It (still) seems a lot less uncertain since 2021M01, even without an ad hoc measure for the pandemic.
Baseline results, using EPU (all items) and Covid deaths per thousands, with all dummies for presidential administrations, and recessions.
Figure 1: Economic Policy Uncertainty Index (blue), and average value with presidential dummies, log of (1+covid death rate per 1000), presidential and recession dummies (tan). NBER defined peak-to-trough recession dates shaded gray. Source: policyuncertainty.com, CDC, NBER, and author’s calculations.
Robustness Tests: Using an alternative covid dummy.
Reader Econned castigates me for a sloppy empirical analysis. I have repeated the analysis, by (1) testing for unit roots, (2) estimating including all presidential dummies, (3) proxying a covid effect with per capita covid deaths, and (4) including a recession dummy.
Re: (1), the full EPU series rejects a unit root using ADF test (constant, trend, 1 lag) at the 0.0000 level. It rejects the Elliott-Rothenberg-Stock point optimal test (trend, intercept) at the 1% msl. It also rejects trend stationarity at 5% msl, using the Kwiatkowski Phillips Schmidt Shin (KPSS) test. I’ll proceed assuming I(0) series.
Figure 2: Covid deaths rates per thousand population (CDPM, blue, left scale), and ln(1+CDPM) (tan, right scale). NBER defined peak-to-trough recession dates shaded gray. Source: CDC, population from FRED, and author’s calculations.
Estimating using OLS on the baseline EPU, I obtain.
EPU = 119.9 + 411.7ln(1+CDPM) + 36.0recession – 15.7GWBush – 34.4Clinton – 30.6GWBush + 11.2Obama + 21.4Trump + 2.7Biden
Adj-R2 = 0.48, SER = 28.7, DW = 0.73, NObs = 464. Bold indicates significance at 10% msl using HAC robust standard errors.
A F-test for equality of the Trump and Biden coefficients rejects with a p-value of 0.32.
The Trump administration is characterized by 21.4 point higher EPU than under Reagan. Adding in a dummy variable (covid20m02-22m02) for the Covid period as suggested by reader Econned (2020M02-2022M02) would reduce the estimated coefficient on the Trump dummy to 20.4 (statistically significant at 5% msl), while the Biden coefficient (6.9) is not statistically significant.
EPU = 118.5 + 57.4covid20m02-22m02 + 33.2recession – 13.7GWBush – 32.9Clinton – 28.6GWBush + 12.8Obama + 20.4Trump + 6.9Biden
Adj-R2 = 0.49, SER = 28.4, DW = 0.72, NObs = 464. Bold indicates significance at 10% msl using HAC robust standard errors.
An F-test for equality of Trump and Biden coefficients would fail to reject at conventional levels (p-value of 0.5).
Personally, I am inclined more towards using a continuous series (deaths per thousands), or a dummy with minimal judgment imposed (dummy for covid from 2020M02 onward) as opposed to an arbitrarily chosen one (2020M02-2022M02).
Uncertainty Indicator: News or World Uncertainty Index
Figure 3: US EPU (all items) (blue, left scale), EPU (news only) (tan, left scale), Ahir-Bloom-Furceri WUI (green, right scale). NBER defined peak-to-trough recession dates shaded gray.Source: NBER, policyuncertainty.com.
There is a question of whether to use the full EPU (including policy changes), or just the EPU using news items. Using the latter, and using Econned dummy for the Covid pandemic, I obtain.
EPUNEWS = 107.5 + 106.9covid20m02-22m02 + 52.5recession – 5.6GWBush – 23.1Clinton – 24.4GWBush + 6.3Obama + 15.5Trump – 16.2Biden
Adj-R2 = 0.35, SER = 45.2, DW = 0.57, NObs = 464. Bold indicates significance at 10% msl using HAC robust standard errors.
An F-test for equality of Trump and Biden coefficients strongly rejects at the 0.005.
In order to cross check these results, I also used the Ahir-Bloom-Furceri World Uncertainty Index (based on Economist Intelligence Unit reports) over this same sample period. This index is available on a quarterly basis, so I aggregate monthly data to quarterly, and run the regression to obtain:
WUI = 0.103 + 0.024covid20m02-22m02 + 0.078recession – 0.053GWBush – 0.033Clinton + 0.065GWBush + 0.127Obama + 0.224Trump + 0.091Biden
Adj-R2 = 0.30, SER = 0.135, DW = 1.12, NObs = 154. Bold indicates significance at 10% msl using HAC robust standard errors.
An F-test for equality of Trump and Biden coefficients rejects at the 0.33 level. An equality test for Trump vs. Obama coefficients rejects at the 0.10 level.
Conclusion: Uncertainty under Biden is lower than that under Trump.
Addendum, 10/1/2023, 5:20pm:
Reader Econned is exercised by my assertion that his 2020M02-2022M02 covid dummy is arbitrary. As he writes:
nothing was ‘arbitrary’ regarding the specification of the COVID dummy. I clearly stated why it was used.
2c) you suggest that “using your arbitrarily selected Covid dummy — uncertainty under Biden was lower than that under Trump. And significantly so in a couple cases.” No sh*T Sherlock… I literally posted the coefficient estimates. Why are you so bad at reading comprehension? You’re an ‘academic’ – isn’t 50% of your career based on reading?!?!
The point is that varying the covid dummy arbitrarily chosen can make the gap between Trump and Biden coefficients larger or smaller. Is covid with us. I still where a mask — and some people at the airport do now too. Did the situation normalize somewhat after 2020M02? Well sure, but I would use 2022M05 (I think with greater justification) than 2022M02. Consider this redraw of Figure 2.
Figure 4: Covid deaths rates per thousand population (CDPM, blue, left scale), and ln(1+CDPM) (tan, right scale). Period proposed by Econned for covid dummy (with implied “leveling off” at 2022M03. Source: CDC, population from FRED, and author’s calculations.
Remember Econned wrote
“Instead, I ran a model specifying COVID as 2020M02-2022M02. These dates were selected by analyzing CDC’s statistics of US monthly COVID deaths. This series show 2022M03 as the month when a noticeable leveling off occurred followed by a sustained period of relatively low deaths (not to diminish any level of deaths). Given the vaccines, etc I think it’s incorrect to keep COVID dummy as “1” in perpetuity which is what you seemingly decided as appropriate.
Just to be clear, in 2022M03, covid deaths narrowly defined by CDC were running at 16,070 per month, and broadly (covid primary or comorbidity) at 26,608. That is, covid deaths per capita were at that time of “leveling off” equal to mid-2021 death rates.
Using a dummy=1 2020M02-2022M05, as deaths per capita hit a local minimum at 2022M05. Then the Trump coefficient (19.6) is significant at the 5% msl, the Biden coefficient (-0.2) is not. F-test for equality of coefficients rejects at 35% msl.