When macro time series meets micro panel data: A clear and present danger

Abstract

This short note addresses a serious problem in combining macro (country-level) time series data, e.g., oil price uncertainty (OPU) or economic/trade policy uncertainty (EPU/TPU), with micro (firm-level) financial and accounting panel data, e.g., corporate leverage, investment, and innovation, to name a few. In most of the applications, the main interest is to assess the impacts of country-level explanatory variable on the firm-level dependent variable, with year fixed effects (along with other firm fixed effects, etc.) being included. Since the macro time series are the same for all firms in each year, it is straightforward to show that the macro time series variable is perfectly correlated with the year fixed effects, and thus unidentifiable. We employ three real data sets to illustrate the perfect multicollinearity issue, and our demonstrations cast doubt on the findings of several existing studies suffering from this issue. Finally, we also offer some practical ways to get around with (at least mitigate) this problem.