EPU- Index(Economic Policy Uncertainty Index) and its transformation

[lal mohan kumar]

Dear all
In one of my study, I am using, Economic Policy Uncertainty Index (epu) as my independent variable. I am using firm-level data(panel), and for each firm-year, I have corresponding economic policy uncertainty index sourced from http://www.policyuncertainty.com/all_country_data.html. Thus for all firms in the year 2010, I have epu=105. My doubt is related to the usage of EPU as an independent variable. My dependent variable is the investment and I want to check the relation between firm-level investment and epu. I referred some papers and in some of them, they directly used epu as an independent variable. I am doubtful whether we can use an index directly as a variable of interest. Hence I transformed epu into logarithmic differences and used it as my independent variable. I ran the following commands

gen delta_log_epu= log(epu)-l.log(epu)
xtreg investment delta_log_epu other_controls i.year, fe vce(robust)

Question 1: Does my transformation of epu makes sense? Is it logically and empirically correct?
Question 2: Should I use delta_log_epu or lag of delta_log_epu?
Question 3: Can I use firm-fixed effects in this setting?

Can someone help me here?
Thanks in advance

[Phil Bromiley]

You didn't get a quick answer. You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters (which you do), readable Stata output, and sample data using dataex.

Whether a particular transformation makes sense is really contingent on your substantive area - we are generally not experts in your (or any specific) area. I think that taking the difference in logs means you're really using something close to the percentage change in uncertainty. In such a case, your parameter is the change in the predicted value for a one unit change in the percentage change of uncertainty (expressed as a fraction/decimal). Note that you will never have a one unit change (100%) in uncertainty from year to year/ I find this hard to think about.

Personally, I find differencing a rhs variable but not the dv hard to interpret and taking logs just makes is a little harder. Since you have fixed effects, you're taking out the level. So, using the variable as original coded makes more sense to me. However, I am not expert in this area so it is just my opinion.

As for lagging or not lagging rhs variables, again this depends on knowledge of the context and data collection. If investment in t might influence the uncertainty measure in t, you may need to either lag uncertainty or control for the endogeneity. Lagging the rhs variables is often done for this reason.

If what you want to explain is within-country variation in investment, then the fixed effects estimator is correct. However, if the main action is in country-level stable differences in uncertainty, then you might also want to look at the between effects or use a hybrid model to look at both.

[lal mohan kumar]

Thanks a lot Phil