We ask whether early educational co-presence with people who later become billionaires predicts legislators' tax-policy behaviour. Using large language models to classify every federal tax bill introduced since 1973 and a new campus-time-overlap measure, we find that exposed legislators originate a larger share of their tax activity, back proposals that are more likely to be regressive, and introduce bills that advance through Congress at higher rates.
Read more
Presented at the IIPF Annual Meeting 2026. An earlier version circulated as “The Precious Networks of the Rich: How the Wealthiest Shape the Tax Agenda”.
Abstract
Despite broad public support for progressive taxation, the U.S. tax system is among the least progressive in its modern history. This paper asks whether early educational co-presence with people who later become billionaires predicts legislators’ subsequent tax-policy behavior. We use large language models to classify the distributional direction of federal tax bills introduced since 1973 and construct a new campus-time-overlap measure. The measure captures entry into a future wealth-bearing contact opportunity structure, not a verified friendship. Exposed legislators accumulate more tax sponsorships over their careers, but this difference disappears when activity is aligned by Congress and chamber. Within common political periods, they instead originate a larger share of their tax activity. Across primary sponsorships and co-sponsorships, the proposals they back are 1.4 percentage points more likely to be regressive rather than progressive. Separately, bills introduced by exposed primary sponsors advance through Congress at higher rates; exposure among same-day original co-sponsors does not account for that relationship. Among the highly selected bills that reach recorded passage votes, exposed and unexposed copartisans do not differ detectably in regressive-direction-aligned voting on the same decision. The sponsorship association persists within university, birth cohort, party, and Congress and is absent for nearby but non-overlapping campus attendance; its point estimate is larger for exposure to inherited than self-made fortunes, although that contrast is imprecise. Campus-time overlap also predicts direct itemized donations, but observed giving does not statistically account for the sponsorship association. Because selection within campus and cohort remains possible, the estimates identify conditional predictive relationships with an educational opportunity structure, not causal influence by particular billionaires.