Evaluate carried-interest imputation using Yale/Love partnership-data method

[MaxGhenis]

Summary

Yale Budget Lab's carried-interest update suggests our US data stack may understate the revenue base for carried-interest reforms. Yale uses Michael Love's K-1/Form 1065 method to estimate carried interest from the gap between partners' profit allocations and capital shares, rather than relying only on public partnership aggregates.

Yale estimates materially higher 2026-2035 revenue from taxing carried interest:

• Biden 2025 Budget proposal: $28.1B new method vs $10.7B old method
• Wyden et al. proposal: $87.7B new method vs $47.5B old method
• Broadest proposal: $87.7B-$104.2B depending on Section 1231 assumptions

Motivation

PolicyEngine currently has broad variables such as long-term capital gains, qualified dividends, and partnership/S-corp income, but no carried-interest-specific income variable or calibration target. This limits both revenue estimates and distributional analysis for carried-interest reforms.

Yale's public GitHub org does not appear to contain a carried-interest-specific microsimulation variable/module, so this may need to be approximated from public data rather than directly ported from their code.

Proposed scope

1. Review Yale/Love methodology and identify what can be approximated with public data.
2. Add a synthetic/calibrated carried_interest_income or similar variable in the enhanced PUF/CPS data pipeline.
3. Allocate carried interest among likely high-income partnership/capital-gains records, with uncertainty bands.
4. Add validation targets based on Yale's published estimates.
5. Cross-link a follow-up policyengine-us issue for reform mechanics once the data variable exists.

Related issues

• Use Forbes 400 data to improve $100M+ AGI bracket synthesis #613: Use Forbes 400 data to improve $100M+ AGI bracket synthesis
• Add general partner vs limited partner distinction for SE tax #480: Add general partner vs limited partner distinction for SE tax

Sources

• Yale Budget Lab: https://budgetlab.yale.edu/research/refining-revenue-estimates-taxing-carried-interest
• Yale table data: https://datawrapper.dwcdn.net/drY3I/1/dataset.csv
• Love paper: https://www.sciencedirect.com/science/article/pii/S0047272725001914

[MaxGhenis]

Suggested PolicyEngine modeling shape:

Data/model boundary

This should probably be implemented as a non-additive classification overlay rather than a new income source.

In policyengine-us-data, add a synthetic variable such as carried_interest_income meaning: the portion of existing preferred-rate income estimated to be carried interest or similar profits-interest compensation.

Important constraint:

carried_interest_income <= existing preferred-rate / partnership-linked income

It should not increase AGI, taxable income, long-term capital gains, qualified dividends, or partnership income in the baseline. It should tag part of income already present in the data so reform simulations can re-rate it without double-counting.

Candidate allocation logic

Allocate the synthetic tag primarily among high-income tax units with combinations of:

• high long-term capital gains
• high qualified dividends, if the proposal definition includes them
• high partnership/S-corp income
• top-tail finance/investment archetypes, where available
• future Forbes/top-tail synthetic records from Use Forbes 400 data to improve $100M+ AGI bracket synthesis #613, if that work lands

The data pipeline should probably support central/low/high assumptions, since we cannot directly reproduce Yale/Love from public K-1/Form 1065 microdata.

Follow-up policyengine-us reform mechanics

Once the data variable exists, a policyengine-us reform could reclassify the tagged amount from preferred-rate income to ordinary income. Conceptually:

preferred_income = adjusted_net_capital_gain + qualified_dividend_income
reclassified_income = min_(carried_interest_income, preferred_income)
ordinary_taxable_income += reclassified_income
preferred_rate_income -= reclassified_income

This would likely need parameters for:

• proposal type: Biden FY2025, Wyden/Reed/Whitehouse/King, broad proposal
• Section 1231 inclusion share: 0%, 25%, 50%, 100%
• income threshold, if applicable
• behavioral response assumption
• carried-interest base scenario: low/central/high

Validation/output expectations

Validation should include sensitivity tables against Yale's published revenue estimates, especially:

• Biden 2025 Budget proposal: $28.1B central new-method score over 2026-2035
• Wyden et al. proposal: $87.7B central new-method score over 2026-2035
• Broad proposal: $87.7B-$104.2B depending on Section 1231 assumptions

Useful simulation outputs would include revenue, affected tax units, average tax increase among affected units, and distribution by income percentile/top-tail group.

[MaxGhenis]

Skeptical view / risks to track before treating Yale/Love as a hard target:

Method caveats

1. Definition mismatch risk. Love's method appears to identify carried interest and similar profits-interest arrangements from partnership allocations. That may be broader than the income targeted by a specific carried-interest bill. A PolicyEngine variable should distinguish the broad inferred profits-interest base from narrower statutory carried-interest definitions where possible.

2. Public-data reproducibility is limited. The strongest part of the Yale/Love method is access to K-1/Form 1065 capital and profit-allocation information. Public PUF/CPS-style data do not include those fields, so any PolicyEngine implementation would be an approximation calibrated to external totals, not a direct replication.

3. Allocation uncertainty may dominate. Even if the aggregate carried-interest base is credible, assigning it to public microdata records is hard. Proxies like high partnership income, high long-term capital gains, qualified dividends, or finance/top-tail archetypes can misclassify entrepreneurs, real estate investors, passive LPs, public-company founders, and other high-income capital-gains filers.

4. Statutory scope matters a lot. Yale's own broad-proposal estimates vary with the Section 1231 inclusion assumption. That implies the model should expose Section 1231 scope as a sensitivity parameter rather than burying it in one central value.

5. Revenue is not just base times rate differential. Taxing carried interest as ordinary income can interact with NIIT, AMT, ordinary brackets, loss limitations, holding periods, entity choice, compensation form, and timing. Some of Yale's score may reflect assumptions that are not visible in the public writeup.

6. Behavioral response is especially uncertain. A reform could shift compensation toward management fees, wages, debt-like returns, corporate blockers, offshore structures, longer holding periods, or other contractual forms. Standard capital-gains realization elasticities may not be the right response model for carried-interest reclassification.

7. Time-period extrapolation risk. Love's empirical estimates cover historical partnership behavior, and Yale projects future revenue. Private-market returns, interest rates, deal volume, and partnership structures can move substantially over a budget window, so 2026-2035 estimates should be treated as scenario targets, not immutable calibration totals.

Implementation implication

Use Yale/Love as an external validation range, not a single hard constraint. A first implementation should probably include low/central/high carried-interest-base assumptions and report sensitivity against Yale's published scores. The central case can target Yale's new-method estimates, but the data pipeline should make the uncertainty explicit.