Quant Flashcards
Heteroskedasticity
variance of the residual (error) term is NOT constant for all observations.
Unconditional Heteroskedasticity
heteroskedasticity is not related to the level of the independent variables, causes no major problems
Conditional Heteroskedasticity
heteroskedasticity IS related to the level of the independent variables, IS a problem, ex. Variance of the residual term increase as the value of the independent variable increases
Effect of Heteroskedasticity & Serial Correlation
SE are unreliable estimates, coefficient not affected, If SE too small, T-stat will be too large and null of no significance is rejected too often, F-test unreliable.
How do you detect Heteroskedasticity?
Breusch-Pagan chi-squared OR examine scatter plot.
How do you correct for Heteroskedasticity?
Robust SE (White-corrected SE) OR generalized least squares
Serial Correlation
residual terms are correlated with one another, common issue with time series data
Positive (Negative) Serial Correlation
Positive regression error in one time period increases the probability of observing a Positive (Negative) regression error for the next time period
How do you detect Serial Correlation ?
Durbin-Watson Stat, DW = 2(1-r), DW < 2 = positively serially correlated, DW > 2 = negatively serially correlated
How do you correct for Serial Correlation?
Hansen Method (Adjust coefficient SE)
Multi-collinearity
two or more independent variables are highly correlated with each other two or more independent variables are highly correlated with each other
Effect of Multi-collinearity
Slope coefficients tend to be unreliable, SE inflated, greater probability that we will incorrectly conclude that a variable is not stat. sig (type II error)
How do you detect Multi-collinearity?
high r^2, sig F-test, but no independent variables are stat sig, suggests the variables together explain the variation but the independent variables do not
How do you correct for Multi-collinearity?
Omit 1 or more of the correlated independent variables
