In the SLRM we assume random sampling and mean independence. Typically random sampling is a reasonable assumption in cross-sectional data. However, random sampling is not a reasonable assumption in time series data. Suppose you have data on GDP growth (gGDP) and interest rates (int) for three years: gGDP1990 = Bo + Bint1990 + 21990 gGDP1391 = Bo + Biint1991 + 21991 gGDP1932 = Bo + Biint1992 + 21932 For our OLS estimate of , to be unbiased we need E(u, lint1990, int1991, int1992) =0 fort = {1990, 1991, 1992}. Explain why if the government chooses ant1991, the interest rate in 1991, based on the GDP growth in 1990, then knowing ant1991 is informative in predicting 1990 and OLS will be biased. (Note: typically the government increases interest rates if GDP growth is higher than some baseline growth rate to prevent overheating") Solution