I am looking for some clarification regarding when multicollinearity (btw i hate spelling that word) is a problem. I understand the, not individual significant but collectively significant t-test/ F-test logic, but I just encountered a QBANK question that only gave info on independant variable correlation coefficients (i.e. no info to calculate an F-test). The highest magnitude rho was 0.43 and the answer implied that due to the high rho of 0.43 multicollinearity was a problem.
Not sure if the QBANK question was simply bad, but is a rho of 0.43 indicative of multicollineraity (strikes me as a bit too low to indicate existence of a linear relationship)?
Mike作者: ayodayo 时间: 2011-7-11 19:54
can you please post the entire question... in absence of more info... what do u expect us to respond?
CP作者: bbtomato 时间: 2011-7-11 19:54
many of us have QBank - you can provide the Qbank ID.
and what are you talking about? Lot's of folks do post QBank questions here - for asking questions..
CP作者: tobeornottobe 时间: 2011-7-11 19:54
Question ID#: 86547
3rd question.
thanks for looking into this.作者: huangxiaoxie 时间: 2011-7-11 19:54
Regression 1:
R^2 was 0.2244
t-stat for T-bill = 1.98
t-stat for S&P 500 Return −0.0161 / 0.032 = 0.5031
t-stat Global index return = 0.0037 / 0.034 = 0.1088
Neither of these is significant for t 0.05,16
so one of these related variables was dropped - which looks like the standard case for Multicollinearity.
