ECON 104 Lecture Notes - Lecture 35: Multicollinearity, Panel Data, Heteroscedasticity
Document Summary
Extending original ols assumptions towards new panel data framework, ols estimators are normally distributed, unbiased. Pick state but collecting data from 82 -> 88. When state has more than average fatalities, it probably has more than average fatalities in the following year too - unlikely to be completely independent from previous year"s results. Assumptions are restated to fit into panel structure. 3rd and 4th moments and multicollinearity is idneitcal to ols case. First assumption = need expectation of error terms to be independent of regressors and obviously for the fixed effect. You want it also to be independent of x"s in other time periods. X must be independent of x"s for that variable and of different time periods. There is no omitted lag effects - today"s x does not impact tomorrow"s error term, otherwise add today in tomorrow"s regression control - otherwise there will be ovb. Need today"s error term to not affect tomorrow"s error term.
