Question

Consider a study where Y={monthly S&P 500 "return" (%change in price)} and X={uncertainty in government economic policies}. The author acknowledges that a regression of Y on X has omitted variable bias because policy uncertainty is generally higher when the economy is doing worse. (For example, think about what you heard in the news during the Great Recession about economic policies compared to a few years earlier.) The author proposes to remedy this by including a dummy variable D=1 during official (NBER-designated) recessions and D=0 otherwise. Which of the following makes you still worry about OVB in the regression model Y=b0+b1X+b2D+b3DX+u?

a. Policy uncertainty was much higher during the more severe Great Recession than during the smaller Mar-Nov 2001 recession, but both only enter as D=1.

b. Policy uncertainty usually increases prior to presidential elections, and elections are scheduled every 4 years without regard to current economic health.

c. Having more policy uncertainty could plausibly decrease stock returns due to lower capital investment by firms, but having less policy uncertainty shouldn't increase stock returns.

d. None of these

Answer 1

The answer is

a)  Policy uncertainty was much higher during the more severe Great Recession than during the smaller Mar-Nov 2001 recession, but both only enter as D=1

The dummy variable should not be an indicator variable, its value should be lying according to the policy uncertainity revolving around the recesssion. Some recesssions are more depressing than other and this factor should be captured properly in the dummy variable D. This is what we still worry about the OVB in regression model.