Page 30 - FINAL CFA II SLIDES JUNE 2019 DAY 3

READING 8: MULTIPLE REGRESSION AND ISSUES IN REGRESSION ANALYSIS
Correcting Multicollinearity

MODULE 8.8: MULTICOLLINEARITY

• Omit one or more of the correlated independent variables (hard to identify the source one though; you can try statistical procedures
like stepwise regression, which systematically removes variables from the regression until multicollinearity is minimized).

• Regression model specification: Select the explanatory (independent) variables to be included in the regression and the
transformations, if any, of those explanatory variables:

Say we wish to predict P/E ratio based on, say, say the stock’s dividend payout ratio (DPO), growth rate (G), and beta (B), and we
know the 3 exhibit multicolliearity:

Specification 1: P/E = b + b DPO + b G + b B + ε
3
2
1
0
If say market capitalization (M) is related to P/E ratio, we create a second specification by including M as an independent variable:

Specification 2: P/E = a + a DPO + a G + a B + a M + ε
1
2
3
0
4
Finally, suppose we conclude that M is not linearly related to P/E, but the natural log of M is linearly related to P/E. Then, we would
transform M by taking its natural log and creating a new variable lnM. Thus, our third specification would be:
Specification 3: P/E = c + c DPO + c G + c B + c lnM + ε
4
1
0
3
2
NOTE: When we change the specifications of the model, the regression parameters change!