Stack of papers / notebook
Research
Econometrics on tax policy and the political economy of peacekeeping.
As team lead at Northeastern, under Prof. Omar Robles, I directed a study of the socioeconomic legacy of the 1981 Economic Recovery Tax Act (ERTA) on U.S. income disparity, running five econometric specifications across 65 years of U.S. data and an 812-observation OECD panel, then turning the findings into a corrective policy proposal. In a separate CU Boulder internship under Prof. Megan Shannon, I examined why autocracies contribute to UN peacekeeping and found that deployments track strategic self-interest over humanitarian need.
- 65 yrs
- of U.S. tax & income data
- 812
- OECD panel observations
- ~12%
- of the income-gap rise tied to ERTA
Northeastern University · Team lead · Prof. Omar Robles
ERTA and the U.S. income gap
As team lead, I directed a study of the socioeconomic legacy of the 1981 Economic Recovery Tax Act on U.S. income disparity, running five econometric specifications over 65 years of U.S. tax and income data and an 812-observation panel of 17 OECD economies. The question is narrower than the usual argument about Reaganomics: not whether inequality rose after 1981, which is not in dispute, but how much of the rise the rate cuts themselves can carry.
- Data: WID.world top-1% pre-tax shares, Tax Foundation marginal rates, FRED real GDP per capita, Census Gini, Saez-Zucman wealth shares.
- Specifications: Welch t-tests, Chow and Quandt-Andrews structural-break tests, a six-country OECD placebo test, and a first-differences regression with Newey-West errors.
- Presented to Prof. Robles with strongly positive feedback; currently under faculty review.
Finding · Structural break
The trend did not steepen in 1981. It reversed.
The easy version of this story is that inequality was already climbing and 1981 made it climb faster. That is not what the series does. Through the New Deal-era rate structure the top 1% share was in slow decline, about a sixth of a percentage point a year. After 1981 the same series turns and rises at roughly a fifth of a point a year. A Chow test puts the break at 1981 with F(2, 61) = 86.03. The sign of the trend, not just its slope, is what changed.
Before 1981 the top 1% share was shrinking. After 1981 it grew.
For two decades the richest 1% were taking a slowly shrinking slice of American income. Around 1981 that reversed, and it has been climbing ever since. The line does not just get steeper. It changes direction.
Fitted least-squares trend in the US top 1% pre-tax income share, before and after the break. A Chow test puts the break at 1981 with F(2, 61) = 86.03, p < 0.001, less than a one-in-a-thousand chance of a split this clean if the two eras really followed one trend.
- Pre-ERTA 1960–1980 · share shrinking 0.156 points a year
- Post-ERTA 1981–2024 · share growing 0.206 points a year
Hover or drag across the chart to read any year. With the chart focused, the arrow keys do the same.
Data & method
Fitted lines only. Each is drawn from the period mean and slope reported in the paper (pre: 11.60% and −0.156pp/yr; post: 16.45% and +0.206pp/yr), the underlying year-by-year WID.world series is not reproduced here. Shaded bands are each period's observed min–max. A percentage point (pp) is one point of the national income pie, so +0.206pp/yr means the top 1% gained about a fifth of a point of all US income every year after 1981.
Finding · Level and variance
Two different distributions, not one drifting
Treating the two eras as samples rather than a single series makes the size of the move legible. The pre-ERTA mean is 11.60% across 21 years; the post-ERTA mean is 16.45% across 44. A Welch two-sample t-test returns t = −10.10. The second thing the comparison shows is less quoted and more interesting: the standard deviation more than doubles, from 1.09 to 2.77. Top incomes did not just get bigger, they got far more cyclical, which is what you would expect once they are tied to asset prices rather than wages.
Two different worlds, not one world drifting
Line the two eras up side by side and they barely overlap. The post-1981 curve sits well to the right, a bigger share, and is also much flatter, which means top incomes swung far harder from year to year than they used to.
The top 1% share averaged 11.6% across 21 pre-ERTA years and 16.45% across 44 post-ERTA years, a 4.85pp shift that a Welch two-sample t-test rejects as chance at t = −10.10, p < 0.001. The spread more than doubled too, from σ 1.09 to σ 2.77.
- Pre-ERTA 1960–1980 · 21 years
- Post-ERTA 1981–2024 · 44 years
The white marker on the chart moves with this slider. Everything to its right is a year the top 1% took more than that.
- Pre-ERTA 1960–1980, years above the threshold
- 0 of 21<1% of the era
- Post-ERTA 1981–2024, years above the threshold
- 31 of 4470% of the era
Above 15.0%: roughly 0 of the 21 years before ERTA, and 31 of the 44 years after it.
Data & method
Normal approximations drawn from the mean and standard deviation reported in Table 1 of the paper (pre: 11.60 ± 1.09; post: 16.45 ± 2.77). The real series is not necessarily normal, this shows the location and spread the test compares, not the empirical histogram. Year counts under the slider are that same approximation applied to each period's length, so they are estimates rather than a tally of observations.
Method · The full specification curve
Seven specifications, two of which found nothing
One specification that agrees with you is a coincidence. The defensible version of this claim is the whole set, so the paper runs five and the methodology companion documents two more, and the two that return nulls are written up at the same length as the ones that do not. A reader who opens the PDF should find no result there that is missing from this page.
Every test we ran, including the two that found nothing
Five of these are in the paper; the panel regression and the event study are in its methodology companion. Open any row for the full finding and where it is weak.
Finding · The placebo test
The result that cuts against the thesis
The hardest test in the paper is the one designed to break it. Applying the same Chow test at 1981 to six OECD economies that enacted no comparable top-rate reduction between 1979 and 1983 should, if ERTA is doing the work, find nothing. It finds a significant break in all six, and the US statistic of 96.2 sits inside their range rather than beyond it. That rules out the strong claim. What survives is the more precise one: the break was global, the magnitude was American. ERTA is not the cause of a worldwide shift, it is the policy architecture through which that shift was channelled into uniquely concentrated US outcomes.
Six countries that never cut their top rate broke in 1981 too
If ERTA caused the break, countries that passed no such tax cut should show nothing in 1981. All six show one, and the US result lands inside their range rather than beyond it. That rules out the simple story, and it is the reason the paper argues the narrower one.
The same Chow test applied at 1981 to OECD economies with no comparable 1979–83 top-rate reduction. Each bar is that test's F-statistic: the longer it runs, the more sharply the country's trend snapped in 1981. New Zealand, which cut nothing, scores 97.1. The United States scores 96.2.
The other 4 countries in the test broke significantly too. The paper reports them as significant without printing a figure, so they are counted in the tally above but cannot be placed on the scale.
- Sweden
- Finland
- Switzerland
- Australia
Data & method
F-statistics at 1981. Published values exist for New Zealand and Canada only; Sweden, Finland, Switzerland, Australia are reported as significant without a figure, so they are counted in the tally but not plotted. An F-statistic here measures how much better the years before and after 1981 fit as two separate trends than as one, Canada's 5.0 is the smallest break in the set and New Zealand's 97.1 the largest.
Mechanism · Financialization
Where the effect was actually hiding
Run the top marginal rate against the top 1% share on its own and it looks irrelevant. Add the FIRE sector's share of GDP as a control and the rate coefficient flips sign and becomes significant, while FIRE itself enters enormous. The financial sector's expansion had been absorbing the effect, and the ERTA-era deregulation that enabled that expansion, SEC Rule 10b-18 above all, which turned buybacks from a legal risk into standard practice, is the link between the tax cut and the concentration. Frydman and Saks put the same shift in compensation: real executive pay grew 0.8% a year from 1936 to 1976 and 8.0% a year from 1977 to 2005.
One missing control was hiding the whole effect
Measured on its own, the top tax rate looks unrelated to what the top 1% take home. Add one thing the model was missing, how large finance, insurance and real estate had grown as a share of the economy, and the relationship appears, pointing the expected way. Finance had been absorbing the effect.
The coefficient on the top marginal rate in a single-country US regression, before and after adding the FIRE sector's share of GDP as a control. It moves from +0.013 (p = 0.778, not significant) to -0.056 (p = 0.015, significant).
Data & method
FIRE share enters with β = 2.97 (SE 0.28, p < 0.001). HAC p-values throughout. From the paper's single-country specification. A coefficient here is how much the top 1% share moves for each one-point change in the top marginal rate; left of zero means cutting the rate raises the share. A p-value is how often a result this large would turn up by luck if the true effect were zero, so 0.778 is "routinely", and 0.015 is "about three times in two hundred".
Estimate · First differences
Putting a number on it
The cleanest specification is a first-differences regression with Newey-West errors, which strips the trend out and resolves the serial correlation that makes the levels regression untrustworthy, Durbin-Watson moves from 0.237 to 1.80. It returns a coefficient of −0.053: a one-point cut in the top marginal rate moves the top 1% share about five hundredths of a point in the same year. Applied to ERTA's actual 19-point cut, that is roughly 1.0pp of the 8pp rise between 1980 and 2024, or about 12%. The remaining 88% belongs to the complementary policies, which is the paper's actual claim and a smaller one than the headline usually gets.
How much of the rise does the tax cut actually explain?
Cut the top tax rate by one point and the top 1%’s share of income rises about five hundredths of a point that year (0.053). ERTA cut it by 19 points. Run that forward and it accounts for roughly an eighth of the 8-point rise since 1980, not the whole story, which is the paper’s actual claim.
Drag to run the same arithmetic on any size of cut. The band under each number is the 95% interval: the range the estimate is actually consistent with, which is wide.
Points of the rise it explains
1.01of 8
somewhere between 0.08 and 1.94
As a share of the whole rise
12.6%
somewhere between 0.9% and 24.2%
At ERTA’s actual 19-point cut the estimate attributes about 1.0 point, roughly 12% of the observed rise. The other 88% is what the complementary policies amplified: the 10b-18 buyback regime, OBRA welfare contraction, the PAC explosion, and antitrust retrenchment.
Team leadership · From findings to policy
Leading the team, and a pivot to policy
I led the team the way Satya Nadella leads: set a clear shared vision, then give people the freedom to decide how to reach it. Balancing full course loads, we produced roughly twenty pages of research in about two months. As the evidence came in, we pivoted the argument from reforming welfare toward implementing a negative income tax, a substantive reframing that the team's creative freedom made possible.
CU Boulder · Research intern · Prof. Megan Shannon
Why autocracies join UN peacekeeping
Selected for a research internship under political science professor Megan Shannon, I synthesized ten years of scholarship and mission-level evidence to explain why non-democratic states contribute troops to UN peacekeeping. The puzzle is that peacekeeping is expensive, and the humanitarian return on it accrues mostly to other people, so a regime that answers to no electorate has thin reason to pay. The evidence points one way: China's and Russia's deployments track strategic self-interest, access, influence over mission mandates, and standing in the institutions that authorize them, rather than humanitarian need. It is the same instinct as the ERTA work, applied to a different subject: take the stated rationale for a policy, and check it against where the resources actually go.