TaxAgent: How Large Language Model Designs Fiscal Policy

Jizhou Wang; Lei Huang; Xiaodan Fang; Yongfeng Huang

REVIEW 4 major objections 5 minor 23 references

TaxAgent: How Large Language Model Designs Fiscal Policy

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A large language model acting as a tax authority designed tax rates that beat classical optimal-taxation theory in a 120-month simulated economy.

desk verdict A novel LLM-agent framework for adaptive tax design, but the headline result rests on a misimplemented Saez baseline and a single stochastic run, so the superiority claim does not stand. read the letter →

arxiv 2506.02838 v1 pith:VZYVFMUZ submitted 2025-06-03 cs.AI econ.GNq-fin.EC

Jizhou Wang , Xiaodan Fang , Lei Huang , Yongfeng Huang This is my paper

classification cs.AIecon.GNq-fin.EC

keywords largelanguagemodelsagent-basedmodelingtaxpolicydesignoptimaltaxationmacroeconomicsimulationequity-efficiencytrade-offfiscalevaluationadaptive

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that a large language model, acting as a government tax authority in a simulated economy, can design bracket tax rates that achieve a better balance between equality and productivity than three established alternatives: the Saez optimal-taxation schedule, the U.S. federal income tax, and a no-tax free market. It builds the comparison into an agent-based macroeconomic simulation in which fifty LLM-based households choose how hard to work and how much to consume under whatever rates the LLM tax agent sets, and the tax agent revises rates every period from the resulting production, wage, price, and unemployment data. Over 120 simulated months the LLM-planned system stabilizes equality and productivity at a higher combined score than the baselines, and the result holds across different underlying LLMs. A sympathetic reader would take the paper's contribution to be a demonstration that iterative LLM planning, without explicit welfare functions or rationality assumptions, is a workable policy-search mechanism.

What carries the argument

The load-bearing mechanism is the iterative feedback loop among three components: H-Agents (household LLMs) that produce $(p_i^w, p_i^c) = H_i(P^{mt}_i, \theta_i^R)$, a macroeconomic environment that aggregates their choices into production $S$, demand $D$, prices, wages, inflation, and unemployment, and the TaxAgent (government LLM) that maps current metrics to a new seven-bracket tax schedule $TX = Gov(P^{mt}, \theta^G, \theta^H)$. The loop is what makes the policy adaptive: the tax agent does not assume a utility function or elasticity; it adjusts rates from observed household reactions. The yardstick is the equality-productivity product, where equality is one minus the wealth Gini index, so every tax proposal is scored by the same social-outcome metric.

What would settle it

Measure the effective labor-supply elasticity of the H-Agents by regressing log work propensity on log net-of-tax rate in the simulation and compare it with empirically documented elasticities of taxable income, roughly 0.1 to 0.5; if the LLM households respond far outside that range, the claimed realism of the behavioral channel fails. A second check is to rerun the 120-month comparison with H-Agents replaced by rule-based households calibrated to those elasticities and see whether TaxAgent still beats the Saez schedule.

Watch

Extended reading notes

Core claim

TaxAgent closes a loop between an LLM government and LLM households in a macroeconomic environment with production, bracketed taxation, consumption, and a Taylor-rule financial market. Each household outputs a work propensity $p_i^w$ and a consumption propensity $p_i^c$; the environment converts those into output, wages, prices, inflation, and unemployment; the TaxAgent reads updated metrics and proposes next-period tax rates for seven income brackets. The performance metric is the product of equality (one minus the normalized Gini index of wealth) and per-capita productivity, and the TaxAgent's adaptive rates keep this product higher than the Saez schedule, the U.S. federal schedule, and the free market, especially after month 40. The paper attributes Saez's shortfall to regressiveness in the simulated schedule and the U.S. system's shortfall to fixed rates, while the free market stagnates without redistribution.

Load-bearing premise

The result collapses if LLM-generated household work and consumption propensities are not a valid stand-in for real taxpayer responses to tax rates, because the entire simulation outcome is built from those propensities.

Editorial extensions

If this is right

If the result holds beyond the simulation, fiscal authorities could use LLM-based sandboxes to stress-test bracket designs before legislation, at a fraction of the cost of pilots.
Dynamic adjustment becomes a prompt-level operation: the same government loop can re-optimize rates when unemployment, inflation, or productivity shifts, without recomputing welfare functions.
The comparison implies that classical sufficient-statistics tax schedules are not necessarily the practical frontier when taxpayer behavior is heterogeneous and boundedly rational.
Because the result is robust to swapping the base LLM, the design principle of closed-loop LLM policy search may transfer to other policy domains, not just income tax.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The paper does not test whether H-Agent work-consumption responses match empirical labor-supply elasticities; if real taxpayers are less responsive, TaxAgent's dynamic flexibility may matter less, while if they are more responsive, its advantage could shrink or grow.
A natural extension is to replace or calibrate the household LLM prompts with data from actual tax returns or surveys, turning the framework from a proof-of-concept into a predictive policy instrument.
The same closed loop could be pointed at other redistribution instruments, such as wealth taxes, corporate taxes, or universal transfers, and compared against behaviorally calibrated agent-based models rather than only classic schedules.
One implicit consequence is that a sufficiently good policy language model could act as a flexible social-welfare maximizer that avoids committing to a cardinal utility function, shifting the modeling burden to validating agent behavior rather than choosing welfare weights.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Bottom line: the TaxAgent combination - LLM households plus an LLM tax planner iteratively setting bracket rates - is genuinely new, and the paper describes it clearly enough to replicate. But the headline claim of "superiority over Saez Optimal Taxation" does not survive contact with the appendix: the Saez baseline is misimplemented. Eq. (32) is not Saez's formula, and the regressive schedules in Fig. 4 contradict the theory the baseline is supposed to represent. The comparison is against a strawman, and that is load-bearing for the abstract's central claim.

What the paper does well: the framework is laid out in enough detail that a reader can see how the pieces fit; the prompts in the appendix give a concrete sense of what the agents see and decide; and the ablation across three LLMs is a reasonable first check on robustness. The idea of modeling heterogeneous, non-rational taxpayer behavior with LLMs is worth taking seriously, and the paper is honest about the limitations of traditional optimal taxation.

The soft spots, in order of severity:

The Saez implementation (Appendix e) is wrong. Saez (2001) gives the optimal marginal rate as (1-G(z))/(1-G(z)+alpha(z)e(z)), with alpha(z)=zg(z)/(1-G(z)). The paper's Eq. (32) adds e*z*g(z) to the numerator and e*g(z) to the denominator; neither is dimensionally consistent with Saez's derivation, and the resulting regressive schedules are the opposite of what the theory implies. So the central benchmark is invalid.

The experiment is a single run with no variance. No seeds, no error bars, no code. The text says "results remain consistent across parameter variations" but nothing backs that up.

The free market result - lowest productivity under zero taxation - is counterintuitive and unexplained. That suggests something is off in the dynamics, and the paper does not address it.

The TaxAgent's prompt contains the evaluation metrics and asks it to optimize "the best for society," so the objective is handed to it. That makes the result partly self-fulfilling, though the comparison to rule-based baselines within the same environment does give some context.

The load-bearing premise - that LLM household choices proxy real taxpayer behavior - is never validated against empirical elasticities or real-world responses.

Who is this for: readers working on LLM-based agents in computational economics or agent-based policy simulation. They will find the framework a useful starting point, and the appendix prompts are instructive. But the current evidence does not support the paper's claims.

Recommendation: I would send it to peer review with a strong request for major revision, so a referee can pin down the baseline error and demand proper statistical support. The idea is worth discussing; the results as reported are not.

Referee Report

4 major / 5 minor

Summary. The paper introduces TaxAgent, an LLM-driven government agent embedded in an agent-based macroeconomic simulation, and compares its tax-rate proposals with three baselines: Saez optimal taxation, the U.S. federal income tax, and a free-market scenario. The central claim is that TaxAgent achieves superior equity-efficiency trade-offs, measured by a product of equality (complement of the Gini index) and productivity (wealth per capita). The authors further claim robustness across two LLM backbones and report that the free market performs worst, while the Saez baseline is described as regressive.

Significance. If the benchmark comparison were valid, the paper would be an interesting proof-of-concept that LLM agents can discover adaptive tax schedules in a stylized economy, and the appendix provides full prompt templates and a two-LLM ablation, which are useful transparency elements. However, the central comparison is not currently established: the Saez baseline is misimplemented in a way that produces a strawman, the results rest on a single unseeded simulation without variance or statistical tests, and the free-market outcome contradicts basic economic intuition. These issues are load-bearing because they directly support the abstract's claim of superiority over Saez optimal taxation.

major comments (4)

[Appendix (e), Eq. (32)] Equation (32) is not Saez's optimal nonlinear tax schedule. Saez (2001) gives T'(z) = (1 - G(z)) / (1 - G(z) + alpha(z) e(z)), with alpha(z) = z g(z) / (1 - G(z)). The paper's formula, T'(z) = ((1 - G(z)) + e z g(z)) / (1 + e g(z)), adds an income-density term to a dimensionless numerator and adds a density term in the denominator, so it is dimensionally inconsistent and does not follow from Saez's derivation. The consequence appears directly in Figure 4, where the 'Saez' schedule is reported as regressive, whereas the Saez optimum is progressive for lower and middle incomes and at worst flat at the top. Because the abstract's headline claim is superiority over Saez optimal taxation, this misimplementation invalidates the benchmark comparison.
[Section IV-B and IV-D] The experimental section reports no number of random seeds, no standard deviations, and no statistical tests. Figures 2-5 appear to plot single trajectories, and Section IV-D1 states that TaxAgent 'performs significantly better' without any significance testing. Given that LLM outputs are stochastic and the simulation has random components, a single run cannot support the paper's quantitative claims, and the absence of released code or data prevents replication despite the claim of 'replicability information' in Section IV-B.
[Section IV-D2 and Appendix (b)] The TaxAgent's prompt explicitly provides the evaluation metrics (productivity as average wealth, equality as Gini complement) and instructs the model to 'build a society that you consider best for society.' The rule-based baselines are not given this objective; in particular, the Saez baseline is computed from an incorrect formula rather than from an optimization of the same social objective. This is not circular in a logical sense, but it makes the comparison unequal: TaxAgent is directly told what to optimize, while the baselines are not. Consequently, the observed superiority may reflect prompt design rather than a genuine policy-discovery advantage.
[Section IV-D2] The free-market scenario, with zero taxation and zero redistribution, is reported to have the lowest productivity and the highest unemployment, 'without taxation and redistribution, societal productivity stagnates.' This contradicts the standard economic prior that removing distortionary taxes should increase, not decrease, labor supply and productivity, especially when agents are assumed to respond to incentives. The paper offers no mechanism explaining this outcome and no validation of the H-Agent decision function in Eq. (1) against empirical elasticities or any sanity check. Since the free-market baseline is part of the central comparison, this unexplained result undermines confidence in the simulation's economic validity.

minor comments (5)

[Figure 1] Figure 1 contains multiple typographical errors that should be corrected: 'Affact' should be 'Affect', 'Institiution' should be 'Institution', and 'Mismatich' should be 'Mismatch'.
[Section IV-C and Eq. (24)] Section IV-C defines productivity as 'the current average wealth of H-Agents,' but Eq. (24) computes it as the sum of wealth over all households. These definitions are inconsistent and should be reconciled.
[Eq. (14) and Section III-A] Equation (14) uses the notation 'pc_j' for the consumption propensity, but Section III-A defines the pair (p^w_i, p^c_i) for work and consumption propensities. The text near Eq. (14) also refers to 'working propensity' where it should refer to consumption propensity.
[Section V and Appendix (f)] There are minor typos in prose, including 'Seaz' instead of 'Saez' in Section V and 'Chatgpt' instead of 'ChatGPT' in Appendix (f).
[References [11] and [12]] References [11] and [12], on climate change governance and power quality enhancement respectively, appear unrelated to the claim about static tax systems lacking adaptability and should be replaced with relevant citations.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the TaxAgent's derivation chain is self-contained, and the Saez-baseline concerns are benchmark-correctness issues rather than circularity.

full rationale

The paper's central claim—that TaxAgent achieves superior equity-efficiency trade-offs—is an empirical simulation outcome, not a derivation from its own outputs. TaxAgent's prompt in Appendix (b) supplies historical productivity and equality metrics and asks the LLM to choose rates for the society it considers best, but the evaluation metric (equality × productivity, Section IV-C) is not identical to the prompt by construction: the LLM's rate choices are not guaranteed to maximize that product, and the comparison is made against rule-based baselines (free market, US federal, and the paper's Saez implementation) within the same simulated economy. No parameter is fitted to a subset and then presented as a prediction. The only self-citation with author overlap, reference [20] (Phyx), appears in the related-work survey on LLM reasoning and is not load-bearing. Appendix (e) Eq. (32) appears inconsistent with Saez (2001) and Figure 4 shows regressive rates, which would undermine the Saez comparison as a benchmark-validity problem, but this does not make the claim circular: TaxAgent's superiority does not reduce to its inputs by definition. No manuscript passage asserts a limitation or omitted proof that would change this assessment.

Assumptions & free parameters 8 free parameters · 5 assumptions · 0 invented entities

The central claim depends on many free parameters and unvalidated domain assumptions. The framework is a stylized model with no calibration to real economies, and the LLM prompts themselves introduce undocumented variation. The free parameters in the macro equations and the unspecified Saez elasticity parameters mean the comparison is not replicable from the text.

free parameters (8)

Price adjustment rate αp
Hand-chosen coefficient in Eq. (20) controlling price response to demand-supply mismatch; no calibration is reported.
Wage adjustment rate αw
Hand-chosen coefficient in Eq. (19) for wage response; not fitted.
Taylor rule coefficients απ and αu
Coefficients in Eq. (17) for interest rate response to inflation and unemployment; values are never stated.
Natural interest rate rn
Input to Taylor rule Eq. (17); chosen by hand, no justification.
Productivity A = 1
Fixed at 1 in Section IV-B and used in Eq. (11); no empirical basis.
Saez elasticity e and Pareto parameter a
Required by Eqs. (31)-(32) for the Saez baseline, but values are never specified; the baseline is therefore underdefined.
Number of households N and months P = 50 and 120
Simulation scale N=50, P=120 set in Section IV-B; no robustness check across scales.
Tax bracket boundaries = US 2024 brackets
Brackets in prompts and baseline come from the US tax schedule; chosen as input.

assumptions (5)

domain assumption Linear production function S = sum l_j * 168 * A
Assumes output is proportional to labor hours with no capital, technology, or complementarities; Eq. (11).
domain assumption Even and implicit redistribution of all tax revenue
Eq. (13) assumes a lump-sum equal rebate of all tax revenue, which is not how real tax systems operate.
ad hoc to paper LLM agents' decisions are a valid proxy for household behavior
Core premise of the framework: H-Agent prompts (Eqs. 1-2) generate realistic work/consumption reactions; never validated against empirical elasticities.
domain assumption Equality times productivity is the right social objective
The social outcome metric is defined as the product of equality and productivity in Section IV-C; this normative choice is taken as given.
standard math Taylor rule and demand-supply price/wage adjustment equations
Borrowed from macro literature (Eqs. 17-20), but parameters are not calibrated.

how reviews work

0 comments

Cite this review

Pith. "Pith review of TaxAgent: How Large Language Model Designs Fiscal Policy." pith.science (2026). https://pith.science/paper/VZYVFMUZ

@misc{pith2026250602838,
  author       = {Pith},
  title        = {Pith review of: TaxAgent: How Large Language Model Designs Fiscal Policy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VZYVFMUZ}},
  note         = {Machine review of arXiv:2506.02838}
}

read the original abstract

Economic inequality is a global challenge, intensifying disparities in education, healthcare, and social stability. Traditional systems like the U.S. federal income tax reduce inequality but lack adaptability. Although models like the Saez Optimal Taxation adjust dynamically, they fail to address taxpayer heterogeneity and irrational behavior. This study introduces TaxAgent, a novel integration of large language models (LLMs) with agent-based modeling (ABM) to design adaptive tax policies. In our macroeconomic simulation, heterogeneous H-Agents (households) simulate real-world taxpayer behaviors while the TaxAgent (government) utilizes LLMs to iteratively optimize tax rates, balancing equity and productivity. Benchmarked against Saez Optimal Taxation, U.S. federal income taxes, and free markets, TaxAgent achieves superior equity-efficiency trade-offs. This research offers a novel taxation solution and a scalable, data-driven framework for fiscal policy evaluation.

Figures

Figures reproduced from arXiv: 2506.02838 by the authors.

**Figure 2.** The social outcomes of all tax systems over 120 [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 4.** Sample tax rates for seven income brackets of the [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

Figures from the paper (2 more)

**Figure 5.** Figure 5: The side-effects of the TaxAgent on the macroeconomic [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]

**Figure 6.** Figure 6: Ablation study of the robustness of the TaxAgent. The [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

23 extracted references · 23 canonical work pages

[1]

Investment in education: do economic volatility and credit constraints matter?,

Karnit Flug, Antonio Spilimbergo, and Erik Wachten- heim, “Investment in education: do economic volatility and credit constraints matter?,” Journal of Development Economics, vol. 55, no. 2, pp. 465–481, 1998

work page 1998
[2]

On the impact of inequality on growth, human devel- opment, and governance,

Ines A Ferreira, Rachel M Gisselquist, and Finn Tarp, “On the impact of inequality on growth, human devel- opment, and governance,” International Studies Review , vol. 24, no. 1, pp. viab058, 01 2022

work page 2022
[3]

The injustice of inequality,

Edward Glaeser, Jose Scheinkman, and Andrei Shleifer, “The injustice of inequality,” Journal of Monetary Economics, vol. 50, no. 1, pp. 199–222, 2003

work page 2003
[4]

Effective policy for reducing inequality? the earned income tax credit and the distribution of income,

Hilary W Hoynes and Ankur J Patel, “Effective policy for reducing inequality? the earned income tax credit and the distribution of income,” Working Paper 21340, National Bureau of Economic Research, July 2015

work page 2015
[5]

The earned income tax credit (eitc),

Austin Nichols and Jesse Rothstein, “The earned income tax credit (eitc),” Working Paper 21211, National Bureau of Economic Research, May 2015

work page 2015
[6]

An exploration in the theory of optimum income taxation12,

J. A. Mirrlees, “An exploration in the theory of optimum income taxation12,” The Review of Economic Studies , vol. 38, no. 2, pp. 175–208, 04 1971

work page 1971
[7]

Optimal taxation and public production: I–production efficiency,

Peter Diamond and James Mirrlees, “Optimal taxation and public production: I–production efficiency,” Ameri- can Economic Review , vol. 61, no. 1, pp. 8–27, 1971

work page 1971
[8]

The design of tax structure: Direct versus indirect taxation,

A.B. Atkinson and J.E. Stiglitz, “The design of tax structure: Direct versus indirect taxation,” Journal of Public Economics, vol. 6, no. 1, pp. 55–75, 1976

work page 1976

Show all 23 references

[9]

Using elasticities to derive optimal income tax rates,

Emmanuel Saez, “Using elasticities to derive optimal income tax rates,” The Review of Economic Studies , vol. 68, no. 1, pp. 205–229, 01 2001

2001
[10]

Econagent: Large language model-empowered agents for simulating macroeconomic activities,

Nian Li, Chen Gao, Mingyu Li, Yong Li, and Qingmin Liao, “Econagent: Large language model-empowered agents for simulating macroeconomic activities,” 2024

2024
[11]

Process and critical approaches to solving the systemic climate change governance prob- lem,

Check Woo Foo, “Process and critical approaches to solving the systemic climate change governance prob- lem,” Politics & Energy eJournal , 2019

2019
[12]

Design and development of ad- vanced control strategies for power quality enhancement at distribution level,

Rajesh Kumar Patjoshi, “Design and development of ad- vanced control strategies for power quality enhancement at distribution level,” 2015

2015
[13]

The case for a progressive tax: From basic research to policy recom- mendations,

Peter Diamond and Emmanuel Saez, “The case for a progressive tax: From basic research to policy recom- mendations,” Journal of Economic Perspectives, vol. 25, no. 4, pp. 165–90, December 2011

2011
[14]

Optimal taxation of top labor incomes: A tale of three elasticities,

Thomas Piketty, Emmanuel Saez, and Stefanie Stantcheva, “Optimal taxation of top labor incomes: A tale of three elasticities,” American Economic Journal: Economic Policy , vol. 6, no. 1, pp. 230–71, February 2014

2014
[15]

Optimal income taxation with un- employment and wage responses: A sufficient statistics approach,

Kory Kroft, Kavan Kucko, Etienne Lehmann, and Jo- hannes Schmieder, “Optimal income taxation with un- employment and wage responses: A sufficient statistics approach,” American Economic Journal: Economic Pol- icy, vol. 12, no. 1, pp. 254–92, February 2020

2020
[16]

The ai economist: Improving equality and productivity with ai-driven tax policies,

Stephan Zheng, Alexander Trott, Sunil Srinivasa, Nikhil Naik, Melvin Gruesbeck, David C. Parkes, and Richard Socher, “The ai economist: Improving equality and productivity with ai-driven tax policies,” 2020

2020
[17]

507–547, University of Chicago Press, January 2018

Susan Athey, The Impact of Machine Learning on Economics, pp. 507–547, University of Chicago Press, January 2018

2018
[18]

Agent based modeling in eco- nomics and finance: Past, present, and future,

R. Axtell and J. Farmer, “Agent based modeling in eco- nomics and finance: Past, present, and future,” Journal of Economic Literature , 2022

2022
[19]

vii–v, Cambridge University Press, 2018

Domenico Delli Gatti, Giorgio Fagiolo, Mauro Gallegati, Matteo Richiardi, and Alberto Russo, Contents, p. vii–v, Cambridge University Press, 2018

2018
[20]

Phyx: Does your model have the

Hui Shen, Taiqiang Wu, Qi Han, Yunta Hsieh, Jizhou Wang, Yuyue Zhang, Yuxin Cheng, Zijian Hao, Yuan- sheng Ni, Xin Wang, Zhongwei Wan, Kai Zhang, Wen- dong Xu, Jing Xiong, Ping Luo, Wenhu Chen, Chaofan Tao, Zhuoqing Mao, and Ngai Wong, “Phyx: Does your model have the ”wits” fo...

2025
[21]

Competeai: Un- derstanding the competition dynamics in large language model-based agents,

Qinlin Zhao, Jindong Wang, Yixuan Zhang, Yiqiao Jin, Kaijie Zhu, Hao Chen, and Xing Xie, “Competeai: Un- derstanding the competition dynamics in large language model-based agents,” 2024

2024
[22]

A survey of large language models for financial appli- cations: Progress, prospects and challenges,

Yuqi Nie, Yaxuan Kong, Xiaowen Dong, John M. Mul- vey, H. Vincent Poor, Qingsong Wen, and Stefan Zohren, “A survey of large language models for financial appli- cations: Progress, prospects and challenges,” 2024

2024
[23]

Agent- based macroeconomics,

Herbert Dawid and Domenico Delli Gatti, “Agent- based macroeconomics,” Handbook of computational economics, vol. 4, pp. 63–156, 2018. APPENDIX a) An example of a complete prompt of a H-Agent: You’re Adam Mills, a 58-year-old individual living in San Antonio, Texas. A tax plann...

2018

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Investment in education: do economic volatility and credit constraints matter?,

Karnit Flug, Antonio Spilimbergo, and Erik Wachten- heim, “Investment in education: do economic volatility and credit constraints matter?,” Journal of Development Economics, vol. 55, no. 2, pp. 465–481, 1998

work page 1998

[2] [2]

On the impact of inequality on growth, human devel- opment, and governance,

Ines A Ferreira, Rachel M Gisselquist, and Finn Tarp, “On the impact of inequality on growth, human devel- opment, and governance,” International Studies Review , vol. 24, no. 1, pp. viab058, 01 2022

work page 2022

[3] [3]

The injustice of inequality,

Edward Glaeser, Jose Scheinkman, and Andrei Shleifer, “The injustice of inequality,” Journal of Monetary Economics, vol. 50, no. 1, pp. 199–222, 2003

work page 2003

[4] [4]

Effective policy for reducing inequality? the earned income tax credit and the distribution of income,

Hilary W Hoynes and Ankur J Patel, “Effective policy for reducing inequality? the earned income tax credit and the distribution of income,” Working Paper 21340, National Bureau of Economic Research, July 2015

work page 2015

[5] [5]

The earned income tax credit (eitc),

Austin Nichols and Jesse Rothstein, “The earned income tax credit (eitc),” Working Paper 21211, National Bureau of Economic Research, May 2015

work page 2015

[6] [6]

An exploration in the theory of optimum income taxation12,

J. A. Mirrlees, “An exploration in the theory of optimum income taxation12,” The Review of Economic Studies , vol. 38, no. 2, pp. 175–208, 04 1971

work page 1971

[7] [7]

Optimal taxation and public production: I–production efficiency,

Peter Diamond and James Mirrlees, “Optimal taxation and public production: I–production efficiency,” Ameri- can Economic Review , vol. 61, no. 1, pp. 8–27, 1971

work page 1971

[8] [8]

The design of tax structure: Direct versus indirect taxation,

A.B. Atkinson and J.E. Stiglitz, “The design of tax structure: Direct versus indirect taxation,” Journal of Public Economics, vol. 6, no. 1, pp. 55–75, 1976

work page 1976