Page 22 - FINAL CFA II SLIDES JUNE 2019 DAY 3

LOS 8.k: Explain the types of heteroskedasticity and how READING 8: MULTIPLE REGRESSION AND ISSUES IN REGRESSION ANALYSIS
heteroskedasticity and serial correlation affect statistical
inference.
MODULE 8.6: ASSUMPTIONS: HETEROSKEDASTICITY

WHAT IS HETEROSKEDASTICITY?

When variance of the residuals is NOT constant (some subsamples are more spread out than the rest of the sample) – 2 Types:

1. Unconditional (UH) : Variance doesn’t systematically change with changes in the value of the independent variable(s) (No Big deal!)
2. Conditional (CH) : Variance systematically changes ‘positively’ (in direct proportion) to changes in the independent variable.

Occurs when larger values X,
create greater dispersion
from best fit line!

Small SE, t-statistics too large:
• You risk Rejecting Ho (when you shouldn’t)
–Type 1 Error
• The opposite is true – Type 2 Error!

4 Effects on Regression Analysis:

• Unreliable Standard Error (SE) estimates The F-test is also unreliable: Why?
• No effect on coefficient estimates For ‘single variable’ regression, F = t squared –recall?

So what? For multiple regression, F = MSR/MSE (what’s MSE? Get it?)