The argument that never settles

Ask any room of financially literate Americans which law better protects the country from another banking crisis, the Glass-Steagall separation of 1933 or the Dodd-Frank capital and activity regime of 2010, and the room will divide before the question is finished. The division is not really about two statutes. It is about two incompatible theories of what makes finance dangerous. One theory says danger lives in the shape of firms, in the combination of deposit taking with securities dealing under one roof, and the law must therefore keep those shapes apart. The other says danger lives in thin cushions against loss, and the law must therefore make those cushions thick, test them under stress, and plan for orderly failure. Everything else in this comparison, the repeal, the crisis, the Volcker Rule, the restoration bills, is commentary on that single fork.

This article resolves the comparison structurally, which no widely read page has yet done. The test it sets is simple. When the reader finishes, the reader can state the fundamental difference between the two regulatory philosophies, structural separation of activities into different firms versus permission of combination under capital requirements and activity limits. The reader can explain which parts of the 1933 separation still stand and which fell. The reader can evaluate, firm by firm, whether restoring the separation would have prevented the 2008 crisis. And the reader can reach a defended verdict on which model better addresses the risk, with the deciding factor named explicitly. That is the whole assignment, and every section below serves it.

The statutory identity of the comparison needs a precise statement at the outset, because imprecision here is the source of most public confusion. On one side stand sections 16, 20, 21, and 32 of the Banking Act of 1933, the provisions conventionally called Glass-Steagall after their congressional sponsors. On the other side stands Public Law 111-203, the Dodd-Frank Wall Street Reform and Consumer Protection Act, signed July 21, 2010. Between them stands the intervening repeal, effected by section 101 of Public Law 106-102, the Gramm-Leach-Bliley Act, signed November 12, 1999, which removed two of the four 1933 provisions and left the other two in force. The popular telling treats the 1999 law as a full repeal and the 2010 law as its answer. The record shows something more interesting: a partial repeal, a crisis concentrated in firms the repealed provisions never governed, and a response statute that chose a different theory of regulation rather than restoring the old one.

Why the comparison matters beyond the seminar room is straightforward. The question of whether to break up large financial firms, to rebuild walls between activities, or to keep the combined firms and make them hold more capital is a live policy choice with consequences for credit availability, for the cost of financial services, and for the exposure of taxpayers to future rescues. Both a bipartisan proposal to restore separation and sustained opposition to it exist in Congress, and the subject returns on social media with every market wobble. A reader who cannot distinguish the affiliation ban from the underwriting ban, or the Volcker Rule from structural separation, cannot evaluate any of those proposals. This article builds the distinction from the statute text up, then follows it through the crisis record, the causal debate, and the verdict.

A note on method before the substance begins. Every causal claim in this article is attributed to a named analyst, because the repeal and the crisis invite confident assertions that the evidence does not always support. The article states plainly what the evidence establishes and what it does not. Restoration proposals are presented by their sponsors and their text, without endorsement, and the article takes no position on pending legislation. The verdict at the end is the article’s own, defended on the evidence and open about the judgment it requires.

The public debate usually arrives in one of three framings, and each framing smuggles in an assumption the evidence does not support. Naming the framings helps the reader see why the comparison has never been resolved, because most arguments are not really about the two models at all. They are about stories in which the models play assigned roles.

The first framing is the morality tale: the 1999 repeal caused the 2008 crisis, and the 2010 act was the deserved response. Its assumption is that chronology establishes causation, that because the repeal preceded the crisis, the repeal produced it. The firm level evidence dismantles this framing without needing to defend the repeal’s wisdom. The repealed provisions governed bank securities affiliations, and the firms that failed were not bank securities affiliates. A morality tale needs a villain whose actions connect to the harm, and the repeal’s mechanism does not connect to the failures. The framing persists because it offers a simple story with a clear legislative remedy, restore the repealed provisions, and simple stories outcompete accurate ones in public argument.

The second framing is the technocratic reassurance: the 2010 act fixed the problem, and the debate is therefore moot. Its assumption is that the control system works as designed, that capital is sufficient, stress tests are honest, living wills are credible, and supervisors are vigilant. This framing’s weakness is the opposite of the first framing’s: it mistakes the design for the outcome. The 2010 model is a set of instruments, and instruments require operators. The capital data show the instruments being built, which is genuine progress, but the supervisory will to use them under political pressure is unproven on the record this article examines, and the measurement problems discussed later in this article, the gaming of risk weights and the possibility of correlated model error, mean the reassurance is premature.

The third framing is the nostalgia argument: the old wall worked, the new machinery is needlessly complex, and restoration would return the system to safety. Its assumption is that the 1933 regime actually governed the financial system that existed before 1999, rather than a shrinking portion of it. The erosion history refutes this. By the 1990s the affiliation ban constrained a declining share of financial activity, the crisis emerged from outside its perimeter, and restoring the wall would rebuild a boundary around a system that no longer fits inside it. Nostalgia for the wall is really nostalgia for a simpler financial system, and no statute can legislate simplicity back into existence.

This article rejects all three framings and offers the structural comparison in their place. The question is not who sinned, whether the technicians have triumphed, or whether the past was better. The question is which theory of risk containment the evidence supports, and the answer requires walking through the provisions, the firms, and the data without the comforting simplifications.

Glass-Steagall’s four provisions and what the 1999 repeal actually removed

The name Glass-Steagall is a synecdoche that misleads. It suggests a single wall, built once and torn down once. The reality is four distinct provisions, aimed at four distinct combinations, with two demolished in 1999 and two still standing. Any serious comparison with the 2010 statute must begin by inventorying all four, because the model that Dodd-Frank replaced was never the whole of the 1933 design. It was half of it, and the surviving half still shapes what banks and securities firms may do.

Why did Congress separate commercial banking from securities dealing in 1933?

Congress acted after the 1929 crash and the Senate hearings of the early 1930s documented how bank affiliates had underwritten and sold securities to the banks’ own depositors, creating conflicts of interest that legislators judged unmanageable and losses that depositors could not see coming.

The affiliate system that provoked the 1933 act deserves a concrete description, because its mechanics are the original case for the separation theory and the template for every later argument about conflicts of interest. In the 1920s, large commercial banks created securities affiliates, legally separate subsidiaries that underwrote and distributed stocks and bonds. The bank’s own trust department and its commercial customers then became natural buyers for the affiliate’s issues. The conflicts compounded at every step. The affiliate had an incentive to underwrite securities of questionable quality because the fees were earned at issuance, regardless of later performance. The bank had an incentive to lend to the issuers whose securities its affiliate was selling, because a loan that kept the issuer afloat protected the value of the paper the affiliate had placed. And the bank’s depositors and trust clients, who relied on the bank’s judgment, could not see that the judgment was compromised by the affiliate’s interest in moving inventory. When the issues soured, the losses fell on the customers while the fees had long since been booked.

The 1933 Congress confronted this machinery after the Senate’s investigation laid its operations bare, and the legislators’ conclusion was not that bankers were unusually corrupt but that the structure made corruption rational. Disclosure could not fix the problem, on this view, because the depositors could not evaluate what they were not told, and conduct rules could not fix it because the temptation was renewed with every underwriting. Only separation, removing the affiliate from the bank’s organization entirely, could remove the incentive. This is the moral and analytical core of the separation theory, and its power explains why the theory retains adherents long after the specific abuses were addressed. The claim is not that every combined firm will abuse its customers. The claim is that the structure makes abuse profitable, that profitable abuse will eventually occur, and that the supervisor cannot watch every transaction in real time. A reader who finds this logic compelling has understood the 1933 model from the inside, which is necessary whether or not the reader ultimately prefers it.

Section 20 of the Banking Act of 1933, codified at 12 U.S.C. section 377, barred member banks of the Federal Reserve System from affiliating with firms engaged principally in the issue, flotation, underwriting, public sale, or distribution of stocks, bonds, debentures, notes, or other securities. The key phrase was engaged principally. A bank could not own, control, or combine with a securities firm whose main business was underwriting and dealing. This was the affiliation wall, the provision that kept commercial banking organizations and investment banking organizations in separate corporate families. Section 32, codified at 12 U.S.C. section 78, reinforced the wall at the level of personnel: it barred officers, directors, and employees of securities firms from serving simultaneously as officers, directors, or employees of member banks. The two provisions worked as a pair. Section 20 blocked the corporate combination, and section 32 blocked the human interlocks that could have recreated the combination informally.

Section 16, codified at 12 U.S.C. section 24(Seventh), operated inside the bank itself. It prohibited national banks from underwriting or dealing in most securities directly, while permitting dealing in bank-eligible securities such as obligations of the United States government and general obligation bonds of states and municipalities. Section 21, codified at 12 U.S.C. section 378, operated from the other direction: it prohibited any person or firm engaged in the business of issuing, underwriting, selling, or distributing securities from engaging in the business of receiving deposits. Together the four provisions built a symmetrical design. Banks could not do securities business directly, securities firms could not take deposits, banks could not affiliate with securities firms, and the people running each could not run the other.

The repeal that everyone cites removed only the affiliation pair. Section 101 of the Gramm-Leach-Bliley Act repealed sections 20 and 32, the provisions barring affiliation and interlocking management between banks and securities firms. Sections 16 and 21 were never repealed and remain in force. A bank still may not underwrite most securities directly, and a securities firm still may not take insured deposits. The wall that fell in 1999 was the wall barring affiliation. The walls barring direct combination of the activities inside a single firm survived. Congressional Research Service analyses of the repeal state this division explicitly, noting that the depository institution subsidiaries of the new financial holding companies remained subject to the securities restrictions of section 16, and that the nondepository subsidiaries remained barred from offering insured deposits by section 21. The surviving interaffiliate transaction limits of sections 23A and 23B of the Federal Reserve Act continued to constrain dealings between a bank and its affiliates after the repeal.

One nuance qualifies the surviving section 16, and the article records it because precision is the point. The 1999 law did not leave section 16 entirely untouched. It amended the provision to permit well capitalized commercial banks to underwrite municipal revenue bonds, the non general obligation variety, creating a single carved exception to the rule that a bank may not underwrite most securities directly. The exception is narrow, and it does not change the provision’s character, but a claim that banks cannot underwrite any securities at all would be false in the absolute, and this article does not make it.

Which parts of the 1933 separation survived the 1999 repeal?

Sections 16 and 21 survived intact in substance, so banks still cannot underwrite most securities directly and securities firms still cannot take deposits, while sections 20 and 32 were repealed, which removed only the bans on affiliation and interlocking management.

The 1999 repeal arrived through the Gramm-Leach-Bliley Act, and its mechanics deserve a brief account because they explain what Congress thought it was doing. The statute originated in the Senate as S. 900, introduced April 28, 1999, by Senator Phil Gramm of Texas under the title Financial Services Modernization Act of 1999, and was enacted under the short title Gramm-Leach-Bliley Act as Public Law 106-102. The conference report cleared the House by 362 to 57 and the Senate by 90 to 8 on November 4, 1999, and President Clinton signed the measure on November 12, 1999. Its long title announced the purpose plainly: to enhance competition in the financial services industry by providing a prudential framework for the affiliation of banks, securities firms, insurance companies, and other financial service providers. Congress was not repealing a philosophy in a fit of deregulatory enthusiasm. It was replacing the affiliation ban with a supervised affiliation regime, the financial holding company, in which a bank holding company could elect to engage in activities financial in nature, including securities underwriting and insurance underwriting, through holding company affiliates subject to umbrella supervision by the Federal Reserve, with functional regulation of the securities and insurance affiliates left to the Securities and Exchange Commission and the state insurance commissioners. The full account of that machinery belongs to the guide to what the 1999 law actually repealed, which this article cites for the repeal’s scope and leaves the detail there.

The forcing event behind the repeal is worth one paragraph, because it shows how the old law died in practice before Congress buried it. In April 1998, Citicorp and the Travelers Group announced a merger combining one of the largest commercial banks with an insurance and securities conglomerate, a combination the unrepealed affiliation provisions would not have permitted. The Federal Reserve Board approved the transaction in September 1998 on the condition that the combined company conform its activities to the Bank Holding Company Act within two years of consummation, including by divestiture as necessary, and the merger was consummated in October 1998. The conformance clock meant the new Citigroup faced the prospect of shedding its insurance underwriting operations by late 2000 unless Congress changed the law. The 1999 act changed the law. The episode is often told as proof that legislation follows deals, and there is truth in that telling, but the fuller truth is that the affiliation ban had been eroding for more than a decade under regulatory interpretations that had already permitted limited securities affiliations, and Congress in 1999 ratified and regularized a market structure that was already emerging.

The regime that replaced the affiliation ban deserves a clear description, because the 1999 law is often caricatured as simple deregulation when it was actually a different regulatory design. Congress did not remove supervision from combined firms. It replaced the ban on combination with a supervised permission structure called the financial holding company. A bank holding company could elect financial holding company status and thereby engage, through holding company affiliates, in activities defined as financial in nature: securities underwriting and dealing, insurance underwriting, merchant banking, and related activities. The election was not automatic. Every subsidiary bank had to be well capitalized and well managed, and each had to hold at least a satisfactory rating under the Community Reinvestment Act. Insurance underwriting and merchant banking could be conducted only in the holding company’s nonbank affiliates, never inside the bank itself, preserving a version of the old separation at the level of the insured depository institution.

Supervision of the new structure followed a deliberate division of labor. The Federal Reserve Board became the umbrella supervisor of financial holding companies, responsible for the consolidated organization, while functional regulation of the affiliates stayed with their specialist regulators: the Securities and Exchange Commission for securities affiliates, the state insurance commissioners for insurance affiliates. The statute directed the Federal Reserve to rely as much as possible on the functional regulators’ examinations and information rather than duplicating them. Meanwhile the old interaffiliate protections survived intact. Sections 23A and 23B of the Federal Reserve Act, which impose quantitative limits, collateral requirements, and arm’s length terms on transactions between a bank and its affiliates, continued to constrain dealings inside the combined organizations. The new powers took general effect in March 2000. For Citigroup, the timing resolved the conformance clock the Federal Reserve had set in 1998: the two year window to shed impermissible activities, which would otherwise have expired in late 2000, was overtaken by a statute that made the combination lawful under supervision.

The significance for the comparison is that 1999 represents a hinge between the two philosophies rather than a simple victory of one over the other. Congress kept the direct activity bans of sections 16 and 21, kept the interaffiliate transaction limits, and replaced only the affiliation ban with a supervised alternative. The 2010 statute then built its far more elaborate control system on top of that supervised combination model. A reader who understands the financial holding company understands why the 2010 act regulates combined firms instead of breaking them up: by 2010 the combined firm was the established legal form, supervised for a decade, and Congress chose to make it safer rather than to unmake it.

With the inventory complete, the comparison can be stated in its cleanest form. The 1933 model said certain combinations of financial activity must not exist inside one organization, because risk migrates across activities and conflicts of interest cannot be managed. The 2010 model says combinations may exist, but the combined organization must hold enough capital to absorb losses, maintain enough liquidity to survive a run, prove its survival under supervisory stress scenarios, accept limits on the riskiest activities, and file plans for its own orderly failure. One model polices the shape of firms. The other polices the resources and conduct of firms whatever their shape. The next section examines what each theory assumes about the nature of financial risk.

The erosion before the repeal

The standard story presents 1999 as the year the wall fell, as though a solid barrier stood until Congress removed it. The legal history is less dramatic and more instructive. By the time the Gramm-Leach-Bliley Act repealed sections 20 and 32, the wall those sections built had been substantially tunneled, and the tunneling had been done by regulators and courts over two decades, not by Congress in a single vote. A Congressional Research Service report prepared before the Senate acted on the 2010 reform bill described the result candidly: Glass-Steagall had imperfectly separated commercial and investment banking, and the Federal Reserve had authorized extensive securities activities for bank affiliates since the 1980s. Understanding the erosion matters for the counterfactual, because restoring the pre-1999 law would restore a regime the industry had already learned to work around.

The tunneling began with three words in section 20: engaged principally. The statute barred affiliation with a company engaged principally in securities underwriting, which left open the question of what principally meant. Beginning in 1987, the Federal Reserve answered by authorizing section 20 subsidiaries, separately organized securities affiliates of bank holding companies that could derive a limited share of their revenue from bank-ineligible securities, the securities section 16 forbade the bank itself to underwrite. The revenue cap started small and rose over the following decade as the Board grew comfortable with the arrangement, reaching 25 percent in 1996. By the late 1990s, the section 20 subsidiaries of the largest bank holding companies were significant participants in corporate debt underwriting and other securities businesses. The affiliation bar still stood as a formal matter, but its economic content had been negotiated down to a revenue percentage.

The Comptroller of the Currency tunneled from the other side. In interpretations dating to 1978 and 1987, the Comptroller concluded that national banks could engage in securitization activities as part of the business of banking, and the courts confirmed bank securitization authority by 1990. This is the legal history Melanie Fein documented in her treatise, and it is why the claim that the 1999 repeal enabled securitization fails at its first step. The securitization market that produced the mortgage-backed securities at the center of the crisis grew up under administrative and judicial interpretations that predated the repeal by decades, not under the permission the repeal granted.

The consequence for the comparison is a correction to both sides. Defenders of the repeal sometimes describe 1999 as a mere ratification of what regulators had already permitted, which understates what changed: the repeal removed the revenue caps, the corporate separateness requirements, and the interlock bar, and it created the financial holding company as a clean legal form for the combination. Critics of the repeal sometimes describe 1999 as the moment banks were first allowed into securities, which mistakes a formalization for an invention. The truth between them is that the 1999 act regularized, expanded, and simplified combinations that the regulatory system had been permitting in constrained form for more than a decade.

The erosion also sharpens the counterfactual. A proposal to restore Glass-Steagall is usually understood as a proposal to restore the 1933 text, but the 1933 text as interpreted through 1999 was already a porous regime. A faithful restoration would have to decide whether to restore the statute as written or the statute as the regulators had narrowed it, and the choice between those two restorations is itself a substantive policy decision that the restoration bills’ sponsors rarely address. The separation model’s historical record, in other words, is not the record of an unbreached wall. It is the record of a wall that regulators spent twenty years finding doors in, which suggests that the pressure to combine banking and securities activity was economic rather than merely legal, and that a restored wall would face the same pressure. The restoration bills, for all their sponsors’ seriousness, contain no answer to the tunneling problem: they reimpose the affiliation bar without addressing the interpretive dynamics that hollowed it out the first time, which means a future Federal Reserve, armed with the same engaged-principally ambiguity, could begin the erosion anew.

Two theories of how risk is contained

The 1933 and 2010 statutes disagree about the nature of financial risk itself, and the disagreement is worth stating as philosophy before it is stated as mechanics. The 1933 Congress believed that risk is contagious across activities and that conflicts of interest, once structurally embedded, cannot be managed by rules of conduct. A bank that both takes deposits and underwrites securities will, on this view, inevitably steer its depositors into the securities its affiliate needs to sell, will lend to prop up the issuers whose paper it underwrote, and will reach for the higher returns of the securities business with money that was entrusted for safekeeping. The misconduct is not a failure of compliance. It is the predictable product of the structure, and the only reliable remedy is to forbid the structure. Separation is therefore preventive rather than corrective. It does not punish the abuse after the fact. It removes the temptation by removing the combination.

The 2010 Congress believed something nearly opposite. It believed that combinations of activity are economically valuable, that diversified firms can be more stable than narrow ones, and that the danger lies not in the combination but in the thinness of the resources standing behind it. A combined firm with abundant equity capital, deep liquidity reserves, tested survival plans, and a credible path to orderly failure can, on this view, safely house activities that a thinly capitalized firm cannot. Risk is therefore managed by measurement and buffers rather than by prohibition. The statute permits the combination and then surrounds it with a control system: minimum capital ratios that rise with the firm’s systemic footprint, liquidity standards that require survival through a thirty day stress, supervisory stress tests that model severe recessions and market shocks, enhanced prudential standards imposed by the Federal Reserve on the largest firms, activity limits on the riskiest trading, and resolution plans, the living wills, that map how the firm could fail without public rescue. Where the 1933 model trusts the wall, the 2010 model trusts the cushion, the drill, and the exit plan.

Each theory has an implicit model of human behavior inside financial firms. The separation theory assumes that incentives overwhelm controls, that a trader with access to insured deposits will use them, and that supervisors cannot monitor every conflict in real time. It is a theory born of the 1920s, when bank affiliates did in fact sell questionable securities to the banks’ customers, and it treats that history as revealing a permanent feature of combined firms rather than a correctable lapse. The capacity theory assumes that incentives can be channeled by making risk expensive, that a firm forced to hold substantial equity against its positions will take fewer reckless positions because its own shareholders bear the loss, and that supervisors armed with data, stress scenarios, and resolution plans can oversee complexity they could never have overseen in the 1930s. It is a theory born of the 2008 crisis, when thinly capitalized firms of many shapes failed together, and it treats that history as revealing a permanent shortage of loss absorbing capacity rather than a problem of any particular corporate form.

The two theories also differ in what they ask of regulators. Separation asks little of supervisors day to day and much of them at the boundary. The ban is binary: the combination is lawful or it is not, and the supervisor’s job is to police the line. This simplicity is the model’s great administrative virtue and also its rigidity. It cannot distinguish a prudent combination from a reckless one, and it cannot adapt when the financial system invents new activities that fit neither side of the old wall. The capacity model asks everything of supervisors continuously. Capital ratios must be calculated, stress scenarios designed, living wills reviewed, liquidity positions monitored, and each of these tasks requires judgment, data, and institutional competence that must be maintained indefinitely. Its virtue is flexibility: the same framework can tighten or loosen as conditions change. Its vulnerability is supervisory failure, because a control system is only as good as the people operating it, and the history of financial regulation offers no shortage of episodes in which supervisors saw the risk and acted too late.

The efficiency case for combination deserves a fair hearing inside the philosophy section, because the 2010 model does not permit combined firms out of indifference. It permits them because combination has genuine economic virtues that separation forgoes. A diversified financial firm draws revenue from multiple businesses, which can stabilize earnings when any single business suffers. It can serve a corporate client across lending, underwriting, and advisory needs within one relationship, which the client may prefer to assembling the services piecemeal. It can achieve economies of scope in technology, compliance, and distribution. And, as the crisis demonstrated, a strong combined firm can absorb a failing specialist, as Bank of America absorbed Merrill Lynch and Countrywide, stabilizing the system in the moment of panic. These are not trivial benefits, and a separation regime sacrifices all of them. The 1933 model implicitly judged the virtues worth sacrificing to eliminate the conflicts. The 2010 model judges the conflicts manageable enough to preserve the virtues. The reader should recognize this as a real trade rather than a rhetorical one, because the costs of separation fall on credit availability and the price of financial services, which are paid by households and businesses far from Wall Street.

A reader tempted to declare one theory obviously superior should pause at the trade each one makes. Separation buys simplicity and the elimination of certain conflicts at the price of forgoing the efficiencies of combination, including the diversification of revenue and the ability of strong firms to absorb weak ones in a crisis, a point that will return with force in the counterfactual section. The capacity model buys flexibility and the preservation of combination at the price of permanent supervisory burden and the risk that the measurements miss the next crisis’s particular shape. Neither theory eliminates financial risk. They distribute it differently, and they fail differently, which is why the comparison table matters more than any slogan about either law.

Each theory also carries a historical pedigree that explains its emotional force, and the pedigree is worth examining because it shows why the debate resists resolution by evidence alone. The separation theory descends from the Progressive era critique of concentrated financial power, given its classic expression in Louis Brandeis’s 1914 indictment of the money trust and its legislative expression in the Pecora hearings of the early 1930s, where Senate investigators documented how the securities affiliates of large banks had sold questionable issues to the banks’ own depositors. That history created a moral narrative that still structures the debate: the banker as a figure torn between service to depositors and profit from speculation, with the structure of the firm determining which loyalty wins. The narrative’s power does not depend on its precise fit to the 2008 facts, which is why it survives the firm level evidence against the causal claim. It is a story about what concentrated financial power does to character, and stories about character are not refuted by balance sheets.

The separation theory also draws strength from the long erosion that preceded the 1999 repeal. For more than a decade before Congress acted, regulatory interpretations had been permitting progressively broader securities affiliations for bank holding companies, and the economic argument for finishing the job was that American banks were losing ground to foreign universal banks and to domestic nonbank competitors that faced no such restrictions. Supporters of repeal argued that the wall had become a competitive handicap that protected no one, since sophisticated customers could get securities services from firms the wall did not cover. Opponents warned that the wall’s erosion proved its necessity rather than its obsolescence. The repeal settled the argument in favor of the erosion’s defenders, but the underlying disagreement about whether the wall protected depositors or merely protected market shares was never resolved. It simply moved into the debate over the 2010 statute’s adequacy.

The capacity theory has its own pedigree, rooted in the supervisory experience of the 2008 crisis as the 2010 Congress understood it. The lesson the drafters took was not that any particular charter had failed but that leverage and liquidity practices had failed across charters: thinly capitalized firms funded with runnable short term liabilities, holding concentrated exposures to housing risk, supervised by regulators who measured compliance with static rules rather than survival under stress. The intellectual bet of the 2010 act is that risk measurement and supervisory technique have advanced far beyond what the 1933 Congress could have imagined, and that a regime of continuous quantitative oversight can do what a static wall cannot: distinguish the safe combination from the dangerous one, and calibrate the constraint to the firm’s actual risk rather than to its legal form. Whether that bet was well placed is the empirical question the capital data and the post enactment supervisory record must answer.

There is a symmetry in the two pedigrees that the comparison table cannot show but the reader should grasp. Each generation wrote its reform against the last crisis’s visible shape. The 1933 Congress saw the 1920s affiliate abuses and wrote a wall against combinations. The 1999 Congress saw the wall’s erosion and the rise of global universal banking and wrote a supervised permission. The 2010 Congress saw the collapse of thinly capitalized firms of every shape and wrote a measurement regime. Regulatory history moves in this pattern, with each reform solving the previous failure and creating the conditions for a failure of a new type. A reader who sees the pattern will be properly skeptical of any claim that either model ends the cycle, and will read the verdict that follows as a judgment about the best available choice rather than a final answer.

The two-model comparison table

The table below is the article’s findable artifact. It places the two models side by side across the five dimensions the brief requires: the philosophy, what is prohibited, what is permitted with conditions, the enforcement mechanism, and the failure mode each model is designed to prevent. Read across the rows for each model’s internal logic, and down the columns for the contrast.

Model	Philosophy	What is prohibited	What is permitted with conditions	Enforcement mechanism	Failure mode each model is designed to prevent
1933 structural separation (Banking Act of 1933, sections 16, 20, 21, 32)	Risk migrates across activities and conflicts of interest are unmanageable, so the law must keep deposit taking and securities dealing in separate firms	Affiliation between member banks and firms engaged principally in securities underwriting and dealing (section 20, repealed 1999); interlocking officers, directors, and employees between banks and securities firms (section 32, repealed 1999); direct underwriting and dealing in most securities by banks (section 16, still in force); deposit taking by securities firms (section 21, still in force)	Banks dealing in bank-eligible securities such as United States government obligations and general obligation municipal bonds; well capitalized banks underwriting municipal revenue bonds under the 1999 amendment to section 16; all non-securities activities of banks and all non-deposit activities of securities firms	Binary prohibition policed by bank supervisors at the boundary; the combination is lawful or unlawful, so compliance is a question of corporate structure rather than of measured risk	A bank using insured deposits to fund securities speculation, or steering depositors into securities its affiliates underwrote, recreating the conflicts of the 1920s that the 1933 Congress investigated
2010 capital and activity model (Public Law 111-203)	Combination is efficient and potentially stabilizing; risk is contained by requiring thick loss absorbing capacity, proven survival under stress, limits on the riskiest activities, and plans for orderly failure	Proprietary trading and hedge fund or private equity fund sponsorship by banking entities, subject to enumerated exemptions (section 619, the Volcker Rule); other targeted prohibitions assigned to agency rulemaking across the statute’s titles	Combined commercial and investment banking, securities underwriting, and dealing inside financial holding companies, provided the firm meets Basel III capital minima, liquidity standards, supervisory stress tests, Federal Reserve enhanced prudential standards under section 165, and files resolution plans under section 165(d)	Continuous supervisory measurement: capital ratios calculated and reported, annual stress tests run by firms and supervisors, living wills reviewed for credibility, liquidity positions monitored, with restrictions triggered by shortfalls	A large combined firm’s losses spilling into its deposit base and the wider financial system, or its disorderly failure forcing a taxpayer rescue; the model aims to make big firms survivable and, failing that, resolvable

Three contrasts in the table deserve emphasis because they are the points most often blurred in public argument. First, the 1933 model prohibits combinations while the 2010 model prohibits selected activities inside combinations. The Volcker Rule’s ban on proprietary trading sits in the prohibition column of the 2010 row, but it restricts what a combined firm may do, not whether the combined firm may exist. That distinction is the entire difference between an activity restriction and a structural separation, and confusing the two is the most common analytical error in this debate. Second, the enforcement mechanisms differ in kind, not just degree. The 1933 mechanism is episodic and structural: supervisors check the shape of the firm. The 2010 mechanism is continuous and quantitative: supervisors check the resources of the firm, quarter after quarter, scenario after scenario. Third, the failure modes differ. The 1933 model was built to prevent the abuse of depositors and the conflicts of combined firms. The 2010 model was built to prevent the collapse of large firms and the public rescues that follow. A reader who asks which model prevents crises must first ask which failure the reader fears, because the models were aimed at different ones.

A worked example makes the contrast concrete. Consider a hypothetical bank holding company with 400 billion dollars in assets that owns both a commercial bank and a broker dealer subsidiary. Under a restored 1933 separation regime, the firm’s existence in its current shape would be unlawful. The affiliation ban would force a choice: divest the broker dealer, or shed the bank charter and operate as a pure securities firm outside the federal safety net. Supervisors would not need to evaluate the firm’s risk models, its funding structure, or its management quality. They would need only to read the organizational chart. The compliance question is binary, the remedy is structural, and the cost is the foregone value of the combination, whatever that value might be.

Under the 2010 regime, the same firm continues to exist and faces a battery of quantitative demands instead. It must hold common equity tier 1 capital of at least 4.5 percent of risk weighted assets plus the 2.5 percent conservation buffer, tier 1 capital of 6 percent, total capital of 8 percent, and a leverage ratio of 4 percent. It must run annual stress tests under supervisory scenarios and submit its capital plans for review, with distributions constrained by the results. It must file a living will mapping its own orderly failure. Its trading desks must comply with the Volcker Rule’s prohibitions. It must maintain liquidity coverage for thirty days of stressed outflows. Supervisors evaluate all of this continuously, and the firm remains lawful only so long as the measurements satisfy them. The compliance question is graduated, the remedy for shortfall is more capital or less risk, and the cost is the expense of the control system and the foregone returns on the equity the firm must hold.

Now reverse the example to see what each regime misses. Consider a pure investment bank, legally separate from any commercial bank, funding long term mortgage assets with overnight borrowing at thirty to one leverage. The 1933 regime sees nothing wrong: the firm is separate, the wall is intact, the organizational chart is clean. Yet this was approximately the shape of the firms that failed first in 2008. The 2010 regime, by contrast, sees the leverage, the runnable funding, and the concentrated exposure, and demands capital, liquidity, and stress survival regardless of the firm’s legal form, at least for firms within its enhanced supervision perimeter. The example reveals the models’ complementary blind spots. Separation polices the chart and misses the leverage. Capacity polices the leverage and depends on the quality of its measurements. A reader choosing between them is choosing which blind spot to live with.

The comparison becomes sharpest when four concrete risks are run through both models in turn, because each model was built for different dangers and the exercise shows exactly where each one bites and where each one is silent. The four risks are proprietary trading losses, conflicts of interest, liquidity runs, and disorderly failure. Every financial crisis in American history has featured at least two of them, and the 2008 crisis featured all four.

Consider first proprietary trading losses reaching insured deposits. A bank that gambles with its own capital on market movements can lose sums that impair its ability to repay depositors, which is the nightmare the 1933 Congress legislated against. The 1933 model attacks this risk at the root of combination: section 16 bars the bank from dealing in most securities directly, and the repealed section 20 barred the bank from affiliating with a firm that did. The risk is prevented by making the organizational form that would take it unlawful. The 2010 model attacks the same risk while permitting the form: the Volcker Rule bans proprietary trading inside banking entities, and the capital requirements ensure that any losses that occur despite the ban stop at the firm’s equity. The difference is between removing the temptation and policing it. The 1933 approach is cheaper to enforce and forgoes the legitimate market making that can resemble proprietary trading. The 2010 approach preserves the legitimate activity and accepts the enforcement cost of distinguishing it from the banned kind.

Consider second the conflict of interest in its classic form: a bank steering its depositors or trust clients into securities underwritten by its affiliate. The 1933 model eliminates the conflict by eliminating the affiliate, which is the cleanest possible solution and the one the Pecora hearings made vivid. The 2010 model manages the conflict through a combination of the surviving interaffiliate transaction limits, the Volcker Rule’s restrictions, fiduciary duties owed to customers, and the general securities law apparatus against fraud and manipulation. The 2010 answer is undeniably more complex, and its complexity is the separationists’ opening: every managed conflict is a conflict that can be mismanaged, and the supervisor cannot review every customer interaction. The capacity advocates’ reply is that the 1933 answer throws away genuine efficiencies, that combined firms can serve customers more cheaply, and that the conflicts can be priced and policed. The reader’s judgment here depends on the reader’s estimate of supervisory capacity, which is why the debate recurs.

Consider third the liquidity run, the risk that short term lenders refuse to roll their funding and a firm dies of illiquidity in days. Here the asymmetry between the models is starkest, because the 1933 model has almost nothing to say about funding. A legally separate securities firm, fully compliant with every Glass-Steagall provision, can still fund long term assets with overnight borrowing and collapse when confidence evaporates, which is approximately what happened to Bear Stearns and Lehman Brothers. The wall polices the organizational chart, not the maturity structure. The 2010 model, by contrast, was built substantially around this risk: the liquidity coverage ratio demands thirty days of survival under stress, the net stable funding ratio addresses longer term stability, and the stress tests model funding shocks alongside credit losses. On this risk the capacity model is clearly the more responsive design, because it measures the thing that actually killed the firms.

Consider fourth the disorderly failure of a giant institution, the risk that a firm’s collapse cascades through the system because no orderly wind down exists. The 1933 model’s answer is indirect: smaller, simpler, separated firms should fail more cleanly, with fewer interconnections to transmit the shock. The logic is plausible but was never tested against a crisis of 2008’s shape, because the crisis’s central firms were already separate and still transmitted shocks through counterparty exposures, funding markets, and asset fire sales rather than through the affiliations the wall governed. Separation reduces some channels of contagion but not the channels that mattered. The 2010 model’s answer is direct: living wills that rehearse the failure, enhanced supervision that monitors the interconnections, and the orderly liquidation authority as the backstop when bankruptcy cannot handle the firm’s scale. Whether the rehearsal survives contact with a real panic is unproven, but the model at least aims at the observed failure mode rather than at an idealized one.

Run together, the four risks vindicate the article’s verdict while honoring the separation model’s strengths. The 1933 model dominates on conflicts of interest, where its structural solution is genuinely cleaner than any conduct regime. The 2010 model dominates on liquidity and disorderly failure, the risks that actually destroyed firms in 2008. Proprietary trading is the contested middle ground, where the Volcker Rule represents the 2010 model’s concession that some 1933 instincts were sound. A reader who weights conflicts most heavily will lean toward separation. A reader who weights runs and collapses most heavily will lean toward capacity. The evidence of 2008 weights the second pair more heavily, because the crisis was a crisis of funding and failure, not of the classic affiliate abuses. But the weighting is a judgment, and the article’s contribution is to make the judgment explicit rather than to pretend the evidence makes it alone.

What the 2010 model controls: capital, stress, activities, and resolution

The 2010 statute’s answer to the separation question is a control system, and the control system has four interlocking parts. Each part addresses a distinct way a large financial firm can endanger itself and the system around it. Taken together they represent the most ambitious attempt in American history to make bigness safe rather than to forbid it.

The first part is capital, and it is the load bearing element of the entire model. Under the United States implementation of the Basel III standards then being phased in, large banking organizations faced minimum common equity tier 1 capital of 4.5 percent of risk weighted assets plus a 2.5 percent capital conservation buffer, for an effective 7 percent before other buffers, with tier 1 capital at 6 percent, total capital at 8 percent, and a minimum leverage ratio of 4 percent. The numbers require translation for the non specialist. A capital ratio is the share of a firm’s funding that comes from its owners’ equity rather than from borrowing. Higher ratios mean the firm can absorb larger losses before its creditors, and ultimately the public, take the hit. The 2010 model bets the system on this arithmetic: if the largest firms hold enough of their owners’ money at risk, the owners will restrain the gamblers, and the losses that do occur will stop at the firm’s equity rather than cascading outward. The Federal Reserve’s stress test publications of the period supplied the enforcement detail, describing the methodology by which supervisors modeled severe economic scenarios and measured whether firms’ capital would survive them.

The statute gave the Federal Reserve a specific mandate to tighten these standards for the largest firms. Section 165 of the act, titled enhanced supervision and prudential standards, requires the Board of Governors to impose more stringent requirements on large bank holding companies and on nonbank financial companies designated as systemically important: tougher capital and leverage rules, liquidity requirements, risk management standards, stress testing, and single counterparty credit limits that cap exposure to any one borrower. The enacted design set the automatic coverage threshold at firms with 50 billion dollars or more in total consolidated assets, a line Congress drew to separate the firms whose failure could threaten the system from those whose failure it could not. A later statute, the Economic Growth, Regulatory Relief, and Consumer Protection Act enacted in 2018, raised the automatic threshold to 250 billion dollars; that change postdates this article and is noted here only as an explicitly dated development. The provision is the clearest expression of the capacity philosophy in the statute’s text. It does not tell large firms to split apart. It tells them to hold more, manage better, and prove it.

The second part is stress testing, the model’s way of verifying that the capital is real rather than notional. Section 165(i) requires annual stress tests, both run by the companies themselves and run by supervisors, under baseline, adverse, and severely adverse scenarios designed by the Federal Reserve. The supervisory program known as the Comprehensive Capital Analysis and Review, whose first round ran in 2009, folds the stress results into a judgment about each firm’s planned capital distributions: a firm that cannot demonstrate survival under the severe scenario cannot pay the dividends and buy back the shares its executives might prefer. This is a subtle but profound shift in the supervisor’s role. Before the crisis, supervision largely checked compliance with static rules. After the 2010 act, supervision models the future, and the model of the future constrains the present. A bank’s capital plan is only as generous as its worst modeled outcome permits.

The third part is activity limits, and here the famous Volcker Rule belongs. Section 619 of the act added a new section 13 to the Bank Holding Company Act of 1956, codified at 12 U.S.C. section 1851, generally prohibiting a banking entity from engaging in proprietary trading, betting the firm’s own capital on market movements, and from acquiring or retaining an ownership interest in, or sponsoring, a hedge fund or private equity fund, the covered funds. The statute carves out enumerated exemptions: market making, underwriting, risk mitigating hedging, trading in government obligations, trading on behalf of customers, and several others. The federal agencies adopted the final implementing rule in December 2013, with the statute scheduling effectiveness for July 21, 2015. The provision’s design is the brief’s partial restoration point made concrete. It restricts an activity inside the combined firm. The firm remains combined. A bank holding company may still own a securities affiliate, still underwrite, still deal. It simply may not run a proprietary trading desk or sponsor the designated funds alongside those businesses. This is a cousin of the 1933 approach, sharing its suspicion of speculative risk inside deposit taking organizations, but it is not a return to structural separation, and calling it one collapses the very distinction this article exists to draw.

The Volcker Rule’s implementation history illustrates both the ambition and the difficulty of the activity limit approach. Congress did not write the definition of proprietary trading in the statute. It directed five federal agencies to write it jointly, listed the activities that must be exempted, required banking entities to establish compliance programs, and set a conformance period for bringing existing activities into compliance. The joint rulemaking drew thousands of public comments, because the central definitional problem is genuinely hard: distinguishing prohibited proprietary trading from permitted market making, the business of holding inventory to serve customer demand, requires regulators to tell the difference between a trader betting the firm’s capital and a trader serving clients, when the two activities can look identical trade by trade. The agencies’ answer relied on metrics, limits, and supervisory judgment rather than on any bright line, which is characteristic of the 2010 model. Where the 1933 regime would have banned the desk, the 2010 regime measures it.

The compliance burden the rule created is itself evidence about the models’ trade offs. Banking entities had to build systems to monitor trading activity against the rule’s metrics, document the purpose of their positions, and demonstrate to examiners that their market making was genuinely customer facing. Supporters of the rule argued this was the necessary price of restraining speculative risk inside deposit taking firms. The separationists argued the complexity proved their point: that policing conduct inside combined firms is so difficult that forbidding the combination would be simpler and more reliable. Both arguments contain truth, and the reader should see the Volcker experience as a controlled experiment in the activity limit philosophy. It works, to the extent it works, by making the prohibited conduct expensive to hide rather than impossible to attempt, which is a different kind of prevention from the wall’s.

How does an activity limit differ from a structural ban?

A structural ban forbids the combination itself, so a bank cannot own a securities firm at all, while an activity limit permits the combination and forbids only specified conduct inside it, which is why the Volcker Rule restricts proprietary trading within combined banking entities rather than breaking those entities apart.

The fourth part is resolution planning, the model’s answer to the too big to fail problem. Section 165(d) requires each covered company to report periodically to the Federal Reserve, the Financial Stability Oversight Council, and the Federal Deposit Insurance Corporation its plan for rapid and orderly resolution in the event of material financial distress or failure. The joint Federal Reserve and FDIC rule implementing the requirement arrived in November 2011, and the first wave of filers submitted their plans on July 1, 2012. The living wills, as they are universally called, force each large firm to map its own legal entities, funding sources, and critical operations, and to explain how it could be wound down without igniting the system around it. Supervisors review the plans for credibility, and the statute gives them leverage when a plan is found wanting. The philosophy is visible in the mechanism. The 1933 model would have made the firm smaller and simpler so that failure was less dangerous. The 2010 model accepts the firm’s size and complexity and demands a rehearsed exit.

Liquidity standards complete the control system, though they arrive through the Basel process rather than the statute’s text. The liquidity coverage ratio requires large firms to hold high quality liquid assets sufficient to cover thirty days of stressed net cash outflows, a full 100 percent coverage requirement, and the net stable funding ratio addresses longer term funding stability. These are Basel Committee standards described here generically, not Dodd-Frank text, but they belong in the picture because they operationalize the same philosophy: the firm may be large and combined, but it must be able to survive a run without emergency public support.

The stress testing regime warrants a closer look at its mechanics, because it is the most novel element of the 2010 model and the one most foreign to the 1933 way of thinking. Each year the Federal Reserve designs scenarios, a baseline reflecting expected conditions, an adverse scenario, and a severely adverse scenario featuring a deep recession, spiking unemployment, collapsing asset prices, and market stress. The covered firms must project their losses, revenues, and capital positions under each scenario over a multi year horizon, and supervisors run their own independent models alongside the firms’ submissions. The results feed directly into the Comprehensive Capital Analysis and Review, under which the Federal Reserve evaluates each firm’s planned dividends and share repurchases. A firm whose capital would fall below the required minima in the severe scenario can be told, in effect, to keep its money. The first round of this supervisory capital assessment ran in 2009, in the crisis’s immediate aftermath, and the discipline it imposed, forcing firms to retain earnings rather than distribute them while the outlook was uncertain, was among the fastest acting elements of the post crisis regime.

The distinction between risk weighted capital ratios and the leverage ratio also deserves explanation, because it reveals how the 2010 model tries to protect itself against gaming. Risk weighted ratios assign different weights to different asset classes, so that a dollar of Treasury securities requires less capital backing than a dollar of corporate loans. The refinement is economically sensible but creates an incentive to accumulate assets the rules treat as safe, which may not be safe in the next crisis’s particular shape. The leverage ratio answers this by ignoring risk weights entirely and requiring equity against total assets, a blunt backstop against the cleverness that risk weighting invites. The two measures work as a pair: the risk weighted ratios calibrate capital to measured risk, and the leverage ratio limits the total scale of the bet. Neither would have been conceivable as a supervisory tool in 1933, and together they embody the capacity philosophy’s faith that measurement, layered and redundant, can substitute for prohibition.

Resolution planning, the living wills, operates on a different logic from the other controls, and its novelty is easily underestimated. Capital, liquidity, and stress tests all aim to prevent failure. The living wills assume failure and demand a rehearsal. Each covered firm must map its legal entities, its funding sources, its critical operations, and its interconnections, and must explain, in a plan reviewed by the Federal Reserve and the Federal Deposit Insurance Corporation, how it could be resolved rapidly and in an orderly manner in material distress without destabilizing the system. The joint agency rule implementing the requirement arrived in November 2011, the first wave of filers submitted in July 2012, and the review process is iterative: supervisors judge the plans’ credibility and push back where the maps are incomplete or the strategies unconvincing. Behind the living wills stands the statute’s orderly liquidation authority, a receivership power for failing systemically important firms as an alternative to ordinary bankruptcy, funded by industry assessments rather than taxpayers. The living will is the plan; the liquidation authority is the tool the plan is meant to make usable. Whether credible plans can actually be executed in the chaos of a real panic remains the most open question in the 2010 design, and the honest answer is that the mechanism is untested at full scale.

Two features of the 2010 design deserve explicit notice because they answer the separation argument on its own terms. First, the statute did not merely permit combination and hope for the best. It built the interaffiliate protections of the old regime into the new one: the Federal Reserve Act’s sections 23A and 23B, which limit transactions between a bank and its affiliates, survived the 1999 repeal and continued to constrain the dealings inside the combined organizations the 2010 act regulates. Second, the statute’s enhanced standards apply to designated nonbank financial companies as well as to bank holding companies, an acknowledgment that systemic risk does not respect the legal form of the firm. That acknowledgment will matter when the article turns to the crisis record, because the firms at the center of the crisis wore precisely the nonbank forms the old affiliation ban never touched. Readers who want the full architecture of the 2010 statute, title by title, should consult the complete guide to the 2010 statute, which carries the detail this comparison summarizes.

The separation alternative Congress considered and rejected

The 2010 law’s philosophy becomes clearest where Congress considered the separation alternative directly and rejected it. The legislative record contains the road not taken, and it is more informative than any retrospective debate, because it shows the choice between form and capacity being made in real time, with the crisis fresh.

The vehicle was the SAFE Banking Act amendment offered by Senators Sherrod Brown, Democrat of Ohio, and Ted Kaufman, Democrat of Delaware, during the Senate’s consideration of the financial reform bill. The amendment would have imposed hard size caps on financial institutions, requiring the largest firms to shrink rather than to capitalize their scale. The amendment’s premise was the separation philosophy’s cousin: that the binding constraint on systemic risk is the size and form of firms, and that firms above a defined scale should not exist.

The Senate rejected the amendment on May 6, 2010. The vote is the clearest legislative expression of the capacity philosophy: Congress declined to limit the form and scale of firms directly and chose instead to require the firms, at whatever scale, to fund themselves with more capital, submit to stress testing, and plan their own orderly resolution. The 2010 law as enacted contains no size caps, no breakup mandate, and no restoration of the affiliation restrictions.

The rejection did not come out of nowhere. During the House’s 2009 consideration of the reform bill, Representative Maurice Hinchey, Democrat of New York, proposed an amendment that would have reenacted Glass-Steagall sections 20 and 32 and prohibited bank insurance activities; it was never brought to a vote. In the Senate, John McCain, Republican of Arizona, and Maria Cantwell, Democrat of Washington, introduced the Banking Integrity Act of 2009, designated S. 2886, on December 16, 2009, which would have reinstated the two repealed sections; it too was never voted on. The restoration argument thus had legislative vehicles across 2009 and 2010, and none of them advanced. The pattern is consistent: the separation alternative was available, visible, and repeatedly set aside.

What Congress enacted instead is the compromise between the two philosophies. Section 121 of the act, codified at 12 U.S.C. section 5331, authorizes the Federal Reserve, with a two-thirds vote of the Financial Stability Oversight Council, to require a financial company to divest assets or off-balance-sheet items if the company poses a grave threat to financial stability. This is a conditional breakup authority, not a structural rule: the firms may exist at their scale unless and until regulators determine they pose a grave threat, at which point the divestiture remedy becomes available. The structure of the compromise reveals the hierarchy of the 2010 philosophy. Capacity is the everyday constraint, applied to all large firms through capital and testing. Form is the emergency backstop, available against individual firms that defeat the everyday constraint. Congress did not forget the separation argument in 2010. It considered the strongest legislative version of it, set it aside, and built the alternative.

The firms that failed and the legal forms they wore

The counterfactual at the heart of this comparison, would restoring the 1933 separation have prevented the 2008 crisis, can only be answered firm by firm. The popular telling imagines giant commercial banks, swollen by repeal-era combinations, collapsing under the weight of securities activities the old law would have forbidden. The record shows a different cast. The firms at the center of the crisis wore legal forms that the repealed provisions never governed, engaged in activities the repeal never authorized, and failed for reasons the affiliation ban never addressed. This section walks the casualty list in the order the crisis revealed it, because the legal form of each failure is the evidence on which the counterfactual turns.

The earliest major failures were independent mortgage originators, firms with no connection to the affiliation debate at all. New Century Financial, an independent subprime mortgage originator organized as a real estate investment trust, filed for bankruptcy on April 2, 2007, one of the first large failures of the crisis. It originated mortgages, sold them into securitization pipelines, and held residual risk that destroyed it when defaults rose. No provision of the 1933 act, repealed or surviving, governed its business model, because it was neither a bank nor a securities firm affiliate. It was a specialist lender operating in a market the separation regime never contemplated. Its failure is the first exhibit for the proposition that the crisis began outside the walls, in the originate to distribute machine that neither the 1933 nor the 1999 framework was designed to supervise.

The thrift failures came next, and they illustrate how legal form determined regulatory fate. IndyMac Bancorp, a thrift holding company and the seventh largest mortgage originator in the country, was seized on July 11, 2008, producing a Federal Deposit Insurance Corporation loss of roughly 12 billion dollars, the most expensive failure in the corporation’s history to that point. Washington Mutual, also a thrift holding company supervised by the Office of Thrift Supervision, was seized on September 25, 2008, with 307 billion dollars in assets, the largest bank failure in American history. Both firms were depository institutions, but they were thrifts, not commercial banks in the Glass-Steagall sense, and their holding company structures sat outside the affiliation provisions the 1999 law repealed. They failed the old fashioned way, by making bad mortgage loans and funding themselves with runnable liabilities, and no restoration of the bank securities affiliation ban would have touched their business models. Their regulator, the Office of Thrift Supervision, had permitted the aggressive lending that destroyed them, which is a supervision story rather than a structure story.

The insurer came next. American International Group, a multinational insurance corporation, was rescued on September 16, 2008, with an 85 billion dollar credit facility from the Federal Reserve Bank of New York in exchange for a 79.9 percent equity stake. The losses that destroyed AIG came from its financial products unit’s portfolio of credit default swaps written on mortgage backed securities, an insurance like activity conducted far from any bank. The affiliation provisions of the 1933 act never applied to an insurer’s derivatives book, and the 1999 repeal changed nothing about AIG’s permitted activities. Its rescue is the purest illustration of the capacity theory’s blind spot and the separation theory’s irrelevance in the same episode: the firm that most threatened the system was not a combined bank at all, and no wall between banking and securities would have constrained a swaps portfolio written by an insurance company’s London subsidiary.

Then came the investment banks, the failures most often misattributed to the repeal. Bear Stearns, a free standing investment bank with no commercial banking parent, was acquired by JPMorgan Chase in a rescue arranged in March 2008 and closed in May, supported by a 29 billion dollar loan from the Federal Reserve Bank of New York. Lehman Brothers, also a free standing investment bank and not part of any bank holding company, filed for bankruptcy on September 15, 2008. Merrill Lynch, likewise a free standing investment bank, was acquired by Bank of America for approximately 50 billion dollars on the same September weekend. Goldman Sachs and Morgan Stanley, the last two independent investment banks, converted to bank holding companies on September 21, 2008, submitting themselves to Federal Reserve supervision in exchange for access to the central bank’s liquidity facilities. The critical legal fact about all five firms is the one the repeal debate most often ignores: the repealed sections 20 and 32 barred affiliations between banks and securities firms, but these firms had no bank affiliates to separate. They were pure securities firms, exactly the kind of firm the 1933 design left free to operate independently. The repeal changed nothing about their permitted activities, their leverage, or their funding models. As Peter Wallison has argued, the provisions Congress removed in 1999 governed banks’ affiliations, not securities firms, and the free standing investment banks were never subject to the affiliation ban in the first place.

Countrywide Financial completes the cast and complicates it usefully. The company was an independent mortgage originator that financed roughly 20 percent of all United States mortgages in 2006, was acquired by Bank of America in a transaction announced January 11, 2008, and completed July 1, 2008, and had converted its bank subsidiary to a thrift charter in March 2007, becoming a thrift holding company under Office of Thrift Supervision oversight. Its trajectory shows how firms shopped for charters and regulators in the pre crisis years, and its acquisition by Bank of America was a crisis era resolution, not a product of the affiliation powers the 1999 law created. Thrift and bank holding company mergers of this kind were achievable under pre 1999 law. The same is true of Bank of America’s acquisition of Merrill Lynch. These were rescues arranged under pressure, not combinations enabled by repeal, and treating them as fruits of the 1999 law reverses the arrow of causation.

Arranged chronologically, the failures tell a story about funding as much as about legal form. New Century’s bankruptcy in April 2007 opened the crisis in the originate to distribute sector, where thinly capitalized lenders depended on continuous access to securitization markets that closed without warning. Countrywide’s March 2007 conversion to a thrift charter showed a giant originator shopping for a more permissive supervisor as its model deteriorated, and its January 2008 acquisition by Bank of America showed the system’s first use of a strong balance sheet to absorb a failing specialist. Bear Stearns’s March 2008 rescue exposed the investment banks’ central vulnerability: firms funding long term assets with overnight repurchase agreements and commercial paper faced immediate death when lenders refused to roll their funding, regardless of the accounting value of their assets. A solvent firm can die of illiquidity in days if its liabilities all come due at once, and the broker dealer funding model made that outcome a daily possibility.

The summer and autumn of 2008 then compressed years of deterioration into weeks. IndyMac’s July seizure demonstrated that the thrift sector’s aggressive mortgage lending had created failures the deposit insurance system would have to absorb at great cost. The September cascade, Lehman’s bankruptcy and Merrill Lynch’s acquisition on the fifteenth, the AIG facility on the sixteenth, the Goldman Sachs and Morgan Stanley conversions to bank holding companies on the twenty first, and the Washington Mutual seizure on the twenty fifth, showed every legal form failing in rapid succession: the pure investment bank, the insurer, the thrift, all within ten days. The variety is the point. No single charter, no single business model, no single regulator’s jurisdiction contained the damage, because the damage came from common exposures, leverage, short term funding, and concentrated housing risk, that cut across every legal boundary the system recognized.

The funding model analysis sharpens the counterfactual further. The investment banks did not fail because they combined commercial and investment banking. They failed because they combined long dated, illiquid assets with overnight liabilities, a maturity mismatch that made them dependent on the continuous confidence of short term lenders. The thrifts did not fail because of securities affiliations. They failed because they made bad loans with insufficient capital and, in Washington Mutual’s case, suffered a depositor run that the insurance system’s limits could not contain. AIG did not fail because of any bank affiliation. It failed because a derivatives book written far from its insurance regulators created exposures its capital could not absorb. In each case the mechanism of failure is visible without reference to the 1933 provisions, and in each case the mechanism is one the 2010 model’s instruments, capital against the assets, liquidity against the runnable funding, stress tests against the concentrated exposures, are designed to measure. That is why the firm record supports the capacity model: the failures were failures of resources, not of form.

Which crisis-era firms would a restored affiliation ban have covered?

Almost none of them. The repealed provisions governed affiliations between banks and securities firms, but the central failures were free-standing investment banks, thrift holding companies, an insurer, and independent originators, none of which the affiliation ban ever directly reached at all.

The honest assessment follows directly. Restoring the affiliation restrictions of sections 20 and 32 would not by itself have prevented these failures, because the restrictions would not have applied to the firms that failed. The investment banks were already separate. The thrifts were already outside the provisions. The insurer and the originators were in businesses the provisions never addressed. This is the core of the case against the popular causal story, and it is why the chronology, repeal in 1999, crisis in 2008, response in 2010, cannot by itself establish causation. A sequence is not a mechanism, and the mechanism the repeal created, supervised affiliation inside financial holding companies, is not the mechanism by which these firms collapsed.

But the honest assessment has a second half, and the article gives it full weight. A defensible argument remains that separation would have limited the scale and interconnection of the largest institutions, even if it would not have prevented the specific failures. That argument does not claim the repeal caused Lehman’s bankruptcy. It claims the repeal permitted the growth of enormous combined firms whose size, complexity, and interconnection made the system fragile, and that a separation regime would have kept the largest firms smaller and simpler. This is the strongest version of the separation case, and it is about scale and complexity rather than about the specific prohibited combinations. The next section presents it through its named advocates, alongside the named analysts who reject it. Readers who want the full failure record, institution by institution, should consult the crisis legislation record, which carries the detailed chronicle this comparison summarizes.

The causal debate, with names attached

The question whether the 1999 repeal caused the 2008 crisis is the most argued question in American financial policy, and it deserves an argument rather than a slogan. This section presents the strongest version of each side, attributed to the analysts who actually made the claims, and then states what the evidence establishes and what it does not. The reader should emerge able to steelman both positions, which is the precondition for choosing between them.

The case that repeal contributed begins with culture. Joseph Stiglitz argued in 2009 that the most important consequence of the Glass-Steagall repeal was changing the culture of commercial banking, so that the investment banking culture of bigger risk came out on top. The claim is not that any particular transaction required the repeal. It is that the repeal legitimized a way of doing business, the aggressive, leveraged, fee driven style of Wall Street, inside institutions that had previously been constrained to the staid business of taking deposits and making loans. Once the combination was permitted and then celebrated, the argument runs, the risk culture of the securities business infected the commercial banking business, and the combined firms pursued returns their predecessors would have considered reckless. Robert Weissman, president of Public Citizen, agreed in a statement issued November 11, 2009, on the tenth anniversary of the repeal, arguing that its most important effect was to change the culture of commercial banking to emulate Wall Street’s high risk speculative betting approach. The culture argument is powerful precisely because it does not depend on tracing any single failure to any single provision. It claims the repeal changed the industry’s norms, and changed norms changed behavior across the system.

Robert Kuttner offered the structural version of the same concern. He acknowledged that regulatory inroads had weakened the separation before 1999, but argued that the repeal permitted the creation of super banks that reenacted the same kinds of structural conflicts of interest endemic in the 1920s. The super bank is Kuttner’s central image: an institution so large, so diversified, and so interconnected that its internal conflicts cannot be managed and its failure cannot be contained. On this view the repeal’s significance lies not in any particular combination it enabled but in the scale it licensed. Elizabeth Warren and the economist Richard Wolff have likewise tied the repeal to the crisis in public argument, emphasizing the growth of concentrated financial power the repeal permitted. The separation advocates’ strongest claim, taken together, is a claim about the system’s architecture: repeal allowed bigness and complexity to compound, and bigness and complexity made the crisis worse than it would otherwise have been.

The case against the causal claim begins with a question Alan Blinder posed in a 2009 essay, and the question has never received a satisfactory answer from the repeal’s critics. Blinder asked what bad practices would have been prevented if Glass-Steagall had still been on the books, and reported that he had yet to hear a good answer. The mortgage underwriting at the center of the crisis, which he called disgraceful, did not rely on any new powers the 1999 law created. The major producers of the dodgy mortgage backed securities were free standing investment banks, not the combined conglomerates the repeal permitted. Blinder concluded that he could not see how the crisis would have been any milder if the 1999 law had never passed. The force of the question is its demand for a mechanism. A causal claim needs a channel, and Blinder’s challenge is that no critic has specified which crisis era practices the unrepealed provisions would actually have forbidden.

Lawrence White and Jerry Markham supplied the legal complement to Blinder’s economic challenge. They argued that the products and practices linked to the crisis were not regulated by Glass-Steagall in the first place, or were already available to commercial banks and their affiliates before the 1999 law repealed sections 20 and 32. The securitization machine, the derivatives exposures, the leveraged funding models: these were creatures of markets and of regulatory decisions far removed from the affiliation ban. White added a counterfactual of his own, arguing that if the separation regime had still been on the books in 2007 and 2008, the large commercial banks could not have absorbed Bear Stearns and Merrill Lynch, which is to say the repeal arguably aided crisis resolution rather than causing the crisis. The acquisitions that stabilized two failing investment banks were possible because the affiliation wall was gone. On this telling the repeal’s absence would have made the panic worse, not better.

Peter Wallison made the most systematic version of the case against causation. In a November 12, 2009, policy brief for the Networks Financial Institute titled in substance Not Guilty, Not Even Close, and again in a 2012 essay on the myths surrounding the repeal, Wallison argued that Glass-Steagall as it related to banks was never repealed at all, that the surviving sections 16 and 21 continued to do the work the public imagines was abandoned, and that the activities at the center of the crisis were available before the 1999 law. The brief’s evidence supports the central legal plank: only the affiliation provisions fell, and the affiliation provisions never governed the firms that failed. Wallison’s broader historical claim is that the crisis was a government housing policy failure compounded by mark to market accounting and other factors, a position this article notes without adjudicating, because the article’s assignment is the comparison of regulatory models rather than a full theory of the crisis.

Melanie Fein, a banking lawyer and author of a leading treatise on the securities activities of banks, supplied the decisive point on securitization, the activity most often assumed to be the repeal’s fruit. Fein demonstrated that the 2008 crisis was not a result of the 1999 law and that the law authorized no securities activities that caused the crisis. Mortgage securitization by national banks had been approved by the Office of the Comptroller of the Currency as early as 1978 for Bank of America and 1987 for Security Pacific, and the courts had confirmed by 1990, when the Supreme Court declined review of a Second Circuit decision, that securitization was part of the business of banking. The claim that repeal enabled securitization is therefore false as a matter of legal history. The machine that manufactured mortgage backed securities was running, with regulatory approval, more than two decades before the affiliation ban fell.

What does the evidence establish, and what does it not? It establishes that the repeal did not cause the specific failures at the crisis’s center, because the legal mechanism of the repeal did not reach those firms. It establishes that the most commonly cited transmission channel, securitization, predated the repeal by decades. It establishes that the surviving provisions continued to bar the direct combinations the public imagines were unleashed. It does not establish that the repeal was harmless, because the scale, complexity, and culture arguments operate at the level of the system rather than the firm, and system level claims are harder to falsify. It does not establish that separation would have made the crisis milder, because the counterfactual requires specifying which combinations would have been forbidden and the crisis’s central firms were already separate. The intellectually honest position, and the one this article defends, is asymmetrical: the affirmative causal claim fails on the firm level evidence, while the scale and complexity claim survives as a plausible but unproven argument about the system’s architecture. A reader who finds the culture and scale arguments persuasive is entitled to favor separation. A reader who demands a mechanism is entitled to find the affirmative case unproven. What no reader is entitled to do is assert the repeal caused the crisis without naming which firms the repealed provisions would have covered, because that is the question Blinder asked and the literature has not answered.

The debate’s resistance to resolution has a methodological root worth naming, because it explains why intelligent analysts can examine the same record and reach opposite conclusions. The culture and scale arguments operate at the level of the financial system, while the evidence against them operates at the level of the firm, and there is no agreed method for adjudicating between levels. Stiglitz’s claim that the repeal changed banking culture cannot be tested by listing the charters of failed firms, because culture is a property of the industry rather than of any institution. Wallison’s claim that the repealed provisions never governed the failures cannot be answered by invoking the atmosphere of the 2000s, because atmosphere is not a mechanism. Each side’s evidence is real, and each side’s evidence is addressed to a different question, which is why the exchange has the character of ships passing rather than of argument engaging.

This identification problem is general to debates about deregulation and crisis, and the reader should recognize its structure wherever it appears. System level claims, that a legal change altered norms, permitted bigness, or encouraged risk taking across the industry, are difficult to falsify because they do not specify which observable outcomes would disprove them. Firm level claims, that specific provisions governed specific failures, are easier to test but may miss the forest for the trees, because a legal change can reshape an industry without governing any particular failure. The intellectually responsible posture, and the one this article adopts, is to report both levels accurately, to demand mechanisms where mechanisms are claimed, and to acknowledge where the evidence runs out. The firm level evidence runs out in favor of the capacity model. The system level argument for separation remains standing, unproven but not disproven, which is why the article presents it at full strength rather than dismissing it.

A final note on the burden of proof clarifies the verdict’s logic. In policy argument, the burden properly falls on those proposing change from the status quo, and the status quo since 2010 is the capacity model. The separation advocates propose rebuilding walls that have been down for a generation, at the cost of forced divestitures, redefined perimeters, and foregone diversification. The evidence they offer must therefore be strong enough to justify the disruption, and on the firm level record it is not: the causal chain from repeal to crisis does not survive examination of which firms failed. The capacity model’s defenders, conversely, must show their instruments work, and the capital data give them a genuine affirmative case. Burdens of proof do not decide the truth, but they discipline the argument, and a disciplined argument is what this comparison has tried to provide.

What the 1933 model got right

An article that verdicts for the capacity model owes the separation model a final accounting of its enduring insights, because a verdict that cannot state what the losing side got right is not a verdict but a dismissal. The 1933 model got three things right, and the 2010 model has not fully answered any of them.

First, it understood that simplicity is a regulatory virtue in its own right. A binary ban can be understood by every bank director, every examiner, and every legislator, without a quantitative staff. The 2010 model’s instruments require expertise to operate and expertise to oversee, which concentrates understanding in a small community of specialists and leaves democratic oversight at a disadvantage. When a rule can only be evaluated by the people it regulates, the political economy of regulation shifts in the industry’s favor. The separationists’ preference for bright lines is not nostalgia. It is a theory about who gets to understand the rules, and a rule nobody outside the industry understands is a rule the industry will eventually shape.

Second, the 1933 model understood the limits of supervisory attention. The capacity model assumes supervisors will run the stress scenarios honestly, object to inadequate capital plans, demand credible living wills, and enforce the Volcker metrics against resistance. Each of these tasks is demanding, and they must all be performed continuously, across every large firm, through political cycles that reward forbearance. The 1933 model’s great economy was to reduce the supervisory task to boundary policing, which is intermittent rather than continuous. A reader who doubts that supervisory courage can be maintained indefinitely has a rational basis for preferring the model that demands less of it, even at the cost of foregone efficiencies.

Third, the separation theory’s account of conflicts of interest has never been refuted, only outflanked. The firm level evidence shows the 2008 failures did not run through the specific conflicts the 1933 Congress feared, but the logic of the 1920s affiliate abuses, that combined firms will use captive customers and insured funding to support their speculative activities, describes a permanent temptation rather than a historical episode. The 2010 model’s answer is that capital, activity limits, and interaffiliate transaction rules manage the temptation. Perhaps they do. But the separationist’s skepticism, that managed conflicts are still conflicts, and that the management will fail under pressure, is a serious position about human behavior under incentives. The verdict for capacity in this article rests on the evidence of 2008, not on a refutation of that skepticism, and a future crisis that vindicated it would require the verdict’s revision.

The measurement problem both models share

Before the restoration proposals and the verdict, the article pauses on a difficulty that afflicts both models, because acknowledging it is what separates analysis from advocacy. Every regulatory regime creates incentives to evade it, and the relevant question is never whether evasion is possible but how it happens and whether the regime can adapt. The two models fail this test in characteristically different ways, and a reader who understands both failure modes will hold the verdict that follows with appropriate humility.

The separation model’s characteristic failure is perimeter erosion. A wall is only as good as its perimeter, and finance is exceptionally skilled at redrawing perimeters. The pre 1999 history is the textbook case: more than a decade of regulatory interpretations progressively widened the securities activities bank holding companies could conduct, until the affiliation ban constrained less and less of the actual market. But the deeper problem is that the wall never covered the whole financial system. Thrifts, insurers, independent mortgage originators, money market funds, and the securitization conduits operated outside the 1933 perimeter entirely, and the 2008 crisis emerged substantially from that outside territory. A separation regime that covers only commercial banks will always face the question of what to do about the institutions that perform bank like functions without bank charters. Extend the wall to cover them, and the regime must define its perimeter anew with every financial innovation. Leave them outside, and risk migrates to the uncovered sector, which is precisely what the 1933 theory said risk would do, except that it migrated around the wall rather than through it.

The capacity model’s characteristic failure is measurement failure, in two forms. The first is gaming: risk weights can be optimized, models can be tuned, and firms with the strongest quantitative teams will find the cheapest way to satisfy any formula. The leverage ratio exists as a backstop against this, but the history of financial regulation suggests that every formula eventually meets its optimizer. The second is model error compounded by supervisory hesitation. The 2008 crisis featured AAA rated securities that were not remotely as safe as their ratings implied, which means the measurements on which a capacity regime depends can be wrong in the same direction at the same time across the whole system. And even accurate measurements require supervisors willing to act on them: to object to a capital plan, to demand more equity, to force a living will rewrite, against the resistance of profitable firms and their political allies. The capacity model’s vulnerability is not technical but institutional. It requires a supervisory apparatus of permanent competence and permanent courage, and the historical record of supervision gives a reader reason to ask whether that requirement is realistic.

The common lesson is that regulation chooses which game will be played. Separation chooses a game of boundary drawing, in which the regulator’s task is to keep the perimeter current against innovation. Capacity chooses a game of measurement, in which the regulator’s task is to keep the models honest against optimization. Neither game is winnable in any final sense. The honest case for each model is comparative: that its game is more winnable than the alternative’s, given the institutions actually available. The separation advocate must argue that boundaries, though erodible, are easier to police than balance sheets. The capacity advocate must argue that measurement, though gameable, captures more of the actual danger than legal form. This article’s verdict rests on the second argument, but the first deserves its standing as the serious alternative rather than as a relic.

The proposals to restore separation

The separation debate did not end with the crisis postmortems. It returned to Congress as legislation, and the restoration proposals deserve a neutral presentation, by sponsor and by text, because they are the vehicle through which the 1933 model might return. This section describes the leading proposal as it stood in the 113th Congress, attributes the claims made for and against it, and takes no position on its merits.

The 21st Century Glass-Steagall Act was first introduced in July 2013, in the 113th Congress, by Senators Elizabeth Warren of Massachusetts, John McCain of Arizona, Maria Cantwell of Washington, and Angus King of Maine. The sponsors reintroduced the measure in the 115th Congress as S. 881 in April 2017; it was referred to the Senate Banking Committee and did not advance to enactment. The bipartisan sponsorship was the bill’s political signature: a Democrat associated with the consumer protection movement, a Republican associated with the critique of concentrated financial power, and two colleagues from across the aisle and the independents’ bench. The bill’s purpose, as stated in its findings, was to separate traditional banks with Federal Deposit Insurance Corporation insurance from riskier financial activities, specifically investment banking, insurance underwriting, swaps dealing, and hedge fund and private equity activities, and to clarify regulatory interpretations that its sponsors argued had undermined the original 1933 act. In substance the proposal would have rebuilt the affiliation wall the 1999 law removed, and extended the separation logic to activities, such as swaps dealing, that did not exist in their modern form when the original wall was built.

The proposal had not been enacted as of this article’s date, and the opposition to it was sustained and substantive. Analysts associated with the American Enterprise Institute argued that no causal link connected the repeal to the crisis, a position consistent with the firm level evidence this article has presented, and contended that a restored separation would sacrifice the benefits of diversification and crisis era absorptive capacity without addressing the actual mechanisms of the 2008 collapse. Supporters of restoration, including the bill’s sponsors, argued that the scale and complexity of the combined firms remained the central danger and that only structural separation could reliably constrain them. The article reports these positions as positions. The sponsors’ claims about the bill’s purpose and effects are the sponsors’ claims. The opponents’ claims about its costs are the opponents’ claims. The reader now possesses the statute level knowledge to evaluate both, which is the article’s contribution to the debate rather than a vote in it.

Two cautions about the restoration discussion are worth stating explicitly, because they are the neutrality flags the subject requires. First, the restoration debate is a social media perennial, and the perennial version of the argument routinely asserts as fact that the 1999 law fully repealed the 1933 act and that the 2010 act restored it through the Volcker Rule. Both assertions are false, as this article has shown, and any evaluation of the restoration bills that proceeds from those premises will misjudge what the bills would actually change. Second, describing a bill’s text is not endorsing it. The article’s verdict, which follows, concerns the regulatory models as evidenced by the crisis record and the capital data. It is not a recommendation about pending legislation, and readers should not treat it as one.

Beyond the politics, any restoration proposal must answer a set of mechanical design questions, and laying them out clarifies what the bills would actually require. First is the transition problem. Rebuilding the affiliation wall would force existing financial holding companies to choose: divest their securities and insurance affiliates, or shed their bank charters and operate without access to insured deposits and the federal safety net. The divestitures would have to occur on a statutory timetable, recreating in reverse the conformance clock the Federal Reserve set for Citigroup in 1998, and the sales would occur under regulatory compulsion, which affects the prices sellers can obtain and the set of willing buyers. The 1999 transition moved from prohibition to permission, which the market welcomed. A restoration would move from permission to prohibition, which the market would resist, and the difference matters for the feasibility of any timetable Congress sets.

Second is the boundary problem in its modern form. The 1933 categories were written for a financial system of banks and securities firms. The modern system includes swaps dealing, securitization pipelines, money market funds, and complex derivatives intermediation that sort poorly into those categories. The 2013 bill answers by listing the activities to be separated, investment banking, insurance, swaps dealing, hedge fund and private equity activities, but every list creates a new perimeter, and the perimeter problem that eroded the original wall would begin anew the day the new wall was built. Supporters of restoration accept this as the permanent work of supervision. Opponents argue it shows the futility of the approach: the wall must be redrawn continuously, while the activities it excludes reorganize themselves around each redrawing.

Third is the international dimension, which the 1933 Congress never faced in its current form. American banks compete with foreign universal banks that combine commercial and investment banking under their home countries’ rules. A restored American separation would not bind those competitors, and the business now conducted inside American financial holding companies would have an incentive to migrate to foreign firms or to the nonbank sector. Whether that migration would reduce American systemic risk or merely relocate it is a genuine analytical question, and the two sides answer it differently: separation advocates argue that relocating risk outside the safety net is itself a gain, while capacity advocates argue that risk outside the regulatory perimeter is harder to measure and therefore more dangerous. The article takes no position on the pending bills, but a reader evaluating them should demand answers to all three questions, transition, boundary, and migration, because a restoration proposal without such answers is a slogan rather than a program.

The verdict: the form-versus-capacity choice

The article has now assembled everything the verdict requires: the four provisions and their fates, the two philosophies, the control system the 2010 act built, the firm by firm crisis record, and the attributed debate. The verdict question is the one the brief poses directly. Is the binding constraint on systemic risk the organizational form of firms or their loss absorbing capacity? The answer determines which model a reader should prefer, and the article now says which the evidence better supports and why.

The form-versus-capacity choice: the entire structural-separation debate reduces to whether risk is best contained by the shape of firms or by the capital they hold against losses, and once a reader sees that, the 1933 and 2010 statutes stop looking like successive attempts at the same thing.

The deciding factor deserves a plain statement before the evidence is weighed. A reader who believes the binding constraint is organizational form, that combined firms will inevitably abuse their combinations no matter how thick their cushions, should prefer the 1933 model and support restoring its affiliation bans. A reader who believes the binding constraint is loss absorbing capacity, that well capitalized firms can safely house combinations and that thinly capitalized firms are dangerous whatever their shape, should prefer the 2010 model and judge it by whether its cushions prove thick enough. There is no neutral ground between these positions, because they disagree about human behavior inside firms, and the evidence can inform that disagreement without settling it by logic alone.

What evidence decides whether form or capital is the binding constraint?

Two bodies of evidence bear on it: the legal forms of the firms that actually failed, which show that form did not bind the crisis, and the measured growth of bank capital under the post-2010 regime, which shows that capacity can be built by rule.

On the evidence, the capacity model is better supported, for three reasons that build on each other. First, the crisis record favors capacity over form. The firms that failed were already separate, already outside the affiliation ban, already in the shapes the 1933 model prescribes for securities firms, and they failed anyway, from thin capital, runnable funding, and concentrated mortgage risk. If organizational form were the binding constraint, the pure investment banks should have been the safe firms. They were the first to fall. The thrift holding companies, the insurer, and the independent originators complete the pattern: danger wore many legal forms, and the common element across the failures was not combination but leverage, the shortage of loss absorbing capacity relative to risk. This is the firm level evidence, and it is the strongest card the capacity model holds.

Second, capacity has proven buildable by regulation in a way that vindicates the 2010 model’s central bet. The Federal Reserve’s stress test data show the common equity capital ratio of the largest banking organizations roughly doubling from about 5 percent in the first quarter of 2009 to about 12 percent by the fourth quarter of 2019, with total common equity rising by more than 720 billion dollars to approximately 1.2 trillion dollars, figures the Federal Reserve reported in releases dated April 2018 and March 2020. A note on measurement is required for honesty: the early 2009 figures use the pre Basel III definition of tier 1 common capital, while the later figures use common equity tier 1, and the Federal Reserve’s own publications flag the definitional change, so the series is not one seamless metric. But the direction and magnitude survive the caveat. The 2010 model set out to make large firms hold dramatically more of their owners’ money at risk, and by the measure the statute itself established, it did. Readers who want the full capital record should consult the analysis of the 2010 act’s banking impact, which carries the detailed evidence this comparison summarizes.

Third, the separation model’s strongest argument, the scale and complexity claim, is real but does not rescue the firm level causal story. Stiglitz’s culture argument and Kuttner’s super bank argument identify genuine costs of bigness: the largest combined firms are harder to supervise, harder to resolve, and more entangled with the system around them. The 2010 model’s answer to those costs is the enhanced prudential standards, the stress tests, and the living wills, and a fair minded reader must ask whether those controls are equal to the complexity they govern. That is the right question, and it is an open one. But it is a different question from whether the repeal caused the crisis, and it does not convert the 1933 model into the proven superior alternative. It converts the debate into what it has always been: a judgment about whether supervisors can oversee complexity, which is the capacity theory’s vulnerability, against a judgment about whether simplicity is worth the foregone efficiencies, which is the separation theory’s cost.

There is a final way to frame the choice that clarifies what is truly at stake, and it comes from thinking about the two kinds of regulatory error. Every regime makes mistakes, and the mistakes come in two types. The separation model’s characteristic error is the false positive: forbidding combinations that would have been safe and productive, sacrificing diversification, economies of scope, and crisis absorptive capacity to prevent abuses that might never have occurred. The capacity model’s characteristic error is the false negative: permitting combinations that prove dangerous, trusting measurements that miss the next crisis’s shape and supervisors who hesitate to act. The 1933 regime erred, if it erred, by being too strict. The 2010 regime errs, if it errs, by being too permissive.

Which error is costlier is the deep normative question beneath the technical debate, and it does not have a technocratic answer. A reader who fears the false negative more, who believes the next crisis will come from a permitted combination the measurements missed, will prefer the wall despite its costs. A reader who fears the false positive more, who believes the costs of foregone efficiency and credit restriction compound year after year while crises are rare, will prefer the measurement regime despite its vulnerabilities. The evidence of 2008 suggests the false negatives of the pre 2010 regime were catastrophic, which is an argument for the capacity model’s instruments rather than against them. But the choice between error types is ultimately a choice about what kind of mistakes a society prefers to make, and an honest verdict names that choice rather than hiding it behind technique.

The verdict, then, is defended but qualified. The evidence better supports the capacity model, because the crisis’s central failures occurred in firms the separation model already governed or never touched, and because the capacity the 2010 model demands has been measurably built. A reader who prefers the 1933 model on scale and complexity grounds holds a respectable position, and this article has presented its strongest version. But that reader should understand what the position requires: a belief that no feasible cushion, no stress test, and no resolution plan can make combined firms safe, a belief the crisis record, in which separate firms failed first, does not obviously support. The form versus capacity choice is ultimately a bet about human behavior under incentives. The evidence suggests incentives respond to capital. It does not prove they respond to nothing else.

The verdict also specifies what would change it, because a defended verdict should state its own falsification conditions. Three developments would weaken the capacity model’s claim. If a future crisis centered on combined firms abusing their combinations in the specific ways the 1933 Congress feared, steering depositors into affiliates’ securities, using the deposit base to fund proprietary speculation, with the combined structure as the mechanism rather than as background, the form theory would gain the firm level evidence it has so far lacked. If the measured capital of the largest firms proved illusory in a stress event, if the ratios said safe and the firms failed anyway, the capacity theory’s central bet would be falsified in the most direct way possible. And if supervisors repeatedly failed to act on accurate measurements, declining to object to capital plans or demand credible living wills when the data warranted it, the institutional critique of the capacity model would move from theoretical to demonstrated. None of these has occurred on the record this article examines. The verdict is therefore conditional on the evidence continuing to look the way it looks, which is the most any verdict about regulatory design can honestly claim.

For readers who want to apply the comparison rather than merely understand it, three questions discipline the judgment of any proposal in this space. First, which failure mode does the proposal target: the abuse of depositors and conflicts of combination, or the collapse of large firms and the rescues that follow? A proposal aimed at the wrong failure mode for the current system is misdirected however elegant its design. Second, which firms does the proposal actually cover: does its legal mechanism reach the firms whose behavior it seeks to change, or does it, like the repealed affiliation ban in 2008, govern a category adjacent to the danger? Blinder’s unanswered question is the permanent form of this test. Third, what evidence would falsify the proposal’s theory: what observable outcome would show its advocates they were wrong? A proposal whose advocates cannot name such an outcome is an article of faith, not a policy. Run any restoration bill, any capital rule, any activity limit through these three questions, and the form versus capacity choice stops being abstract. It becomes a method.

One final consideration belongs in the verdict, concerning the asymmetry the article noted in the causal section. The affirmative claim that repeal caused the crisis fails on the firm level evidence, while the scale and complexity claim survives as plausible but unproven. This asymmetry has a practical consequence for how the reader should weigh the two models going forward. The capacity model must be judged by its instruments: are the capital ratios high enough, the stress scenarios severe enough, the living wills credible enough, the supervisors willing enough? These are measurable, debatable, adjustable questions, and the 2010 design’s virtue is that it makes them askable every year. The separation model must be judged by a counterfactual that can never be run: would the system have been safer if the largest firms had been smaller and simpler? The question is legitimate, and Stiglitz’s cultural version of it may capture something real about how norms propagate inside large organizations. But a regulatory philosophy that cannot specify its falsification conditions asks the reader for trust rather than offering evidence, and trust is a scarce commodity in financial regulation for good historical reasons.

Which to study first: a reading order

This article carries the series’ study path recommendation for the comparison, and the recommendation depends on the reader’s purpose. A student meeting the two models for the first time should study the 1933 model first. It is the simpler machine: four provisions, a binary ban, a philosophy stated in one sentence. Its vocabulary, affiliation, underwriting, separation, is the vocabulary the later debate assumes, and a reader who learns it first will read the 2010 statute with comprehension rather than bewilderment. Then the student should study the 2010 model, which is the more complex machine and the live law, and ask at each provision which 1933 problem it answers differently. A reader whose purpose is practical, a policymaker, a journalist, an investor, should reverse the emphasis without reversing the order: learn the 1933 model for the vocabulary, then spend the serious effort on the 2010 model’s control system, because that is the regime governing the firms in the reader’s portfolio or jurisdiction.

For readers deciding what to study first across the whole series, the series study guide lays out the recommended sequences by goal, and this comparison sits in the financial regulation cluster as the capstone that should be read after the two statute guides it compares. Students working through the cluster should keep your statute notes, citations, and case chronologies together free on VaultBook, and those preparing for coursework or examinations should practice and revise US government and civics material on ReportMedic, which carries structured revision material for the regulatory frameworks this article compares.

The comparison this article set out to resolve is now resolved, on the terms the brief required. The fundamental difference is structural separation of activities into different firms versus permission of combination under capital requirements and activity limits. Two of the four 1933 provisions still stand, and the wall that fell was the affiliation wall. Restoring the separation would not by itself have prevented the 2008 crisis, on the firm by firm evidence, though the scale and complexity argument for separation remains the strongest version of its case. And the defended verdict is that the evidence better supports the capacity model, with the deciding factor named: whether a reader believes the binding constraint on systemic risk is organizational form or loss absorbing capacity. A reader who can state that sentence, and defend a choice under it, has passed the article’s test.

A concrete study sequence makes the recommendation usable. Begin with the four provisions themselves, reading the United States Code citations as the anchors: section 377 for the affiliation ban, section 78 for the interlocking ban, section 24(Seventh) for the underwriting restriction, section 378 for the deposit taking ban. A student who can state what each provision forbade, and which two survive, has the foundation everything else builds on. Next, study the 1999 repeal mechanics: section 101’s surgical removal of the affiliation pair, the financial holding company election and its requirements, and the Citicorp Travelers episode as the forcing event. Then turn to the crisis firm by firm, assigning each failure its legal form, investment bank, thrift holding company, insurer, independent originator, until the pattern of forms outside the repealed provisions becomes visible rather than asserted.

Only then take up the 2010 control system, and take it up in the order the statute presents its logic: capital first, as the load bearing element, then stress testing as the verification, then activity limits as the targeted prohibitions, then resolution planning as the rehearsal for failure. At each stage, ask the comparative question the article has modeled: which 1933 problem does this provision answer, and does it answer it better or worse than the wall did? Finally, read the debate literature in chronological order, starting with the 2009 essays, Blinder’s question, Stiglitz’s culture argument, Weissman’s anniversary statement, Wallison’s policy brief, because the arguments were sharpest when the wreckage was freshest, and later commentary largely elaborates positions staked in those first two years. A student who completes this sequence will not merely know the two models. The student will own the comparison, which is the series’ aim for every article it publishes.

Frequently Asked Questions

Q: What is the difference between Glass-Steagall and Dodd-Frank?

Glass-Steagall, the conventional name for sections 16, 20, 21, and 32 of the Banking Act of 1933, used structural separation: it forbade certain combinations of financial activity inside one organization, on the theory that risk migrates across activities and conflicts of interest cannot be managed. The 1999 Gramm-Leach-Bliley Act repealed only the affiliation provisions, sections 20 and 32, leaving sections 16 and 21 in force. Dodd-Frank, Public Law 111-203 of 2010, uses a different philosophy entirely: it permits combined firms and controls risk through capital requirements, liquidity standards, supervisory stress tests, activity limits such as the Volcker Rule, and resolution planning. One model polices the shape of firms, the other polices their resources and conduct.

Q: Would Glass-Steagall have prevented the 2008 crisis?

On the firm by firm evidence, restoring the repealed affiliation provisions would not by itself have prevented the crisis. The central failures wore legal forms the repealed provisions never governed: Lehman Brothers, Bear Stearns, and Merrill Lynch were free standing investment banks with no bank affiliates to separate; Washington Mutual and IndyMac were thrift holding companies; AIG was an insurer; and New Century and Countrywide were independent mortgage originators. Sections 20 and 32 barred affiliations between banks and securities firms, so they changed nothing about these firms’ permitted activities. The honest counterpoint is that separation might have limited the scale and interconnection of the largest institutions, which is the strongest version of the separation argument, but it is about architecture rather than the specific failures.

Q: Is the Volcker Rule a return to Glass-Steagall?

No. The Volcker Rule, section 619 of Dodd-Frank codified at 12 U.S.C. section 1851, is an activity restriction inside the combined firm, not a structural separation between firms. It generally prohibits a banking entity from engaging in proprietary trading and from sponsoring or owning hedge funds and private equity funds, subject to exemptions for market making, underwriting, risk mitigating hedging, and similar activities. The firm remains combined: a bank holding company may still own securities affiliates and still underwrite and deal. Glass-Steagall’s repealed provisions, by contrast, forbade the affiliation itself. The Volcker Rule is best understood as a cousin of the 1933 approach, sharing its suspicion of speculative risk inside deposit taking organizations, rather than a restoration of it.

Q: Which is stronger, Glass-Steagall or Dodd-Frank?

Strength depends on what is being measured, which is why the question needs the form versus capacity framing. Glass-Steagall’s separation was stronger as a prohibition: its bans were binary and simple to enforce, since the combination was either lawful or not. Dodd-Frank is stronger as a measurement system: it imposes graduated capital requirements, liquidity standards, annual stress tests, enhanced prudential standards, and resolution plans, and it reaches nonbank firms the old law never touched. The 1933 model cannot distinguish a prudent combination from a reckless one, while the 2010 model cannot function without competent continuous supervision. On the crisis evidence, the capacity model addresses the actual failure mechanisms, thin capital and runnable funding across many legal forms, more directly than restored separation would.

Q: Why has Glass-Steagall never been restored?

Restoration has been proposed but never enacted, and the reasons are substantive rather than merely political. The 21st Century Glass-Steagall Act, first introduced in July 2013 by Senators Warren, McCain, Cantwell, and King, would rebuild the affiliation wall and extend separation to newer activities such as swaps dealing, but it did not advance to enactment. Opponents argue, with analysts such as Peter Wallison and Alan Blinder, that the repeal did not cause the crisis, since the failed firms were never governed by the repealed provisions, and that restored separation would sacrifice diversification and crisis era absorptive capacity. Supporters counter that only structural separation can constrain the scale and complexity of the largest firms. The debate persists because it is ultimately a judgment about whether supervision can oversee complexity.

Q: Does Glass-Steagall separate firms while Dodd-Frank only limits activities?

That is a useful first approximation but it needs two qualifications. First, Glass-Steagall’s separation was only ever partial: even before 1999 it permitted banks to deal in bank eligible securities, and after the 1999 repeal of sections 20 and 32, combined firms were permitted under the financial holding company regime, with sections 16 and 21 still barring direct underwriting by banks and deposit taking by securities firms. Second, Dodd-Frank does more than limit activities: its capital, liquidity, stress testing, and resolution planning requirements are resource and planning controls, not just activity bans. The cleaner statement is that the 1933 model prohibited combinations while the 2010 model permits combinations under measured constraints, with the Volcker Rule’s activity limits as one component.

Q: Which parts of Glass-Steagall are still law?

Sections 16 and 21 of the Banking Act of 1933 remain in force. Section 16, codified at 12 U.S.C. section 24(Seventh), still prohibits banks from underwriting or dealing in most securities directly, with dealing permitted in bank eligible securities such as United States government obligations and general obligation municipal bonds, plus a narrow 1999 amendment allowing well capitalized banks to underwrite municipal revenue bonds. Section 21, codified at 12 U.S.C. section 378, still prohibits securities firms from taking deposits. Sections 20 and 32, the affiliation ban and the interlocking management ban, were repealed by section 101 of the Gramm-Leach-Bliley Act in 1999. Congressional Research Service analyses confirm this division explicitly.

Q: Should a student study Glass-Steagall or Dodd-Frank first?

Study Glass-Steagall first. It is the simpler machine: four provisions, a binary ban, and a philosophy expressible in one sentence, which makes it the ideal introduction to the vocabulary of affiliation, underwriting, and separation that the entire later debate assumes. A student who learns the 1933 model first will read Dodd-Frank with comprehension rather than bewilderment, asking at each provision which 1933 problem it answers differently. Then study Dodd-Frank as the live law and the more complex machine, spending the serious effort on its capital, stress testing, and resolution framework. The series study guide lays out recommended sequences by goal, and this comparison is designed as the capstone to be read after both statute guides.

Q: Why do so many accounts describe Glass-Steagall as fully repealed?

The full repeal story is simpler and more dramatic than the truth, which is why it survives. Saying Congress tore down the wall between banking and commerce in 1999 makes a cleaner narrative than saying it repealed the affiliation provisions while leaving the direct activity bans in force, and political argument on both sides has found the simple version useful: critics of deregulation cite it as the original sin behind the crisis, while defenders of the 1999 law rarely correct a story that makes their achievement sound larger. The legal reality, confirmed by Congressional Research Service analyses, is that only sections 20 and 32 fell, while sections 16 and 21 continue to bar banks from underwriting most securities directly and bar securities firms from taking deposits. Any argument built on the full repeal premise misjudges what actually changed.

Q: Did the repealed provisions regulate investment banks, or only banks’ affiliations?

Only banks’ affiliations. Sections 20 and 32 of the Banking Act of 1933 barred member banks from affiliating with firms engaged principally in securities underwriting and dealing, and barred interlocking management between the two. They said nothing about the conduct of independent securities firms. Lehman Brothers, Bear Stearns, Merrill Lynch, Goldman Sachs, and Morgan Stanley were free standing investment banks with no bank affiliates, so the affiliation ban never applied to them and its repeal changed nothing about their permitted leverage, funding, or activities. This is the central legal fact of the counterfactual debate, and the core of Peter Wallison’s argument that the repeal could not have caused the failures most often attributed to it. The provisions governed the banks’ side of combinations, not the securities industry as such.

Q: What does the Volcker Rule prohibit inside a banking entity?

The Volcker Rule, Dodd-Frank section 619 codified at 12 U.S.C. section 1851, generally prohibits a banking entity from two things: engaging in proprietary trading, meaning trading securities and derivatives for the firm’s own profit rather than for customers, and acquiring or retaining an ownership interest in, or sponsoring, a hedge fund or private equity fund, the so called covered funds. The statute then exempts specified activities, including market making, underwriting, risk mitigating hedging, trading in government obligations, and trading on behalf of customers. Federal agencies adopted the final implementing rule in December 2013, with effectiveness scheduled for July 21, 2015. The design is deliberately an activity restriction within combined firms, which is why it is a cousin of the 1933 philosophy rather than a return to structural separation.

Q: What failure mode was the 1933 structural model built to prevent?

The 1933 model was built to prevent the abuse of depositors and the conflicts of interest the 1933 Congress had investigated in the wake of the 1929 crash. The specific failure mode was a bank using insured deposits to fund securities speculation, or steering its depositors into securities underwritten by its own affiliates, so that the bank profited from the underwriting while the depositor bore the risk. Sections 16 and 21 attacked the direct combination by barring banks from underwriting most securities and barring securities firms from taking deposits, while sections 20 and 32 attacked the indirect combination by barring affiliations and interlocking management. The model assumes that once the combination exists, the conflicts are unmanageable, so the only reliable remedy is to forbid the structure itself rather than police conduct within it.

Q: What failure mode was the 2010 capital model built to prevent?

The 2010 model was built to prevent the collapse of large financial firms and the taxpayer rescues that follow, the too big to fail problem the 2008 crisis exposed. Its failure mode is a large combined firm’s losses spilling into its deposit base and the wider financial system, or its disorderly failure forcing public rescue because no orderly wind down exists. Each component addresses part of that mode: capital requirements ensure losses stop at the firm’s equity, liquidity standards ensure survival through a run, stress tests verify survival under severe scenarios, activity limits restrain the riskiest trading, and resolution plans map an orderly failure. Unlike the 1933 model, which aimed at depositor abuse and conflicts, the 2010 model aims at systemic collapse, which is why asking which model prevents crises requires specifying which failure is feared.

Q: What is Joseph Stiglitz’s argument that repeal helped cause the crisis?

Joseph Stiglitz argued in 2009 that the most important consequence of the Glass-Steagall repeal was cultural rather than transactional. In his account, the repeal changed the culture of commercial banking so that the investment banking culture of bigger risk came out on top, legitimizing aggressive, leveraged, fee driven behavior inside institutions previously constrained to deposit taking and lending. The claim does not depend on tracing any single failure to any single provision. It holds that the repeal altered industry norms, and altered norms altered behavior across the system, encouraging the pursuit of returns the old commercial banks would have considered reckless. Robert Weissman of Public Citizen endorsed the same cultural mechanism in November 2009. The argument’s strength is its breadth; its vulnerability is that system level cultural claims are difficult to falsify with firm level evidence.

Q: What is Alan Blinder’s counter-question about the repeal-caused-the-crisis claim?

In a 2009 essay, the former Federal Reserve Vice Chairman Alan Blinder posed a challenge to critics of the repeal that has never received a satisfactory answer: what bad practices would have been prevented if Glass-Steagall had still been on the books? Blinder noted that the disgraceful mortgage underwriting at the crisis’s center did not rely on any new powers the 1999 law created, and that the major producers of dodgy mortgage backed securities were free standing investment banks, not the combined conglomerates the repeal permitted. He concluded that he could not see how the crisis would have been any milder if the 1999 law had never passed. The question’s force is its demand for a mechanism: a causal claim needs a channel, and Blinder’s challenge is that no critic has specified which crisis era practices the surviving provisions would actually have forbidden.

Q: Can a bank still underwrite securities under the surviving 1933 provisions?

Mostly no, with defined exceptions. Section 16 of the Banking Act of 1933, still in force, prohibits banks from underwriting or dealing in most securities directly. Banks may deal in bank eligible securities, principally obligations of the United States government and general obligation bonds of states and municipalities. The 1999 Gramm-Leach-Bliley Act added one narrow exception, amending section 16 to permit well capitalized commercial banks to underwrite municipal revenue bonds, the non general obligation variety. Broader securities underwriting by banking organizations happens in holding company affiliates under the financial holding company regime, not inside the bank itself, and remains subject to section 16’s restrictions at the bank level. The absolute claim that banks cannot underwrite any securities is therefore false, but the surviving provision still bars the direct combinations the 1933 Congress feared.

Q: Who sponsors the 21st Century Glass-Steagall Act, and what would it do?

The 21st Century Glass-Steagall Act was first introduced in July 2013, in the 113th Congress, by Senators Elizabeth Warren of Massachusetts, John McCain of Arizona, Maria Cantwell of Washington, and Angus King of Maine, a bipartisan group spanning the parties and the independents’ bench. According to its sponsors, the bill would separate traditional banks carrying Federal Deposit Insurance Corporation insurance from riskier financial activities, specifically investment banking, insurance underwriting, swaps dealing, and hedge fund and private equity activities, and would clarify regulatory interpretations the sponsors argued had undermined the original 1933 act. In substance it would rebuild the affiliation wall the 1999 law removed and extend separation logic to activities that did not exist in modern form in 1933. The bill had not been enacted as of this article’s date.

Q: How much did large-bank capital increase after the 2010 act, according to the Federal Reserve?

According to Federal Reserve stress test data, the common equity capital ratio of the largest banking organizations roughly doubled from about 5 percent in the first quarter of 2009 to about 12 percent by the fourth quarter of 2019, with total common equity capital rising by more than 720 billion dollars to approximately 1.2 trillion dollars, as reported in Federal Reserve releases dated April 2018 and March 2020. One measurement caveat is required: the early 2009 figures use the pre Basel III definition of tier 1 common capital, while the later figures use common equity tier 1, and the Federal Reserve’s own publications flag the definitional change, so the series is not one seamless metric. The direction and magnitude nevertheless survive the caveat, and they represent the 2010 model’s central bet paying off in measured form: the data show large firms holding dramatically more of their owners’ money at risk.

Q: What does section 165 of the Dodd-Frank Act require of the largest banks?

Section 165, titled enhanced supervision and prudential standards, requires the Federal Reserve’s Board of Governors to impose more stringent standards on large bank holding companies and on designated nonbank financial companies than apply to smaller firms. The mandated categories are capital, leverage, and liquidity requirements, risk management standards, stress testing, and single counterparty credit limits that cap exposure to any one borrower. The enacted design set automatic coverage at firms with 50 billion dollars or more in total consolidated assets, the line Congress drew to separate firms whose failure could threaten the system from those whose failure it could not. A later statute enacted in 2018 raised the automatic threshold to 250 billion dollars, a dated development outside this article’s coverage period. Section 165(d) adds resolution planning, the living wills, requiring covered firms to file plans for rapid and orderly resolution in material distress. The provision is the clearest expression of the capacity philosophy in the statute’s text.

Q: What deciding factor should a reader use when choosing between the two models?

The deciding factor is whether the reader believes the binding constraint on systemic risk is organizational form or loss absorbing capacity. A reader who believes combined firms will inevitably abuse their combinations, no matter how thick their cushions, should prefer the 1933 separation model and judge restoration proposals by whether they rebuild the affiliation wall. A reader who believes well capitalized firms can safely house combinations, and that thinly capitalized firms are dangerous whatever their shape, should prefer the 2010 model and judge it by whether its capital, stress testing, and resolution requirements prove adequate. On the evidence this article presents, the capacity model is better supported: the crisis’s central failures occurred in firms the separation regime already governed or never touched, and the demanded capacity has been measurably built.