Is System Welfare Circular? Diagnosis, Metric, and the Independence of the Welfare Estimand

The circularity objection fails because diagnosis, measurement, and data sources are separate

The paper concedes that W appears in both Hollow Win diagnosis and β_W measurement. It rejects circularity by separating a sign claim from a magnitude claim, then showing that ΔW is estimated from harm channels independent of revenue.

• Hollow Win is a diagnosis: W is degraded below threshold C = 0 while A and B gain
• β_W = −ΔW/Π is a metric: welfare loss per dollar of audited revenue
• ΔW is built from mortality, morbidity, environmental, governance, and intergenerational harm channels
• Π enters only as the denominator after ΔW has been estimated

The strong objection says ΔW is just revenue in disguise

The paper names three versions of the circularity objection and focuses on the two that matter. The fatal version would hold only if the welfare loss estimate were determined by revenue.

• Weak form: W is defined inside the framework, as theoretical quantities usually are
• Moderate form: Hollow Win and β_W both use W, so the framework may be untestable
• Strong form: ΔW = f(Π), making β_W = −f(Π)/Π a scale-only ratio
• The paper answers the moderate form with sign-versus-magnitude and the strong form with a falsification test

The paper admits the repeated W and then narrows what must be proved

The objection has force only because W appears in two places. The answer is not that the symbol differs, but that the claims and estimands differ.

• In the eight-outcome taxonomy, Hollow Win (0,1,1) means C = 0 while both bilateral parties gain
• In the metric, β_W measures welfare destroyed per unit of private revenue
• If the framework only said W is destroyed and β_W measures destroyed W, it would be circular
• The paper’s burden is to show that ΔW is constructed independently of the diagnosis and of Π

A Hollow Win says W is negative; β_W says how large the loss is per revenue dollar

The diagnosis and the metric are independently contestable. One can accept that a threshold was crossed and dispute the β_W estimate, or accept a β_W estimate and dispute whether C = 0 was reached.

• Diagnosis: the sign of dW is negative and the system coordinate has crossed the degraded threshold
• Measurement: β_W reports the magnitude of ΔW divided by audited annual revenue
• The paper uses hypertension as the analogy: elevated blood pressure and a 145/95 reading are different claims
• This separation makes the framework testable rather than self-confirming

ΔW is estimated from five harm channels, not read off the Hollow Win label

The decisive move is empirical: the numerator is constructed domain by domain from named mechanisms and external data sources, while the denominator is audited revenue.

• Premature mortality: VSL from labor-market and stated-preference literature
• Morbidity: DALYs or QALYs from epidemiological studies
• Environmental damage: contamination concentrations, persistence half-lives, ecosystem services, biodiversity loss
• Governance and trust losses: fraud dollars, enforcement actions, risk premia, compliance costs
• Intergenerational harm: present value of harms transferred to future periods

Three framework features keep Π outside the ΔW estimate

The paper identifies three safeguards that make the numerator and denominator come from different data-generating processes.

• Revenue, never profit: Π is audited gross revenue, not a residual affected by litigation, regulation, or accounting choices
• Non-derivability: W-Independence says W cannot be expressed as a function of agents’ payoffs
• Exogenous simulation inputs: Monte Carlo low, central, and high values come from cited empirical literature
• The simulation can be run with revenue withheld and still return the same ΔW

The simulation estimates ΔW before β_W is calculated

Appendix B defines ΔW as a channel-based simulation. Revenue is not a parameter in the input distributions; it is used only after the welfare-loss estimate exists.

• Identify harm channels H1 through Hk from peer-reviewed literature and regulatory impact analyses
• Specify distributions: lognormal for mortality counts, beta for morbidity DALYs, triangular for cost estimates
• Draw samples and compute ΔW = Σj wjHj
• Use social cost weights such as VSL and social cost of carbon, not revenue weights
• Divide by Π only at the final β_W step

β_W is circular if and only if harm channels are revenue-determined

The paper turns the objection into a testable condition. If revenue determines harm, ΔW should be explained almost entirely by Π across domains.

• Definition: harm is revenue-determined if there exists f such that ΔW = f(Π) for all covered activities
• Regression: ΔWi = α + βΠi + εi
• Circularity prediction: β > 0 and R² ≈ 1
• Independent-identification prediction: β may have any sign, but R² << 1
• Reported result: R² is below 0.15, with large heteroskedastic residuals

Comparable revenues produce very different β_W values

is the empirical signature that ΔW is not Π in disguise. Activities near the same revenue scale have different welfare-loss rates because their harm mechanisms differ.

• Opioid analgesics, US 2017: Π = 12.5B, ΔW = 438B, β_W = 35.0
• Coal-fired power, US 2019: Π = 18.2B, ΔW = 312B, β_W = 17.1
• PFAS manufacturing, global 2020: Π = 2.8B, ΔW = 187B, β_W = 66.8
• Sugar-sweetened beverages, US 2018: Π = 78.4B, ΔW = 89B, β_W = 1.1
• Fast fashion, global 2022: Π = 1.2T, ΔW = 1.8T, β_W = 1.5

The strong objection would require mortality, pollution, and fraud to be deterministic revenue functions

The paper rejects the strong form by naming what would have to be true. Revenue would need to determine every harm channel across all domains, which the examples contradict.

• Mortality case: a life-saving drug and an addictive opioid can have the same revenue but opposite mortality effects
• Environmental case: a solar panel manufacturer and a coal plant can have the same revenue but different emissions, mining, and waste profiles
• Governance case: an honest bank and a fraudulent bank can have the same revenue but different trust losses
• The mechanisms are chemical properties, production processes, oversight, use patterns, and biological effects, not revenue itself

β_W can rank harmful activities only where harm channels are independently measurable

The policy claim is bounded. β_W is usable when mortality, morbidity, environmental, governance, or intergenerational harms can be estimated from sources independent of revenue, and it remains exposed to the regression test.

• Regulators can compare harmfulness across activities with different scales because β_W is not a scale-only metric
• Critics must show that the relevant harm channels are revenue-determined, not merely that W appears twice
• Future applications would weaken the framework if cross-domain R² moved close to 1
• Sparse harm data can be handled through Monte Carlo uncertainty, but the inputs still must come from outside Π