1
$\begingroup$

when there is simultaneous causation between dependent and independent variables often in papers we see a lagged variable being used in place of contemporaneous variable. For example, Clemens, Radelet, Bhavnani and Bazzi (2012) say - “We avoid poor-quality instrumental variables and instead address potential biases from reverse and simultaneous causation by the more transparent method of lagging". Hayo, Kutan, and Neuenkirch (2010) - “The vector of controls contains lagged returns...Contemporaneous U.S. returns are excluded to avoid simultaneity problems” However what is the rationale for lagging? This short article looks at estimators of lagged variables used in such cases. It deduces that either there is no serial correlation in the variable and the value of the lagged estimator is 0. Or that there is serial correlation and the value of the lagged estimator is quite a meaningless value of  ((c+bf)/(1-be)), where b is the coefficient of x(t) on y(t), f is the serial correlation, e is the effect of y(t) on x(t) and c is the true effect of x(t-1) on y(t). Further the article finds that these estimators are consistent but converge to true value only for n>100 approximately, further leading to wrong significance interpretations. Given this, what is the rationale of using a lag to deal with simultaneity? If the lagged variable is not serially correlated its coefficient on y(t) is 0, it does not capture the effect of x(t) on y(t) and one only risks to make false significance inferences in small samples. If on the other hand the variable is serially correlated than it does capture the effect of x(t) on y(t), but only by the fraction of serial correlation and further that result is distorted in non linear way because the estimator also captures the effect of 1 - (y(t) on x(t) times x(t) on y(t)) in the denominator. So it is just a meaningless result. Henceforth, why do people use the lagged variables to address simultaneity endogeneity? It seems to make no sense.

The aforementioned article is this one: https://www.otago.ac.nz/economics/news/otago075947111111111.pdf

Your thoughts are much appreciated!

$\endgroup$
  • $\begingroup$ Hi: I only glanced at the paper and In not well versed in simultaneity bias but it sounds like an important finding which should be in a well known respected journal. If that is not the case, then the result is either incorrect or already well known. So, your best best is to ask an econometrics professor who is an expert in simultaneity bias and ask them what they think of the result. I would think something like that would have been discovered before ? $\endgroup$ – mark leeds Jul 24 at 23:19

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.