GPT 6 Astra Ultra attempt on Econ-409

Optimal top-income taxation with avoidance and mobility

Research target and a restricted accounting result

The target optimizes a top marginal rate jointly with enforcement and reclassification policy, accounting for all displaced tax bases and a fixed initial population. A local elasticity of reported top income cannot identify that global optimum. We derive a transparent two-base marginal formula, show why identical reported-income responses imply different welfare effects, and provide a conditional robust-policy certificate. The formula is a benchmark, not a universal tax recommendation.

A two-base economy with real and shifting responses

Use one unit mass of identical taxpayers for the first calculation. Let $L$ be real earnings above a normalized zero threshold, $S$ earnings reclassified to a second base taxed at fixed rate $\sigma$, and $X=L-S$ the top-rate base. For fixed enforcement $e$, each taxpayer chooses feasible $(L,S)$ to maximize \[ V(\tau,e)=\max_{L,S}\{(1-\tau)L+(\tau-\sigma)S-c(L)-k(S,e)\}. \tag{1} \] Assume an interior unique smooth optimum, with convex real effort cost $c$ and shifting resource cost $k$. Taxpayer utility is quasilinear in the common numeraire. The top rate changes no other primitive. Revenue and welfare are \[ R=\tau X+\sigma S,\qquad W=gV+\lambda[R-C(e)], \] where $g\ge0$ is a fixed social weight and $\lambda>0$ is the public-funds weight. All receipts are included once. In particular $V$ is already after-tax private welfare; adding pretax income separately would double-count it. Real earnings reductions destroy revenue at marginal rate $\tau$, whereas pure shifting loses only the rate difference $\tau-\sigma$. This distinction is an accounting consequence conditional on the model, not an empirical claim that shifting is harmless. The resource cost $k$ and public enforcement cost $C$ remain present in welfare.

Same observed elasticity, different marginal welfare

At a common baseline let $X=100$, $\tau=0.5$, $\sigma=0.4$ and $X_\tau=-100$. Consider local model A with $L_\tau=-99,S_\tau=1$ and model B with $L_\tau=-1,S_\tau=99$. The reported tax base and its derivative coincide. With $\lambda=1,g=0.7$, (2) equals $30-49.5-0.1=-19.6$ in A and $30-0.5-9.9=19.6$ in B. These local responses can be generated by the quadratic primitives in the next section, choosing intercepts to match baseline $X$. Thus even the sign of the marginal welfare gain is not identified by the reported-income elasticity.

A mobility margin adds a further distinction. If the resident mass is $N(\tau)$, per-resident bases are $x,s$, and revenue is $N(\tau)(\tau x+\sigma s)$, then \[ R_\tau=N\{x+\tau x_\tau+\sigma s_\tau\} +N_\tau(\tau x+\sigma s). \tag{3} \] The last term is lost domestic revenue, while the welfare of emigrants stays inside a fixed-initial-population criterion. Dropping them from the utility integral changes the question. Their optimized migration costs and foreign incomes require a residence model; neither (2) nor a taxable-income elasticity identifies them. We do not add (3) to (2) while silently retaining its fixed-mass envelope formula.

A computable joint rate/enforcement benchmark

Let \[ c(L)=\frac{(L-L_0)^2}{2b},\quad k(S,e)=\frac{(\kappa_0+e)S^2}{2},\quad C(e)=\gamma e^2/2, \] with $b,\kappa_0,\gamma>0$. Restrict the policy to a compact rectangle $\tau\in[\sigma,\bar\tau]$, $e\in[0,\bar e]$, $\bar\tau<1$, and choose constants ensuring $0\le S<L$ on it. Then strict concavity of (1) gives \[ L=L_0+b(1-\tau),\qquad S=\frac{\tau-\sigma}{\kappa_0+e}. \] The optimized utility is \[ V=(1-\tau)L_0+\frac b2(1-\tau)^2+ \frac{(\tau-\sigma)^2}{2(\kappa_0+e)}. \] Substitute these rational expressions into $W$ to obtain a continuous function on the rectangle, so an optimum exists. From the envelope and revenue formulas, \[ W_e=-gS^2/2+\lambda(\tau-\sigma)\frac{S}{\kappa_0+e} -\lambda\gamma e. \tag{4} \] The first term records the extra real private shifting cost for given choices; the second records revenue recovered because shifting falls. Its sign is not fixed by saying enforcement reduces avoidance.

There is no need to assume global concavity to certify an approximate optimum. Bound the two absolute partial derivatives of the explicit $W$ over the compact rectangle by $L_\tau,L_e$, using $\kappa_0+e\ge\kappa_0$. On a rectangular mesh with spacings $h_\tau,h_e$, choose the largest computed value, with each value error at most $\eta$. Rounding any policy to a mesh point changes welfare by at most $L_\tau h_\tau+L_e h_e$. Hence the chosen policy has regret at most \[ L_\tau h_\tau+L_e h_e+2\eta. \tag{5} \] This is a finite global certificate, albeit computationally expensive in a higher-dimensional institution-specific problem.

Identified models and robust choice

If parameters lie in a data-consistent compact set $\Theta$, define $\underline W(p)=\inf_{\theta\in\Theta}W(p;\theta)$ and $\overline W(p)=\sup_{\theta\in\Theta}W(p;\theta)$. For any candidate policy $p_*$, \[ \sup_{\theta\in\Theta} [\sup_pW(p;\theta)-W(p_*;\theta)] \le\sup_p\overline W(p)-\underline W(p_*). \tag{6} \] Proof: each term in brackets is bounded by the right side before taking the supremum. This sufficient certificate is conservative because its upper and lower endpoints may use different models. If every $W$ is within $\delta$ of a fitted $\widehat W$ uniformly and $p_*$ is $\eta$-optimal for the latter, its regret is at most $2\delta+\eta$ by adding and subtracting fitted values. Neither bound removes uncertainty without credible data restrictions.

Linked all-base records, independently varied enforcement, migration histories and a consistent welfare boundary are necessary to evaluate the expanded problem. Bargaining rent transfers require modeling the recipients who lose those rents. The present proof addresses real responses and reclassification but not that channel. The substantive unresolved task is to justify a joint identified class and obtain useful small regret under those richer margins.

Completion Estimate

45%

This is a post hoc, heuristic estimate for the full catalogue target. The attempt proves a two-base marginal welfare formula, demonstrates elasticity-based welfare ambiguity, and certifies joint rate and enforcement optimization in a quadratic benchmark. A data-supported joint model covering mobility, bargaining transfers, evasion, and all displaced tax bases remains necessary for the full policy target.

References

• [source] T. Piketty, E. Saez and S. Stantcheva (2014), Optimal Taxation of Top Labor Incomes: A Tale of Three Elasticities, AEJ: Economic Policy 6(1), 230--271. https://www.aeaweb.org/articles?id=10.1257/pol.6.1.230. Publisher abstract verifies the real, avoidance and bargaining distinctions; (1)--(6) are a separately specified and derived benchmark, not the paper's full optimal-tax model or a claimed new general formula.