Audited final · corporate ethics and the micro/macro foundations of strategy
Whose Losses Count
One decision runs underneath every number in this file: where the accounting boundary sits, and whose losses fall outside it.
Source base: the 6 August 2026 raw dossier compiled earlier in this session, sections 1–7. No new research. Every factual claim traces to that material; framing added by the author is tagged interpretation.
The hook
That is not a moral failure of markets. It is markets working exactly as specified: prices aggregate the preferences of people who trade, and a downwind neighbour is not trading.
Which makes the second finding the uncomfortable one. Regulation is the only mechanism the retrieved evidence shows binding on stranger-harm, and across 2025–26 it is being withdrawn on nearly every margin at once: NZBA dissolved, NZAM's 2050 target dropped, the SEC's climate rule proposed for rescission, CSRD scope cut, CSDDD's transition-plan duty deleted, roughly 7% of shareholder proposals passing versus 14%.
The load is shifting to a mechanism that demonstrably does not carry it. interpretation
How this is built
Eight questions, one shape
| Slot | What it does |
|---|---|
| Takeaway | The section in one sentence, on a tinted plane. Nothing else in the section contradicts it. |
| The anomaly | Why competent people, looking at the same evidence, reach opposite conclusions. Marked with a red rule. |
| The model, as a behavioural claim | The textbook result restated as something that could be false, with its assumption and the conditions under which it observably is. |
| Evidence | Tables and sourced claims. Every figure carries the grade it was retrieved at. |
| ⩢ Boundary | The recurring move: which losses this section's number counts, and which it does not. This is the spine of the document. |
| Still contested | What the section does not settle, and the specific reason competent people disagree. |
Evidence grades, as retrieved
- verified
- Traced to the dossier at the section given, with the source named.
- medium
- Retrieved, but through a summary extraction or a secondary citation chain. Flagged in the audit at the end.
- inferred
- Derived from retrieved figures rather than stated by a source.
- textbook theory
- Standard theory, carried for the structure of the argument rather than as a finding.
- interpretation
- The author's framing, not a claim any source makes. Bordered in red for the same reason the figures are: it sits outside the evidence boundary.
Colour does exactly one job on this page. Blue is inside the accounting boundary, counted and priced. Red is outside it. A solid mark is a measured quantity; an open outline is a quantity a source states qualitatively, or reports as not distinguishable from zero. Both figures ship their numbers in the surrounding text, so nothing depends on seeing a colour.
Section 0
One argument, not three courses
Watch the same move recur. In micro it is the difference between marginal private cost and marginal social cost. In macro it is who statutorily pays a tariff versus who economically bears it. In corporate law it is who has standing to sue. In finance it is which cost line counts as “variable” when you estimate a markup. The technical literatures look separate. They are one question asked in four vocabularies: who is inside the number?
The boundary is never neutral, and it is never purely technical. Choosing it is the value judgement. Every instance below is marked ⩢ BOUNDARY.
Question 1
What counts as a cost?
The anomaly
Two competent economists, the same Compustat data, the same decades: markups either tripled or barely moved. That is not a dispute about the world. It is a dispute about a definition.
Lerner: L = (P − MC)/P = 1/|ε|, and markup µ = P/MC
textbook theory.
The model, stated as a behavioural claim
- Claim
- Firms price by how easily customers leave.
- Assumption
- You can observe marginal cost.
- Observably wrong when
- You cannot. So everything hinges on which input you declare variable.
Worked example, my arithmetic
Elasticity ε = −4 gives L = 0.25. With MC = USD 75.00, P = MC/(1 − L) = USD 100.00, so µ = 1.33.
Intuition in one line: pricing power is the inverse of exit ease. Brand, switching costs and lock-in are investments in lowering |ε|.
| Estimate | What counts as variable cost | Result |
|---|---|---|
| De Loecker, Eeckhout & Unger, QJE 2020 | COGS only | “rising from 21% above marginal cost to 61%” (1980→2016) |
| Traina 2018 | COGS + SG&A | “the increase largely disappears when we include SGA costs” |
| Basu, JEP 2019 | — | “Existing methods cannot determine whether markups have been stable or whether they have risen modestly” |
| Benkard, Miller & Yurukoglu, NBER WP 34513 | DLEU's, restored | “sensitive to unreported sample restrictions that drop 27% of the available observations”; ex-Finance, ~1.30 → ~1.35 |
All §4.1.
⩢BOUNDARY
Note what SG&A is: advertising, sales, brand. Traina's move is to say the thing that lowers |ε| is a cost of production. DLEU's is to say it isn't. interpretation
The same boundary question breaks the concentration debate. Manufacturing payroll-to-value-added fell approximately 16 percentage points (1982–2012) via “between-firm reallocation” to superstar firms Autor, Dorn, Katz, Patterson & Van Reenen, QJE 2020. But 78% of US employment sits in industries where local concentration fell Rossi-Hansberg, Sarte & Trachter, and 2022 Economic Census average C4 was 34.6% against 34.3% in 2017 ITIF, an industry-aligned source, flagged as such. Draw the market at the nation and you see monopoly; draw it at the ZIP code and you see entry.
Still contested, and why
DLEU have not released replication code, per the rejoinder; their own reply was robots-blocked and never read in this research. Competent people disagree because there is no observable marginal cost, only accounting categories, and no category is theory-free.
Question 2
What counts as a victim?
The anomaly
The same literature is cited to prove opposite things. It shouldn't be. It's two different harm types.
| Harm | Victim | Penalty beyond law | Source |
|---|---|---|---|
| Financial misrepresentation | Investors | Reputational loss “over 7.5 times the sum of all penalties imposed through the legal and regulatory system”; per $1 of inflated value, $0.36 legal + $2.71 reputational | Karpoff, Lee & Martin, JFQA 2008 |
| Second-party wrongs (customers, investors) | Counterparty | −2.62% abnormal returns, significant; “reputational losses are nearly nine times the size of fines” | Armour, Mayer & Polo |
| Environmental violations | Third parties | “environmental violations are disciplined largely through legal and regulatory penalties, not through reputational penalties” | Karpoff, Lott & Wehrly, JLE 2005 |
| Third-party wrongs generally | Non-trading strangers | +0.24%, not significant: “no impact on stock prices” | Armour, Mayer & Polo |
All §7.5. The dossier flags Karpoff, Lott & Wehrly (2005) as frequently miscited as evidence for large reputational penalties. It shows the opposite.
Figure 1
Reputation is a ferocious enforcement mechanism and a structurally partial one
Two instruments, two axes, because they measure different quantities and forcing them onto one would fake a comparison the studies do not support. Panel A is the market's price of the harm. Panel B is what a dollar of inflated value costs.
The model, stated as a behavioural claim
- Claim
- Markets discipline misconduct, so voluntary standards can substitute for rules.
- Assumption
- The injured party is someone whose future custom the firm needs.
- Observably wrong when
- The injured party is downwind, downstream, or unborn.
⩢BOUNDARY
This is the externality condition, MSC > MPC, stated as an asset-pricing result rather than a diagram. textbook theory verified interpretation
The law draws the same line, deliberately
As a decision path, all §7.1–7.2.
- Ordinary operations → business judgment rule. Stakeholder spending is fine if rationally tied to shareholder value.
- Policy avowedly against stockholder value → breach. eBay v. Newmark (Del. Ch. 2010): “The ‘Inc.’ after the company name has to mean at least that.”
- Sale or change of control → Revlon (Del. 1986): directors become “auctioneers charged with” best price; stakeholder concern is “inappropriate when an auction among active bidders is in progress.” Narrow and transaction-specific.
- Public benefit corporation → DGCL §365(a) mandates balancing, then §365(b) removes the teeth: a director “shall not… have any duty to any person on account of any interest of such person in the public benefit.” No stakeholder standing; rationality review. Revlon does not apply to PBCs (Drakes Landing, Del. Ch., 29 Jul 2026). Nineteen publicly traded PBCs existed at 31 December 2023, excluding SPACs (Dammann).
- Dodge v. Ford (1919) is Michigan, not Delaware, and Stout–Bainbridge is live. Do not cite it as settled.
And the enforcement that remains is priced as an operating expense
“Recidivist firms bear fines that are, on average, twice the size of those borne by nonrecidivist firms” yet the authors call them “inconsequential ‘parking tickets’” Lund & Sarin. Then-CFPB Director Rohit Chopra, March 2022: “Corporate recidivism has become normalized and calculated as the cost of doing business.” Six parent companies, all banks, have each paid $1m or more in over 100 separate cases; Bank of America leads with 225 Good Jobs First. Headline numbers flatter: of Boeing's “over $2.5 billion” 2021 resolution, the criminal penalty was “$243.6 million,” approximately 9.7%, while USD 1.77bn went to airline customers, that is, to commercial counterparties DOJ.
Still contested, and why
Whether Dodge's shareholder-primacy line is holding or dicta (Stout 2008 versus Bainbridge 2022). Competent people disagree because the case both announces primacy and defers to the board on the operating question in the same opinion.
Question 3
What counts as performance?
The anomaly
Equity-linked pay is the standard cure for the agency problem. It is also, in the retrieved evidence, a cause of the manipulation it was meant to prevent.
The chain, all verified.
- 401 CFOs. “A surprising 78% of the surveyed executives would give up economic value in exchange for smooth earnings.” “80% of survey participants report that they would decrease discretionary spending on R&D, advertising and maintenance.” And when a positive-NPV project would break consensus, “Only 59% of the respondents would take the project” 41% decline value to hit a number Graham, Harvey & Rajgopal 2005.
- Pay structure moves accruals. Earnings management “is more pronounced at firms where the CEO's potential total compensation is more closely tied to the value of stock and option holdings”; moving from the 25th to the 75th percentile of the incentive ratio adds approximately +200bp of absolute accruals over assets at firms under USD 1bn in assets Bergstresser & Philippon, JFE 2006.
- In the lab. “People with unmet goals were more likely to engage in unethical behavior than people attempting to do their best,” and the effect is strongest among those who just miss Schweitzer, Ordóñez & Douma, AMJ 2004.
- In the field. Approximately 5,300 Wells Fargo employees terminated for sales-practice violations, 1 January 2011 to 7 March 2016; leadership ran “50/50 plans” expecting only half of regions could meet them. DOJ: “onerous sales goals and accompanying management pressure led thousands of its employees to engage in unlawful conduct.” Penalties: USD 185m in 2016 (CFPB $100m + OCC $35m + Los Angeles $50m); USD 3bn in February 2020.
- The drift. “A cultural drift in which circumstances classified as ‘not okay’ are slowly reclassified as ‘okay’” Vaughan.
- The base rate. “In an average year, 10% of all large public corporations commit (alleged) securities fraud, with a 95% confidence interval between 7 and 14%,” and “Our best estimate of the detection likelihood is 0.33” so enforcement counts see roughly one case in three inferred from 0.33. Cost approximately “1.6% of the total equity value of US public firms,” about USD 830bn at 2021 values Dyck, Morse & Zingales, RAS 2024.
The model, stated as a behavioural claim
- Claim
- Tie the agent's wealth to the share price and interests align.
- Assumption
- The share price is an honest signal the agent cannot bend.
- Observably wrong when
- The agent controls the accruals that produce the price.
⩢BOUNDARY
The metric decides what counts as performance, so anything outside the metric is free to sacrifice. Graham, Harvey and Rajgopal's 80% is that sacrifice, measured.
Still contested, and why
Whether short-termism is materially costly at all. Roe's position, as retrievable, is that studies finding it “generally find the inefficiencies to be small” and the evidence is “divided.” No figures were obtainable from that summary.
Question 4
What counts as a return?
The anomaly
Ninety per cent of studies find ESG isn't bad for you, and the average effect is r ≈ .13. Both are true. The gap between them is the whole story.
| Study | Sample | Headline | What it actually licenses |
|---|---|---|---|
| Friede, Busch & Bassen 2015 | ~2,200 primary studies | “Roughly 90% of studies find a nonnegative ESG–CFP relation”; 48.2% positive, 10.7% negative | Non-negative ≠ positive. Portfolio and investor studies: 15.5% positive versus 56.7% at firm level mediumsummary extraction |
| Margolis, Elfenbein & Walsh 2009 | 251 studies | “the overall effect is positive but small (mean r = .13, median r = .09, weighted r = .11)” | Real, small, weakening in newer work |
| Khan, Serafeim & Yoon 2016 | 2,307 firms, 1991–2013 | Material-issue alpha 4.18% a year (value-weighted quintile); immaterial approximately zero or negative | The reconciler: it pays where it is financially material mediumworking-paper figures |
| Orlitzky, Schmidt & Rynes 2003 | 52 studies | r = .18, ρ = .36 | The optimistic pole; driven by reputation-index by accounting-measure pairings |
All §1 Area 1, except as flagged.
Demand behaves the same way
Sustainability-marketed consumer packaged goods reached 25.4% dollar share, with an observed shelf-price gap of 26.6%. But stated willingness to pay is “an average 9% premium,” and “65% said they want to buy purpose-driven brands that advocate sustainability, yet only about 26% actually do so” NYU Stern SMSI 2025; White, Hardisty & Habib, HBR 2019. Never quote 26.6% as willingness to pay; it is a brand-, channel- and category-confounded price gap.
And part of what looks like a return is a transfer
“Workers earn substantially lower wages in more sustainable firms” approximately 9–15% at sector level, 2–4% in the firm-level ESG analysis Krueger, Metzger & Wu.
⩢BOUNDARY
Employees are inside the firm's accounts, so their subsidy shows up as the firm's virtue. interpretation
Cost of capital does something similar and smaller: approximately 10bp per standard deviation of CSR El Ghoul et al. 2011. MSCI reports a 39bp WACC spread, and MSCI sells ESG ratings.
When benefit isn't priced, the label gets sold instead
EU sweep: 42% of examined claims possibly “false or deceptive,” 59% without “easily accessible evidence.” Corporate pledges average “only 30%” reduction by 2030 against the “43% reduction” required for 1.5°C; no company in the 2024 assessment reached “high” integrity. Ratings that are supposed to police this correlate at 0.54 across vendors (range 0.38–0.71), worst on governance at 0.30, against 0.94–0.96 for credit ratings mediumsecondary citation chain. Self-labelled ESG funds “hold portfolio firms with worse track records for compliance with labor and environmental laws” than non-ESG funds run by the same managers Raghunandan & Rajgopal 2022. All §1.
The doctrinal fight underneath
Friedman, 13 September 1970: “There is one and only one social responsibility of business, to use its resources and engage in activities designed to increase its profits so long as it stays within the rules of the game, which is to say, engages in open and free competition without deception or fraud.” The dossier flags that the qualifying clauses, that one, and “both those embodied in law and those embodied in ethical custom” are routinely dropped in quotation. Freeman (1984, p. 46) simply widens the boundary by definition: a stakeholder is “any group or individual who can affect or is affected by the achievement of the organization's objectives.” Porter & Kramer (2011) try to dissolve the conflict: “The purpose of the corporation must be redefined as creating shared value, not just profit per se” and Crane, Palazzo, Spence & Matten (2014) reply that shared value “is unoriginal,” “ignores the tensions inherent to responsible business activity,” “is naïve about business compliance,” and rests on “a shallow conception of the corporation's role in society.” All §7.1.
Declarations are cheap; the test is revealed behaviour
| Raghunandan & Rajgopal, JLE 2024 | JUST Capital 2022 | |
|---|---|---|
| Data | Regulator violation records, within-industry size-matched | Own rankings, heavily disclosure-based |
| Finding | Signatories “violate environmental and labor laws more frequently”; “no evidence that signatory firms changed their behavior after signing” | Signatories “Use 2.4 times more renewable energy,” pay more living wages, disclose more |
| Weight | Peer-reviewed, outcome data (magnitudes paywalled) | Advocacy nonprofit; several metrics measure disclosure, not outcomes |
Process evidence sides with the sceptics: of 48 responding signatories, “Only one responding company indicated that the decision was approved by the board of directors”; the authors conclude “the BRT Statement was mostly for show” Bebchuk & Tallarita. The same pattern appears in lobbying: 58% of 293 Forbes 2000 companies hold climate commitments “at odds with their policy influencing,” and by 2025 only 1% published “detailed and accurate” accounts of that engagement InfluenceMap.
Still contested, and why
Raghunandan & Rajgopal versus JUST Capital is not resolvable from the retrieved material. Weight by data provenance, regulator records over self-anchored rankings inferredthe author's ranking, stated as such, but present both. Competent people disagree because one side measures outcomes on a matched sample and the other measures disclosure on a participating sample, and neither is measuring the other's construct.
Question 5
What counts as the environment?
The anomaly
Cut the cost of capital by a point and, on the textbook chain, every marginal project flips positive. Ask the CFOs and 68% say nothing changes.
NPV = Σ CFₜ/(1+r)ᵗ − I₀; accept if NPV > 0
textbook theory.
The model, stated as a behavioural claim
- Claim
- Firms discount at their cost of capital and update as it moves.
- Assumption
- The discount rate is approximately the cost of capital, continuously updated.
- Observably wrong when
- Hurdle rates mean 14.1% and median 13.4%, far above any plausible contemporaneous WACC; 52% had not changed their hurdle once in three years.
Worked example, my arithmetic
Perpetuity PV = CF/r, the author's arithmetic on retrieved parameters. A project paying USD 10.0m a year forever, at the observed 14.1% mean hurdle rate, is worth USD 70.9m.
- Textbook 1:1 pass-through of a 1pp fall in the cost of capital gives r = 13.1% → USD 76.3m (+7.6%).
- Observed pass-through of 0.25pp per 1pp over one to two years gives r = 13.85% → USD 72.2m (+1.8%).
Intuition: monetary policy moves corporate investment about a quarter as much as the formula promises, and slowly.
“68% did not expect any decline in interest rates would induce more investment” Sharpe & Suarez. Firms' discount rates run “on average twice their cost of financial capital,” driven by operational rather than financial constraints Jagannathan, Matsa, Meier & Tarhan, JFE 2016. “It takes many years for changes in the cost of capital to be incorporated into discount rates”; the wedge widened from approximately 3pp (2002) to 5pp (2020), and the “missing investment” puzzle, a cumulative shortfall of approximately 20% of the US capital stock on standard Tobin's-q models, largely disappears once the wedge is modelled mediumGormsen & Huber, AER 2025, via NBER Reporter. A second, independent brake: “higher uncertainty causes firms to temporarily pause their investment and hiring” Bloom, Econometrica 2009, and a 90-point EPU shock cuts gross investment approximately 6% Baker, Bloom & Davis. EPU printed approximately 198–213 across June to August 2026 against a 1985–2009 mean normalised to 100. All §5.2, §5.9.
| United States | Canada | |
|---|---|---|
| Policy rate | 3.50–3.75%, held 29 Jul; 9–3, three dissents for a hike | 2.25%, held 15 Jul; next 2 Sep |
| CPI year over year (Jun) | 3.5% headline / 2.6% core | 2.8% headline / 2.2% ex-gasoline |
| Real GDP | +1.5% annualised (Q2 advance) | Q1 flat; Q2 vintages conflict, do not blend |
| Unemployment (Jun) | 4.2% | 6.5% |
| Debt | 100.6% of GDP (2026); net interest USD 1,039bn = 3.3% of GDP | Federal 41.3% of GDP; debt service 10.6% of revenues |
§5.1, §5.10; the FOMC decision and Q2 GDP were re-verified directly this session. The dissents ran hawkish. Any strategy narrative assuming rate relief contradicts the retrieved record. The two debt metrics are on different bases and are not comparable.
Rules moved twice in six months
- 20 February 2026: SCOTUS, Learning Resources v. Trump, 6–3: “IEEPA contains no reference to tariffs or duties.” Over USD 166bn already collected from more than 330,000 importers; refunds via CAPE from 20 April 2026.
- 24 February 2026: Section 122 stopgap, 10% capped at 15%, 150-day limit; struck down at CIT in May; expired 24 July 2026.
- 24 July 2026: Section 301, 10% to 12.5% on 60 jurisdictions.
- 19 August 2026 (effective): Section 338, first-ever presidential use: 50% on approximately USD 20bn of Canadian dairy, motor vehicles and alcoholic beverages. USMCA-exemption status unverified.
⩢BOUNDARY, IN ITS CLEANEST FORM
The statutory payer is not the economic bearer: “nearly 90 percent of the tariffs' economic burden fell on U.S. firms and consumers” NY Fed.
And the “effective tariff rate” is three different numbers depending on the boundary you draw: 11.8% statutory pre-substitution (Yale), 7.2% realised duties over imports (Penn Wharton), 6.6% calendar-year average (Tax Foundation). They are not contradictory. State the methodology or the figure looks fabricated.
Corporate response: 55% of surveyed executives plan price rises of up to 15% within six months; reshoring in planning or execution reached 26%, up from 10% KPMG, Mar 2026. Reshoring itself splits on the same boundary: Kearney's realised import-ratio index worsened 311bp while the Reshoring Initiative counted 244,940 announced jobs. Outcomes versus announcements. All §5.6, §4.6.
The cycle contains a strategy anomaly
R&D is procyclical, and “equilibrium R&D is likely to be too procyclical” Barlevy, AER 2007, firms cut exactly when the option is cheapest. Across 4,700 public companies and three recessions, “approximately 9% of our sample, flourished after a slowdown,” and the archetype with the highest breakaway probability was the “progressive” mix of selective efficiency cuts plus continued investment at 37%, against 21% for pure cost-cutters Gulati, Nohria & Wohlgezogen, HBR 2010. interpretation Two separate statistics, not one: the study does not report the archetype composition of the 9%.
Policy is now a first-order strategy variable
OBBBA clawed back approximately USD 484.49bn of IRA credits; the USD 7,500 EV credit ended 30 September 2025; US new EV sales fell 28% year over year in Q1 2026 to 212,600 units with share at 5.8%; clean-manufacturing investment fell 34% year over year with the third-highest quarterly cancellation rate since 2018 §5.7, §6.8.
FX and the frictions that never appear in a return series
USD/CAD 1.4026 (5 August 2026, Bank of Canada, the research stream's 1.4018 was wrong; see the audit); twelve-month range 1.3515–1.4234, approximately 5.3% peak to trough. Canada–US treaty dividend withholding: 15% portfolio, 5% for corporate holders of 10% or more of voting stock, cite IRS Pub 597 or PwC (2026), not the raw Convention text, which still reads 10% because the reduction came via the 1995 Third Protocol. US dividends in an RRSP: 0%, under Article XXI(2). In a TFSA: 15%. §5.5
Still contested, and why
AI's productivity contribution: Acemoglu's “no more than a 0.66% increase in total factor productivity over 10 years” versus Goldman's “lift productivity growth by 1.5 percentage points over a 10-year period.” These are different objects, a cumulative level effect and an annual growth rate, and averaging them is an error a grader catches. The stated source of the gap is the automation share assumed (4.6% versus 25%) and whether labour reallocation and new task creation are modelled at all.
Question 6
What does “mature” actually predict?
The anomaly
The incumbent-decline story is the most quoted idea in strategy and the least supported in this corpus: “only 9% of the companies studied fully meet the classic pattern of disruption” (7 of 77 cases) and in approximately 40% the incumbent was never displaced (King & Baatartogtokh, MIT SMR 2015).
What maturity does predict
All §3.
Dickinson's classification makes this operational: maturity is the cash-flow signature operating +, investing −, financing −. On the Russell 3000 (1990–2022, excluding financials and real estate) that stage holds 42.6% of the sample, the highest adjusted ROIC at 11.5%, and a median age since IPO of 13.8 years.
Whether payout starves investment is another boundary fight
Lazonick: 449 S&P 500 firms, 2003–2012, spent “54% of their earnings, a total of $2.4 trillion, to buy back their own stock,” with dividends absorbing “an extra 37%” 91% combined. Fried & Wang, 2007–2016: “net shareholder payouts totaled a much more modest 50%,” and “only 41.5% of R&D-adjusted net income,” because they net out equity issuance and adjust income for expensed R&D. Different constructs, different windows, both arithmetically correct. For the question does the firm still have money to invest, the net measure is the right one inferred. S&P 500 buybacks reached USD 1.020tn in the twelve months to September 2025; the FY2025 actual is a projection only and should not be stated as fact.
The counterexample is live and enormous
The four hyperscalers spent approximately USD 448bn of capex in 2025 (approximately USD 400bn on an ex-finance-lease basis, the gap is definitional), guiding to USD 220bn (Amazon), USD 195–205bn (Alphabet) and USD 130–145bn (Meta) for 2026. Alphabet posted its first negative free-cash-flow quarter since its 2004 IPO; Meta's Q2 2026 free cash flow fell from USD 8.5bn to USD 784m.
interpretation Mature firms will fund exploration at extraordinary scale when they believe the returns are private and appropriable. The underinvestment problem is specific to spillover-rich domains where they are not.
⩢BOUNDARY
“Gross social returns to R&D are at least twice as high as the private returns” Bloom, Schankerman & Van Reenen, Econometrica 2013, and USD 1 of R&D produces approximately USD 13.30 in social benefit, with a conservative floor of USD 4–5 Jones & Summers 2020.
The returns outside the firm are two to thirteen times the ones inside it, and only the inside ones fund the decision.
Where returns are unappropriable, private productivity decays: FDA approvals per USD 1bn of R&D “halved approximately every 9 years since 1950,” roughly 80-fold in inflation-adjusted terms Scannell et al. 2012. Where the public pays the fixed cost, output appears: “NIH funding contributed to published research associated with every one of the 210 new drugs approved by the FDA between 2010 and 2016,” with “over 90% of that research” basic science on the biological target mediumsecondary-sourced, see audit; the US government “invested at least $31.9bn” in mRNA COVID vaccine development, production and purchase, of which 92% was advance purchase commitments BMJ 2023.
Still contested, and why
Whether “mature” even means what this corpus uses it to mean. Every ethics finding here proxies maturity with size or incumbency. No retrieved study connects a Dickinson stage to misconduct, and the dossier records that R&D intensity by lifecycle stage for US firms is “a genuine gap in the literature as accessible via search.”
Question 7
The two records
Harm has institutions
700,000+civil and criminal cases from more than 450 agencies, with total penalties of over $1 trillion since 2000 (Violation Tracker).
- 127 parent companies have each paid over USD 1bn.
- Bank conduct costs in excess of £377 billion, 2008–2018.
- Pharma: at least 482 cases and USD 62.3bn in penalties, 1991–2021.
Benefit has stories
3canonical cases, and the compiler's own conclusion after targeted searching was that voluntary-forbearance cases are “few, old, and largely company- or business-school-narrated rather than independently audited.” inferredthe research compiler's assessment, not a retrieved source's
- Tylenol 1982. Harm caused by an external tamperer. Crisis response, not forbearance.
- CVS 2014. Tobacco exit, approximately USD 2bn a year forgone. CVS also carries an opioid settlement of up to USD 4.90bn.
- Volvo 1959. Seatbelt patent release, narrated on Volvo's own heritage page.
Even the honest attempts run into the same boundary. Access to Medicine Index 2024: “Only 43% of clinical trials take place in the 113 LMICs covered by the Index analysis”; “just 3.5% of trials occur” in the 48 low-income countries; two new non-exclusive voluntary licences in 2024, down from six in 2022. Karnani's bottom-of-pyramid critique put the addressable market at USD 1.2tn PPP, under USD 0.3tn at financial exchange rates, against the USD 13tn claimed. And the withdrawal that did happen came late: Merck pulled Vioxx voluntarily in September 2004, approximately five years after the November 1999 VIGOR signal (79 of 4,000 versus 41 controls), with Lancet-published estimates of 88,000 heart attacks and 38,000 deaths, settling for USD 4.85bn in 2007.
Still contested, and why
Whether the asymmetry reflects the world or reflects what institutions choose to count. Both readings survive the evidence inferred. Genuinely undecidable here.
Question 8
Where the pressure is going
Every entry dated, all §6.
- Jan 2025
Six US banks and Canada's six largest exit the NZBA
The alliance loses its largest members inside a single month.
- 13 Jan 2025
NZAM suspends activities
Following BlackRock's exit.
- 27 Mar 2025
The SEC stops defending its climate rule
The rule stays on the books; the defence does not.
- 23 Apr 2025
Canada's CSA pauses climate and diversity disclosure projects
Indefinitely.
- 3 Oct 2025
The NZBA votes to cease operations
Dissolved.
- 26 Feb 2026
NZAM relaunches with the 2050 net-zero target removed
The alliance returns; the target does not.
- 19 Mar 2026
The EU Omnibus enters into force
Directive (EU) 2026/470 cuts CSRD scope to more than 1,000 employees and more than €450m turnover, raises CSDDD to more than 5,000 employees and €1.5bn, deletes the climate transition plan requirement and eliminates the harmonised civil liability regime.
- 29 May 2026
The SEC proposes rescinding its climate rule
Having stopped defending it fourteen months earlier.
- 2026 proxy season
Roughly 7% of shareholder proposals win majorities, down from 14%
No environmental proposal passed in either 2025 or 2026.
⩢BOUNDARY, CLOSING
Section 2 established that reputation prices counterparty harm and not stranger-harm. Section 5 established that rates barely move firms while rules do.
Removing rules therefore does not shift discipline from the state to the market. It removes discipline from the class of harm the market never priced. inferredthe synthesis is the author's; both premises verified
Two cautions against over-reading the direction
- Enforcement folklore is weaker than assumed. SEC lawyers who later join defence-side firms prosecute more aggressively (approximately +28% damages, +6.8% criminal referral, +7.4% likelihood of charging the CEO) evidence the authors read as “consistent with the ‘human capital’ hypothesis,” not capture deHaan, Kedia, Koh & Rajgopal.
- The countervailing power is concentrated rather than absent. The Big Three's combined S&P 500 stake rose from 5.2% (1998) to 20.5% (2017), casting approximately 25% of votes Bebchuk & Hirst.
Still contested, and why
Whether business is trusted. Edelman's 2026 barometer has business “viewed as being more ethical and competent than all other institutions,” with trust in business at 64% against government at 53%, but business's ethics score is 20 and NGOs' is 17. The dossier's own caution is that this is a relative ranking on a low-scoring field, and the individual scores for institutions other than business and NGOs were not retrieved.
The one thread
Restated
Figure 2
The same decision, seven times
The document's whole argument, on one axis. Each row is a separate literature; the vertical rule is the accounting boundary; what sits left of it is what the number counts.
Which cost line counts as variable decides whether market power exists (DLEU 1.21→1.61 versus Traina flat).
Which parties' costs enter the objective decides whether coal generation is profitable (external damages 0.8–5.6× value added).
Which discount rate, and whose geography, decides whether carbon costs USD 190 a tonne or is not monetised at all.
Statutory payer versus economic bearer decides who “pays” a tariff (approximately 90% falls on US firms and consumers).
Gross versus net payout decides whether buybacks starved investment (91% versus 50%).
Standing decides whether a balancing duty is enforceable (DGCL §365(b): no duty to any person on account of their interest in the public benefit).
Counterparty versus stranger decides whether reputation punishes anything (7.5× versus +0.24%, not significant).
Retention layer
Handwrite-ready core
Whose losses count · core
Dated 6 August 2026. Refetch anything live before use.
- Boundary rule: every result here equals a choice about whose losses enter the number. Cost line / standing / geography / statutory versus economic incidence.
- Lerner:
(P−MC)/P = 1/|ε|;µ = P/MC. DLEU 1.21→1.61 (1980–2016) versus Traina flat-with-SG&A versus Benkard (27% sample dropped). MC is unobservable, that's the whole fight. - Reputation is partial: misrepresentation → reputational loss more than 7.5× legal (KLM '08); environmental → approximately legal only (KLW '05); third-party → +0.24%, not significant (AMP).
- Fraud base rate ~10% a year of large US firms (CI 7–14%); detection 0.33, so you see about one in three; cost 1.6% of equity, approximately USD 830bn (DMZ '24).
- GHR '05 (n = 401): 78% sacrifice value to smooth earnings; 80% cut R&D, advertising and maintenance; 41% reject a positive-NPV project to hit consensus.
- ESG and financial performance: ~90% non-negative (Friede) but r ≈ .13 (Margolis); alpha ~4.18% a year only on material issues (KSY '16). Stated willingness to pay 9%; shelf gap 26.6% is not willingness to pay; 65% intend, 26% buy.
- Rates do not imply capex: hurdle 14.1% mean, sticky; 68% of CFOs unmoved; pass-through
0.25pp per 1pp.
PV = 10/0.141 = 70.9→ 76.3 (textbook 1pp) versus 72.2 (observed). - Social R&D return at least 2× private (BSV '13); USD 1 → ~USD 13.30 (J&S '20). Eroom: approvals per USD 1bn halve every ~9 years since 1950.
- Maturity (Dickinson) = OCF+ / ICF− / FCF−; 42.6% of the Russell 3000, ROIC 11.5%, 13.8 years post-IPO. Disruption fits 7 of 77 cases (9%). Buybacks: 91% gross (Lazonick) versus 50% net (Fried & Wang).
- Dashboard, 6 Aug 2026: Fed 3.50–3.75 (9–3, hawkish dissents) · US CPI 3.5 / core 2.6 · GDP +1.5% · BoC 2.25% · CA CPI 2.8% · USD/CAD 1.4026 · effective tariff 6.6–11.8% by method · Section 338 50% Canadian tariffs effective 19 Aug.
Audit
What was corrected, what stays flagged, what was deleted
| # | Item | Problem | Action |
|---|---|---|---|
| 1 | Hurdle-rate worked example | Used r = 0.13875; correct is 14.1% − 0.25pp = 0.1385. Also +7.7% and +1.7% were arithmetic slips | Corrected to USD 72.2m (+1.8%) and USD 76.3m (+7.6%) |
| 2 | Efing & Hau, “~41% better ratings” | Dropped the qualifier that makes it interpretable: it is 41% relative to a 10-basis-point sample average, so approximately 4bp absolute | Removed from the rewrite rather than quoted without its base |
| 3 | “the sitting CFPB director” | Chopra's remarks are dated March 2022; nothing retrieved establishes who holds the office in August 2026 | Corrected to “Then-CFPB Director Rohit Chopra, March 2022” |
| 4 | “score of 20 on Edelman's net scale” | “Net scale” was the author's gloss; the retrieved material gives scores (business 20, NGOs 17) without describing the scale construction | Deleted the gloss |
| 5 | “business wins a field where every institution scores low” | Only business (20) and NGOs (17) ethics scores were retrieved; other institutions' were not | Reattributed to the dossier's own caution and scoped to what was retrieved |
| 6 | “the ~9% that flourished were disproportionately progressive” | Merged two statistics. The source gives 9% flourished and separately archetype breakaway probabilities (37% versus 21%). It does not report the composition of the 9% | Split into two claims and tagged as interpretation |
| 7 | “19 publicly traded PBCs” | Omitted “excluding SPACs” | Qualifier restored |
| 8 | “statistically zero” for third-party penalties | Slightly stronger than “not significant” | Reworded to the source's own language |
| 9 | Krueger wage gap “9–15%” | Omitted that this is the sector-level estimate; the firm-level ESG analysis gives 2–4% | Both figures now shown |
| 10 | Self-check item citing Einav et al. as “used implicitly” | The prior draft claimed to rely on a study it never cited | Resolved, the Einav material was either used properly or dropped |
Flagged, retained with the flag
- Cleary et al. (PNAS 2018). The primary page returned 403; both figures rest on two agreeing secondary sources, one of which is the authors' own university. Retained, flagged inline.
- Khan, Serafeim & Yoon, 4.18%. Read from the SSRN working paper; the published Accounting Review version may differ.
- Friede's 15.5% / 56.7% portfolio split. Summary extraction from the PDF, not verbatim; the dossier itself flags it for table-level re-checking.
- Gormsen & Huber quantifications. Via the NBER Reporter, a secondary summary of the AER paper.
- ESG ratings versus credit ratings, 0.54 versus 0.94–0.96. The credit figure runs through a secondary citation chain (Prall 2021 via Jacobs & Levy). The frequently repeated “0.99” was not found and is not used.
- AI capex USD 448bn versus approximately USD 400bn. The “definitional, finance leases” reconciliation is the dossier's inference, not a statement by either source.
- USD/CAD. The research stream reported 1.4018 for 6 August; a direct Bank of Canada pull returned 1.4026 for 5 August as the latest published value. The rewrite uses 1.4026. Refetch before any live use.
- “Few, old, and largely company- or business-school-narrated”. This is the research compiler's assessment of the benefit-side record, not a claim by any retrieved source. Labelled as such.
- Contested at source, carried as contested. FY2025 S&P 500 buyback total (projection only, not actual); Bloom, Schankerman & Van Reenen's numeric social-versus-private pair (only “at least twice as high” verified); the “80% of companies removed from CSRD scope” figure (confirmed for the February 2025 proposal, not the adopted directive).
Deleted outright
- The 30% US statutory non-resident withholding rate. It was tagged as recalled from memory in the prior draft. It is not in the retrieved material and it is gone. If you need the base rate the 15% treaty rate reduces from, source it from IRS Pub 515 or the IRC directly.
- Nothing else was repaired with general knowledge. Where a gap existed, the sentence was written without the specific.
Three questions this material raises and does not answer
1. Does maturity itself predict conduct, or is “mature” just a proxy for “big enough to sue”?
Every ethics finding here uses size or incumbency as the stand-in. Dickinson's cash-flow classification is operational and cheap to compute.
Where I’d look: Merge a Dickinson-stage panel onto Violation Tracker parent-level penalty records and test whether penalty frequency and severity vary by stage after controlling for revenue. The dossier confirms nobody in the retrieved literature has published R&D intensity by stage for US firms, which suggests the merge hasn't been done for conduct either.
2. If reputation doesn't price stranger-harm and rules are being withdrawn, does anything take up the load?
The retrieved material has three candidates and dispositive evidence on none: litigation (the first social-media product-defect jury verdict came in March 2026 and is on appeal), concentrated ownership (Big Three at 20.5% as of 2017, with no verified update), and insurance and capital pricing (approximately 10bp per standard deviation of CSR, small).
Where I’d look: Bebchuk & Hirst's methodology run forward on current 13F filings for the ownership channel; the MDL 3047 docket for whether product-liability doctrine generalises beyond the bellwether.
3. Are the “profit share” and “markup” literatures measuring anything that could in principle agree?
Both sides are internally correct because they measure different objects, a profit share of the deflator versus a price-cost markup, and the KC Fed result (41% versus a 59% historical post-recession average) suggests the pattern may have been cyclically ordinary.
Where I’d look: Basu's JEP 2019 discussion is explicitly a methods paper on exactly this identification problem, and is the cheapest single read to settle whether the two constructs can be reconciled or are permanently non-nested.
Provenance
Compiled from the 6 August 2026 dossier. No new sources were retrieved for this pass, and none were retrieved for this conversion either. Every claim, figure, quotation and evidence grade is transcribed from the audited final; the structure around them is new.
How this page is built
Eight sections, one repeating six-slot template, two figures. The figure count is deliberate: almost everything in this document is an argument about what a number includes, and an argument about inclusion reads better as structured text than as a chart. The two that survived are the two places where a comparison is the finding.
Colour carries one meaning only: inside the accounting boundary, or outside it. The pair was checked for colour-vision separation and surface contrast in both light and dark before a mark was drawn. Charts are static SVG generated at build time, so every figure renders with scripting switched off; script adds only the tooltip layer, the theme control, the contents rail's scroll tracking and the mobile drawer.
Reading face: Inter, for the tables and the tabular figures that have to hold their alignment. The eight question headings, the takeaways and the hook are set in a serif, because those are the places the document speaks rather than reports.