Specification And Data Issues

Question 1

What is FF misspecification?

Answer 1

When the relationship between dependent and explanatory variables has wrong functional form

Question 2

What does misspecification of FF lead to?

Answer 2

Biased estimators

Question 3

What general misspecification test is used?

Answer 3

RESET: REgression Specification Error Test

Question 4

What must hold for RESET?

Answer 4

MLR 1-4 and we must have a large sample

Question 5

What other misspecification tests are used?

Answer 5

Mizon-Richard
Davidson-Mackinnon

Question 6

What can be said about Mizon-Richard misspecification test?

Answer 6

• not as general as RESET
• to combine models they must have exact same dependent variable
• can only use Ftest to compare after combining as equations are non-nested
• results sometimes unhelpful

Question 7

What are outliers?

Answer 7

Extreme values

Question 8

Why are outliers are potential problem?

Answer 8

Because OLS is based on the squaring of residuals
- outliers will have large residuals and when squares will have a large impact on estimated

Question 9

If outliers are he result of a keying in error what should be done?

Answer 9

Affected observations should be discarded

Question 10

How do we deal with outlier if we choose to exclude it?

Answer 10

Make use of dummy variable:
Set dummy variable = 1 for outliers and 0 otherwise
- the estimates OLS coefficients calculated will exclude the outliers

Question 11

What is coefficient on the dummy variable equal to?

Answer 11

Residual for observation

Question 12

What can be done with the coefficient on the dummy variable?

Answer 12

Can use a t-test with t n-k-2 d.o.f to see if the outlier is significantly far away from regression line
- big significant coefficient seen as evidence of an outlier

Question 13

Why might we exclude a variable?

Answer 13

• excluding outliers can improve our model
• estimates can become significant and hence, overall explanatory power can improve