| paper | d4-journals-jet-2003-0019 |
| authors | — |
| venue | ? |
| filed under | unclassified |
| judged by | gpt-6-luna / xhigh (triple__luna__xhigh__c6r1) |
| judge confidence | high |
| authors would recognise it | yes |
Proposition 2.1 proves that finite-support allocations attain the same value as general allocations, but does not establish that an optimum can be computed or give a complexity result. The proposed LP is a plausible computational extension, yet it works over a population the paper already models as a continuum.
fails bit a — no named computational result to mirror
The proposed problem covers a bounded, rational piecewise-linear restriction of the Section 2 adverse-selection planner; it leaves the paper’s smooth, unbounded cases and Section 3 moral-hazard results outside.
The strongest positive case is a computational continuation of the paper’s adverse-selection planner problem. It is a weak case for population continuization as a novelty: Section 2 already assumes a continuum of agents. What the paper does not supply is a bit-complexity classification for computing its optimal allocations from finite input.
My sole anchor is Proposition 2.1, proved in this paper. It says that restricting the planner’s dual problem \((D)\) to allocations with finite support gives a problem \((D_F)\) that has an optimal solution and has the same optimal value: \(\nu(D_F)=\nu(D)\). The paper has no named result saying that a problem is in P, NP-hard, or in another complexity class. Proposition 2.1 is an optimization result, not a polynomial-time bound; it makes a finite-output computational question well grounded.
A plausible high-multiplicity regime is a large group insurance pool with two hidden risk classes. Policyholders within each class share the same risk, endowment, preferences, and contract terms. Millions of people may therefore be summarized by two rational population shares, \(\xi_L\) and \(\xi_H=1-\xi_L\), so \(\tau=2\) even as the number of people grows. This is an instance regime, not a claim that every insurance market has so few types.
Here is a precise mirror, the Two-Risk Mass Insurance problem. An instance gives rational \(\xi_L,\theta_L,\theta_H,\omega_1,\omega_2,B,\gamma\), and rational concave piecewise-linear utilities \(U_L,U_H\) on \([0,B]\), plus a rational welfare threshold \(K\). A contingent contract is \(c=(c_1,c_2)\in[0,B]^2\), and its expected utility and expected net trade for type \(i\) are
\[
E_i(c)=\theta_i U_i(c_1)+(1-\theta_i)U_i(c_2),\qquad
r_i(c)=\theta_i(c_1-\omega_1)+(1-\theta_i)(c_2-\omega_2).
\]
The question is whether there are lotteries \(\pi_L,\pi_H\) over contracts such that
\[
\int E_L\,d\pi_L\ge\int E_L\,d\pi_H,\qquad
\int E_H\,d\pi_H\ge\int E_H\,d\pi_L,
\]
\[
\xi_L\int r_L\,d\pi_L+(1-\xi_L)\int r_H\,d\pi_H\le 0,
\]
and
\[
\gamma\int E_L\,d\pi_L+(1-\gamma)\int E_H\,d\pi_H\ge K.
\]
A solution is a pair of contract lotteries satisfying these constraints; the search version returns an optimal pair.
This is the paper’s adverse-selection planner problem with a finite binary description of preferences and a bounded contract domain. The population object is the type mass \(\xi\); the lotteries are the paper’s allocation choice, not the claimed source of continuity. The problem is in Class A for this representation: take the union \(S\) of the utility breakpoints and the endpoints \(0,B\). On each rectangle between adjacent points of \(S\times S\), both expected utilities and net trades are affine in the contract. Any contract in the rectangle can therefore be replaced by a lottery over its corners with the same utility and resource moments. The optimization reduces exactly to a rational LP over \(S\times S\), of size polynomial in the input bit length and number of breakpoints. It gives an optimal finite-support allocation, in line with Proposition 2.1.
The scope is Section 2’s two-type adverse-selection result and its planner allocation, not the paper’s moral-hazard results. The weakest point is that this is not a new continuous-population model: the paper already has one. And the P result covers the bounded, piecewise-linear computational representation, not the paper’s full smooth, unbounded LSIP. The positive claim is narrower: the paper identifies an optimization problem and finite-support optima, while leaving its algorithmic complexity for explicit rational inputs unclassified. Further questions include whether useful complexity bounds survive removal of the contract cap, and what changes when the number of privately informed types is part of the input.
The proposed mirror fails at the population-continuization gate: this paper has already made the population continuous. Section 2 explicitly has a continuum of agents with type shares \(\xi_L\) and \(1-\xi_L\); its allocation measures describe the distribution of contracts within each type. Section 3 likewise starts with a measure-one continuum of agents. The group-insurance scenario is therefore a plausible high-multiplicity regime, but it is the paper’s own model, not a discrete model being mirrored.
Proposition 2.1 does not supply the missing bridge. It says that optimal allocation measures can be restricted to finite support. That is a statement about how many contracts an allocation needs, not about how many voter or agent types describe the society. Nor does it give a bit-complexity result: the contracts have real-valued coordinates, and the paper permits smooth utility functions without specifying finite encodings for them. The proposition is an optimization theorem, not a computational classification.
The proposed Two-Risk Mass Insurance problem makes those encoding choices explicit and obtains a clean result. With two types, two states, bounded contracts, and rational piecewise-linear utilities, the corner-grid LP is a natural algorithm. But this is a computational formalization of an economy whose society is already continuous. Its tractability comes from the piecewise-linear representation of utility; population continuization does no new work in the reduction.
A more faithful computational version could retain the paper’s smooth utilities, but then it needs a specified representation and access model. An oracle model needs precision and query assumptions; a formula-based model makes complexity depend on the chosen formula language and its size. Either could support a worthwhile study of the planner’s LSIP, but neither makes this paper a continuous mirror. Adding a growing number of risk types or states could also create a richer complexity landscape, yet that changes the paper’s fixed two-type, two-state setting.
This is not an argument that computational work on the economy would be uninteresting, or that its answer being polynomial makes it unworthy. It is the narrower scope objection: the paper’s named result already concerns a continuous population, and Proposition 2.1 concerns finite support over assignments. The positive case itself calls its proposal a “computational continuation.” That may be a good research extension, but it does not identify a population continuization mirror of this paper. The negative case is strongest on that classification point; it does not rule out valuable computational research on the model.
The adversarial triple: the proponent anchors on up to three named results; the opponent sees that case and must defeat every anchor; the judge decides which case convinced it. These are the pipeline’s own outputs, generated by tools/triple_run.py — no human edited them. The paper’s own text is not reproduced here beyond the quoted statement above.