Measuring the damage — Capital Asset Pricing Model to System Asset Pricing Model

As Chapter 1 establishes, the system-welfare coordinate C is not a function of the parties' payoffs, so the Hollow Win outcome¹ (C=0, A=1, B=1) can go undisplayed. A binary label is enough to diagnose that this is happening. It is not enough to manage it. A group decision support system built on the three requirements that follow from the MST needs to answer quantitative questions rather than one categorical one: how much system welfare degrades per unit of private gain, where on the feasible welfare boundary a given outcome sits, and what price an institutional designer should attach to system welfare to reach the efficient tradeoff. The System Asset Pricing Model (SAPM) supplies those continuous quantities and collects them into one measurement framework. The previous chapter gave a label; this one gives a rate: diagnosis tells you the coordinate is missing, measurement tells you how much its absence costs.

Welfare beta (β_W) is the first of those quantities. It measures the average rate at which system welfare is destroyed per dollar of annual industry revenue, and is defined as β_W = ΔW/Π, where ΔW is a domain's total annual system-welfare loss and Π is annual industry revenue, never profit². The choice of denominator carries the weight here: revenue is the gross private flow the system permits, so a welfare cost expressed per dollar of revenue is comparable across industries that run very different margins, while a profit denominator would not be. β_W is the quantity a group decision support system needs to separate mild Private-Systemic Tension from the catastrophic kind, because it converts a categorical worry into a rate that can be ranked and compared. Across the 58-domain ranked revenue-ratio portfolio, the revenue-weighted β_W is 2.61: each dollar of industry revenue is associated with about $2.61 of system-welfare loss. That destruction is reported across eight channels — a mortality channel valued at the value of a statistical life ($15.8T per year)³, health and disease ($12.7T per year), social and informational harm ($10.1T per year), climate and carbon ($8.4T per year), environmental and ecosystem loss ($6.5T per year), economic extraction and deadweight loss ($6.8T per year), governance and systemic failure ($5.4T per year), and lost productivity ($3.3T per year). These channels are contemporaneous annual flows; together they account for the roughly $72 trillion ranked-panel total. The ratio is an average quantity; the separate marginal expression −dW/dΠ answers the change-at-the-margin question for a corrective levy and is not the definition of average β_W.

β_W uses revenue because the numerator and denominator must share the same domain, time period, and activity boundary. The numerator is annual system-welfare loss, ΔW. The denominator is annual industry revenue, Π. Profit is a weak denominator because accounting choices, financing structure, tax strategy, and depreciation can move it away from the activity scale. Revenue is the activity denominator that can be matched to the harm boundary. The hard part was never the division. The hard part was making ΔW legitimate: naming the welfare channels, bounding the industry, using public sources, stripping transfers, correcting overlap, reporting uncertainty, and matching the harm to the revenue associated with that harm. Economists had externality estimates, social-cost estimates, health burdens, cleanup costs, fiscal losses, environmental accounts, and cost-benefit methods. The missing step was to organize those pieces into a repeatable industry-level ratio: annual system-welfare loss per dollar of revenue.

The canonical annual portfolio target is $69.0T⁴ — the net figure; the gross is $72.3T, from which $3.3T of cross-domain double-counting is removed once. It is scoped to the 58-domain ranked revenue-ratio panel. the exact total depends on the prices chosen to value each kind of harm, so this chapter teaches it as an order-of-magnitude measurement whose exact total depends on the shadow prices chosen (the value of a statistical life, the social cost of carbon)⁵.

The whole estimate is conditional on the price set. The headline $72 trillion is planner-relative and capture-sensitive: it moves with the dollar values a planner attaches to each kind of harm — a death, a cleanup, a lost year of health — and with the political pressures that shape those values. The domain ranking is conditional in the same way: a large enough change in a single price, such as the value of a statistical life, can move the total and reorder the domains alike, because different domains carry their harm through different channels. β_W measures welfare destroyed per dollar of annual industry revenue. The practical guidance for a regulator is therefore about direction, not decimals: act on the domains that come out clearly worst across a range of reasonable price sets, and read every figure — the total and the ordering both — as an order-of-magnitude measurement rather than a settled number.

A very high β_W is a diagnostic signal, and firearms show why it can be real. The numerator is annual system-welfare loss from the firearm activity being measured: mortality, nonfatal injury, emergency care, long-term disability, policing, legal-system costs, lost productivity, trauma, insurance burden, and other source-backed public costs, with overlap corrections so the same harm is counted once. The denominator is the revenue boundary for that same activity. For firearms, the current revenue-basis calibration uses ΔW = $509.9B and Π = $23.2B for the U.S. civilian firearms revenue boundary, producing β_W = 21.98. The 100,000-draw Monte Carlo⁶ (a simulation that recomputes the estimate many times with randomly drawn inputs) interval is [17.5, 27.7], so the conservative lower bound reported by the simulation is 17.5. Even near the bottom of the modeled range, firearms impose about $17.50 of system-welfare loss per $1 of civilian firearms revenue. The methodological answer is reproducibility: show the numerator, show the denominator, show the boundary, show the interval, and show the conservative lower bound.

Paper 7 — the public-data asset-pricing test — has now been run on public data⁷, and the result is mixed but theoretically cleaner than a uniform confirmation would have been. It tests four predictions. Prediction 1 (that the most system-destructive firms earn no unusual return in normal times, before markets learn to price system welfare) survives: the high-minus-low β_W portfolio shows no detected normal-times alpha. The honest reading is that this is the theory's own structural prediction confirmed, not a negative result: system welfare W is not a function of the parties' payoffs, so a market that prices the bilateral payoff vector shows no β_W premium until W is forced into the price from outside — through Decision Accounting, financial-statement integration (FASB/IASB—the US and international accounting-standards boards), and the resulting cost-of-capital advantage — a partial, multi-year process that never reaches every firm. Prediction 2 (that welfare-beta predicts how sharply a firm's stock reacts when a restoration event forces the hidden system harm into view) survives strongly: β_W predicts the magnitude of restoration-event returns even after the Fama-French-Carhart four-factor model factor⁸, sin-stock, litigation,⁹ and green-minus-brown controls¹⁰. Prediction 3 (that the most system-destructive firms should have more downside, negative-skew returns) fails: high-β_W skew is more positive, not more negative, which is the opposite of what a hidden-tail-risk story would require. Prediction 4 (that firms sharing the same damaged system should have returns that move together) fails: shared-system membership does not produce a positive within-system residual covariance load. The conclusion the paper draws is bounded: the data support SAPM and the MST as a structural no-premium and restoration-loading account, but without yet establishing a broad hidden-tail-risk or non-diversifiable system-covariance asset-pricing theory. The four outcomes are results from the public-data test, awaiting independent replication.

The finance beta and the policy beta are distinct, and the cleanest way to see it is to write the actual pricing relation. In the asset-pricing paper, the System Asset Pricing Model (SAPM) is a two-factor equation for a single firm's expected return: E[R_i] - R_f = β_i^mλ_m + β_i^Wλ_W Read it for one company, firm i. Its expected return above the risk-free rate (E[R_i] - R_f) is explained by two pieces. The first, β_i^mλ_m, is ordinary market risk: how much firm i's returns move with the overall market (β_i^m), times the market's risk premium (λ_m = E[R_m] - R_f). That first term, all by itself, IS the Sharpe-Lintner CAPM¹¹. The second, β_i^Wλ_W, is the new part: how much firm i's returns move with system-welfare shocks (β_i^W), times the price the market puts on system-welfare risk (λ_W). Now set the system-welfare price to zero, λ_W = 0, and the entire second term disappears: E[R_i] - R_f = β_i^mλ_m That is CAPM, exactly. SAPM does not replace CAPM; it contains it — CAPM is the special case you get when the market prices system welfare at nothing¹², under the paper's stated special-case restriction (λ_W = 0); that restriction is separate from the policy-average ratio β_W. Notice that every β here carries the subscript i: this is a firm-by-firm relation, because each company has its own market loading and its own welfare loading. You run it for one company — not, as you never would, for a whole industry at once. The policy β_W used throughout this curriculum is a different quantity. It is not a return loading: β_W = ΔW / Π, annual system-welfare loss divided by annual industry revenue on the same boundary — an industry-level ratio of how much welfare an activity destroys per dollar of revenue. So the two 'betas' answer different questions. The finance beta (β_i^W) asks how a firm's stock returns load on a welfare factor; the policy beta (β_W) asks how much welfare loss an activity produces per dollar of revenue. Litigation, regulation, insurance repricing, disclosure, financing limits, or reputational events can make system-welfare exposure enter expected cash flows and market prices — pushing λ_W and β_i^W away from zero — but those channels are partial and event-dependent.

The Environmental Protection Agency finalized a PFAS (per- and polyfluoroalkyl substances, long-lasting synthetic 'forever chemicals') drinking-water rule and estimates its compliance cost at about $1.63 billion a year; water-sector groups put it higher, at $3.8–5.5 billion a year¹³. That cleanup is one visible channel, not the whole system-welfare cost. PFAS carries a measured welfare-beta of 0.96: against roughly $32 billion in annual PFAS-related revenue, the annual system-welfare loss is on the order of $31 billion, so welfare cost sits close to revenue. That is what a β_W near 1 means, and it is why PFAS is a control-group boundary case rather than a large multiple of revenue. The $1.5 trillion figure often attached to PFAS is a different quantity — the Lancet Countdown's estimate of annual health losses from plastics generally — not PFAS cleanup.

Field 15 is ALTERNATIVES. Field 16 is PREDICTION. Field 17 is SYSTEM WELFARE.¹⁴ Field 14 records the alternatives considered; Field 16 records a scoreable prediction with its metric, horizon, uncertainty, and action; Field 17 records the system boundary, evidence, uncertainty, time horizon, and review trigger for the welfare consequence. The numbering is fixed at seventeen fields.