Page 33 - FINAL CFA II SLIDES JUNE 2019 DAY 3

Misspecification #3: Incorrectly Pooling Data READING 8: MULTIPLE REGRESSION AND ISSUES IN REGRESSION ANALYSIS

R = b + b B + b lnM + b lnPB + b FF + ε MODULE 8.9: MODEL MISSPECIFICATION, & QUALITATIVE DEPENDENT VARIABLES
0
3
2
4
1
Regression coefficients being different from one period to the next, but we pool and develop one regression model over the
entire period, instead of 2 separate ones over each subperiod; tests and predictions then become misleading.

Misspecification #4: Using a Lagged Dependent Variable as an Independent Variable

A lagged variable comes from a prior period: R = d + d B + d lnM + d lnPB + d FF + d R t – 1 + ε
4
3
5
0
2
1
If ε is serially correlated due to the inclusion R t – 1 we get biased & inconsistent estimates, hence unreliable tests & return predictions.
Misspecification #5: Forecasting the Past

Using variables measured at the end of say July to predict a variable measured during July for a future period!.

Misspecification #6: Measuring Independent Variables with Error

Free Float above is rather a ‘proxy’ for corporate governance quality and portfolio returns (as we can’t easily measure
“corporate governance quality,”). The higher the level of FF, the more influence the capital markets have on management’s
decision making process and thus, the more effective the corporate governance structure (hence P/E). As FF is a proxy only, it
will have error, causing biased and inconsistent estimates, hence our testing and predictions will be unreliable.

Another common example when an independent variable is measured with error is when we want to use expected inflation in
our regression but use actual inflation as a proxy.