normality: The residuals/errors should be normally distributed.
• No or little multicollinearity: The residuals/errors must be independent.

OLS is also affected by the presence of outliers in the data. Outlier treatment is necessary before one proceeds with linear regression modeling using OLS linear regression.