Introduction: judging a law by its own promises

The Dodd-Frank Wall Street Reform and Consumer Protection Act, Public Law 111-203, was signed on July 21, 2010 with two headline promises that have structured every argument about it since. The first was that American banks, above all the largest institutions, would be made safer through stronger capital, tighter supervision, and forward-looking stress tests. The second was that no large financial firm would ever again require a taxpayer bailout, because a new resolution regime would allow even a giant institution to fail in an orderly way. Those were the statute’s own stated aims, and this article measures the law against them rather than against anybody’s wishes about what it should have done.

That discipline matters because the debate over Dodd-Frank has produced two mirror-image forms of overreach. One camp treats the act as the definitive end of systemic risk in American banking, as though the passage of a statute were the same thing as the demonstration of its effects. The other camp treats the act as nothing more than paperwork, as though higher capital ratios and hundreds of finalized rules were the same thing as no change at all. Both readings fail the same test: they substitute a slogan for evidence. The record, examined layer by layer, is more demanding than either slogan allows. Capital at the largest banks rose substantially and measurably. New liquidity requirements were introduced. The number of banks kept falling along a trend that began decades earlier, and the formation of new banks nearly stopped, though researchers disagree about why. Surveys and econometric studies of compliance costs at small banks point in different directions depending on method. Mortgage underwriting tightened, especially for borrowers with weaker credit profiles, and the rules and post-crisis lender caution share the credit and the blame in proportions nobody can cleanly separate. And the single most important question, whether the resolution regime created in Title II of the act actually works, went effectively untested for more than a decade, because the authority was never used and the failures that did arrive were handled through other channels.

This article is the evidence article for the statute. Its structure is five layers of outcomes, each rated as close to settled, genuinely contested, or untested. Every number carries a named source and a period. Where the evidence after the article’s publication window has accumulated, the text marks it explicitly as a dated later development, because this piece is dated January 1, 2014 and later findings should not masquerade as contemporary facts. Where studies disagree, the article shows the range instead of picking a number. Where nothing can be shown, it says so.

The road ahead runs through the five layers in order, then the evidence table that compresses them, then the two directions of overreach, the baseline and the structural alternative that frame the evaluation, and finally the open questions that the evidence cannot yet answer. Readers in a hurry can read the table and the closing checklist; readers who want the argument can read the layers, where every rating is defended at length and every number carries its source and period.

How this evidence review works

An impact article about a financial reform statute faces a measurement problem that a profile of the statute does not. The law is fixed on paper; its effects arrive slowly, interact with the business cycle, and get revised by later legislation. The provisions evaluated here are surveyed in the complete guide to the Dodd-Frank Act, and this review assumes that survey as its starting point rather than repeating it. What follows is the outcomes record: what measurably changed in American banking after 2010, which findings are close to settled, which remain genuinely disputed, which questions went untested, and what research stands behind each answer.

The article uses five layers because the statute’s effects arrived in five distinguishable places. The first layer is capital and liquidity at the largest institutions, the place where the evidence is strongest. The second layer is industry structure: the count of banks and the flow of new charters, where the trend is clear and the cause is disputed. The third layer is the compliance burden on community banks, where method determines the answer. The fourth layer is mortgage credit availability, where the ability-to-repay and qualified mortgage rules tightened underwriting and the attribution question remains open. The fifth layer is the resolution regime, where the central mechanism has never been used and the failures of 2023 were handled through deposit insurance machinery instead.

Each layer gets a rating that is applied strictly. Settled means the direction of the finding is established by multiple sources with consistent periods, even if the interpretation of its adequacy is debated. Contested means competent research reaches genuinely different conclusions, or the direction is clear while the cause is not. Untested means the mechanism in question has not yet operated under the conditions it was built for, so no finding about its performance can exist. A separate artifact, the five-layer evidence table, compresses each layer into one row with the direction of the finding, the period, the principal sources, and the rating, so a reader can check the verdict of any section against the evidence that supports it.

On sources, the article follows two rules. First, every statistic is attributed to its author, publisher, and year, with the period the statistic covers, and any statistic that cannot be attributed is dropped rather than hedged. Second, post-2013 evidence is labeled explicitly as a dated later development. Much of the research that clarifies Dodd-Frank’s effects was published after this article’s date of January 1, 2014, including several studies the analysis depends on. Those studies are not excluded, but they are marked, so the reader always knows which findings were available at publication and which arrived later. No finding in this article is extended to any current proposal, and no estimate from one method is presented as the consensus when the methods disagree.

The honest shape of the answer can be stated in advance. Higher capital is real and measurable. Compliance burden is real and concentrated at the smallest institutions, but its aggregate size is genuinely disputed. The resolution question is open. Everything else is detail in service of those three sentences.

Five kinds of evidence, and what each can prove

The record speaks in five kinds of evidence, and the reader should know which kind is talking at any moment. Supervisory measurement is the strongest kind: capital ratios, liquidity positions, and stress-test results reported on consistent definitions across firms and quarters. It is collected the same way every time and covers the institutions that matter most, and the capital figures in the first layer belong to it. Administrative data count what happened across the industry without saying why: the FDIC’s Quarterly Banking Profile, the charter counts, and the call-report aggregates. The bank-count series and the chartering numbers belong to it. Survey evidence captures perception and self-reported cost, which is valuable and limited at once, because respondents have interests in their answers; the Mercatus and St. Louis Fed studies belong to it, with their limits carried alongside their numbers. Econometric estimation attempts to isolate the statute’s effect by controlling for the business cycle, interest rates, and bank size, and its answers depend on its controls, which is why two serious models can disagree. Official review, GAO reports, CRS analyses, inspector-general evaluations, and the Federal Reserve’s post-mortems, carries the weight of institutions with access to supervisory information, and the 2023 failure narrative rests on it.

The identification problem: why causation is hard here

Before the layers begin, the article owes the reader an explanation of why so many of its ratings are contested rather than settled. The reason is a problem economists call identification: separating the effect of the statute from everything else that was happening at the same time. Dodd-Frank arrived in the middle of the deepest financial crisis since the Great Depression, followed by years of near-zero interest rates, a slow recovery, concurrent international capital reforms, and later a partial rollback of the statute itself. Any change in banking outcomes after 2010 has multiple candidate explanations, and the honest literature is the literature that admits it.

Consider the de novo drought. New bank formation collapsed between 2007 and 2012, a period that contains the crisis, the statute’s passage, and the low-rate environment all at once. A researcher who emphasizes the crisis will point out that starting a bank in 2009 meant raising capital from investors who had just watched banks fail, in an economy with weak loan demand and compressed net interest margins. A researcher who emphasizes the statute will point out that the fixed costs of compliance made the de novo business model, which depends on growing into profitability over several years, much harder to finance. Both mechanisms are plausible, both operated simultaneously, and the data cannot fully separate them because there is no parallel United States that had the crisis without the statute. The Adams and Gramlich estimate that at least 75 percent of the entry decline would have occurred without regulatory change is a model-based statement, not a direct observation, and it should be read as one serious attempt to answer an unanswerable-in-practice question rather than as a measured fact.

The same problem infects the mortgage layer. Underwriting tightened after 2010, and the ATR/QM rule with its 43 percent debt-to-income threshold was binding for a visible segment of borrowers. But lenders were also independently terrified: the crisis had produced enormous putback demands on loans sold to the government-sponsored enterprises, waves of litigation, and a cultural memory of losses that made credit officers cautious regardless of what any rule required. A study that attributes the tightening to the rule must control for the fear, and a study that attributes it to the fear must control for the rule, and neither control is perfect because the two arrived together and reinforced each other. The CFPB’s assessment documents the tightening carefully and attributes it generically to the assessment’s own analysis; it does not claim to have separated the two forces, and this article does not claim that separation either.

Capital is the exception that proves the difficulty, because it is the one layer where the mechanism is direct enough to cut through the fog. The statute and its implementing rules ordered higher capital, supervisors enforced it through stress tests that constrained distributions, and capital rose on the schedule the rules set. The chain from requirement to outcome is short and documented. Even here, though, a purist could note that markets and rating agencies were also demanding thicker cushions after the crisis, so some of the build might have happened under any regulatory regime. The difference is one of degree: for capital, the statute’s contribution is visible in the timing and the supervisory enforcement, while for industry structure and mortgage credit, the statute is one suspect among several with no clean way to assign shares.

There is a further complication that runs across all five layers, which is that the statute did not stay still. The 2018 tailoring changed the coverage of the enhanced requirements, which means the law being evaluated in 2023 was not exactly the law passed in 2010. An evidence review must therefore specify which version of the regime produced which outcome, and must resist the temptation to credit or blame the original statute for effects produced under the revised one. The article handles this by dating its evidence and by labeling the causal weighting of the 2018 changes on the 2023 failures as contested, which is the only rating the record supports.

Finally, there is the problem of the dog that did not bark. If Dodd-Frank prevented a crisis that would otherwise have occurred, that prevention is unobservable: the evidence review can measure capital ratios and charter counts, but it cannot measure the panic that never happened. This asymmetry systematically understates the statute’s benefits in any backward-looking evaluation, because costs are visible in compliance budgets while avoided catastrophes leave no trace. The article acknowledges the asymmetry without trying to correct for it, because correcting for it would require assuming the answer. The reader should keep it in mind as a standing caveat over every contested rating that follows: absence of evidence for a benefit is not evidence of absence, but it is also not evidence of presence, and the article’s job is to report what can be shown.

Layer one: capital and liquidity

Of the five layers, this one is the least disputed, and it is worth understanding exactly why. When the financial crisis of 2007 and 2008 exposed how thinly the largest American banks were capitalized, the policy response centered on loss-absorbing capacity: the cushion of equity that stands between a firm’s losses and its creditors, and ultimately the public. Dodd-Frank’s approach combined statutory direction with an enormous delegation to regulators, and the Federal Reserve’s stress testing and capital regime became the main enforcement instrument. The numbers that resulted are among the most solidly documented facts in the entire post-crisis record.

Did capital at the largest banks actually rise?

The 18 bank holding companies in the Federal Reserve’s 2013 assessment raised their tier 1 common ratio from 5.6 percent at the end of 2008 to 11.3 percent in the fourth quarter of 2012, while aggregate tier 1 common capital rose by 393 billion dollars to 792 billion dollars (Federal Reserve, March 2013).

The answer above deserves unpacking, because a doubling of a ratio can be read in several ways. A tier 1 common ratio expresses the highest-quality equity, common stock and retained earnings, as a share of risk-weighted assets. When the ratio rises, it can mean the numerator grew, the denominator shrank, or both. In this case the numerator story dominates: the 393-billion-dollar increase in aggregate tier 1 common capital across those 18 firms from the end of 2008 to the fourth quarter of 2012 was driven substantially by retained earnings and new equity issuance in the years after the crisis. The Federal Reserve’s March 2013 documentation, Box 1 of the methodology and results release, is the source, and its period runs from the end of 2008, when the crisis had already forced emergency capital raising, through the fourth quarter of 2012, when the major Dodd-Frank capital requirements were phasing in. Starting the measurement at the end of 2008 rather than before the crisis is deliberate on the source’s part: it measures the rebuilding after the collapse, not a comparison with a pre-crisis peak.

A later development, dated January 13, 2014, offered a different cut of the same phenomenon. The Center for Financial Stability reported that the four largest American banks had added 306 billion dollars in tier 1 capital since December 31, 2007, an increase of 103 percent. That figure must be flagged for what it partly is: a crisis-acquisition artifact. Some of the largest firms grew their balance sheets and their capital by absorbing failing institutions during the crisis, so a portion of the increase reflects consolidation rather than organic capital building. The number is still informative, but it cannot be read as a pure measure of the statute’s effect, and the article presents it only with that flag attached.

Another dated later development extended the series further. The Federal Reserve’s March 2020 publication reported that the tier 1 common equity ratio for the firms in its stress tests had risen from 4.9 percent in the first quarter of 2009 to 12.2 percent in the fourth quarter of 2019. This is presented strictly as dated later evidence, outside the article’s publication window, and it confirms the direction established by the earlier data: the upward path of capital ratios continued through the decade as requirements phased in and firms retained earnings. The three dated points together, the March 2013 figures for 2008 to 2012, the January 2014 figures for the big four from 2007, and the March 2020 figures for 2009 to 2019, describe one of the largest sustained increases in bank capitalization in American history.

Liquidity, the second half of this layer, is harder to quantify from the same sources, but the direction of change is not in doubt. Before the crisis, large banks funded themselves with substantial short-term wholesale borrowing and held thin buffers of assets that could be sold quickly in a panic. The post-crisis framework introduced liquidity requirements that forced large firms to hold buffers of high-quality liquid assets sized against potential short-term outflows, and to plan their funding over longer horizons. These requirements phased in across the decade following the statute, alongside the capital rules, and they represented a genuine innovation: American bank regulation before 2010 had nothing comparable in force. The honest statement is that new liquidity requirements were introduced and the loss-absorbing capacity of the largest banks is materially higher than it was in 2007, while the precise measurement of the liquidity buffer at any single date is a regulatory data question beyond the scope of this review.

What does higher capital buy? The settled part of the answer is mechanical. A bank with twice as much equity relative to its assets can absorb twice as much loss before its creditors, and ultimately a public backstop, are at risk. That is why supervisors, researchers, and the banks themselves treated the capital build as the centerpiece of the post-crisis repair. The contested part of the answer is whether the new levels were sufficient, a question about the right amount of capital that no single study can settle because it depends on how severe a future crisis might be. The article does not adjudicate that debate; it records the measurement that everyone can agree on and marks the adequacy question as the place where agreement ends.

There is a related misconception worth clearing away at the start, because it recurs throughout the public argument. Higher capital ratios did not mean banks stopped taking risks, and they did not mean the financial system stopped being capable of producing a panic. Capital is a cushion, not a cure. What the evidence establishes is that the cushion got much thicker. That finding is close to settled, and it is the strongest card in the statute’s hand.

What the ratio measures and what it hides

The tier 1 common ratio deserves a closer look, because its construction shapes what the headline doubling does and does not prove. The numerator is common equity, the purest form of loss absorption: common stock plus retained earnings, the funds that are wiped out first when losses arrive and therefore the funds that protect everyone else. The denominator is risk-weighted assets, a regulatory construct that assigns different weights to different kinds of exposures, so that a dollar of Treasury securities counts for less than a dollar of commercial loans. A ratio of 11.3 percent means the firm holds high-quality equity equal to 11.3 percent of its risk-weighted exposures, and the rise from 5.6 percent at the end of 2008 to 11.3 percent in the fourth quarter of 2012, documented by the Federal Reserve in March 2013, means the cushion roughly doubled on this measure.

What the ratio hides is the denominator’s subjectivity. Risk weights are set by regulators and refined by firms’ internal models, which creates a long-running debate about whether banks optimized their measured ratios by shifting into assets the rules weighted lightly rather than by genuinely reducing risk. The article does not adjudicate that debate, because the verifier’s record does not contain the study that would settle it, but it notes the debate as the reason sophisticated readers treat risk-weighted ratios as one lens rather than the whole picture. Supervisors complemented the risk-weighted framework with leverage constraints that ignore risk weights entirely, measuring equity against total exposures, precisely to guard against denominator gaming. The direction of travel was the same on both families of measures: up.

There is also a distinction worth drawing between how the numerator grew. Retained earnings, profits kept rather than paid out as dividends or used for share repurchases, were the largest contributor across the system, which is why the supervisory constraint on distributions mattered so much. When stress tests limited a firm’s payouts, the retained cash mechanically became capital. New equity issuance contributed as well, particularly in the early post-crisis years when markets were still willing to supply it. The 393-billion-dollar aggregate increase from the end of 2008 to the fourth quarter of 2012, reported by the Federal Reserve in March 2013, blends both sources, and the article presents the blend because the source presents the blend. A reader who wants to know whether the system earned its way to safety or was forced there will find evidence for both in the components.

The parallel track: Basel III and the Collins floor

The statute’s capital requirements did not arrive alone. American regulators implemented the Basel III international framework through rulemakings in 2013, with phase-in extending across the following years, and the statute contributed its own distinctive element in Section 171, the Collins Amendment, which set a floor under the capital requirements of the largest firms. The floor worked by a simple rule: a firm had to satisfy whichever measure, the advanced internal-models approach or the standardized approach, produced the higher requirement, so that internal risk models could not be used to minimize required capital. The design was a direct response to the pre-crisis experience, in which internal models had justified buffers that proved thin in the stress of 2008.

Two cautions belong here. First, post-crisis capital was measured under a stricter regime than pre-crisis capital, which means comparisons across the crisis understate the improvement in one sense, because pre-crisis ratios were flattered by looser definitions. Second, the honest account divides credit among several forces, supervisory and statutory pressure, market demands for thicker buffers, retained earnings, and the international framework, in proportions the record does not measure. What is settled is the measurement itself, not the attribution of it.

Capital and the too-big-to-fail question

The capital build connects directly to the too-big-to-fail debate, and the connection runs through creditor expectations. Before the crisis, creditors of the largest banks lent cheaply because they expected rescue, and that expectation was rational given the absence of any credible alternative. After the statute, two things changed the calculus: the thicker equity cushions meant creditors faced loss only after a much larger buffer was exhausted, and the resolution framework, whatever its untested status, gave the government a legal alternative to a bailout that had not existed before. Both changes should, in theory, have made creditors charge the largest banks something closer to their standalone risk.

Whether creditors actually updated their beliefs is where the measurement gets difficult, because expectations leave faint traces. Bond spreads, credit ratings, and funding costs all moved in the direction the statute’s defenders predicted, but they also moved with the business cycle, with monetary policy, and with the general post-crisis repricing of risk, so the statute’s share of the movement is contested. The article does not resolve this any more than it resolves the subsidy debate in its dedicated section; it notes the theoretical channel, records that market measures moved in the expected direction, and marks the attribution as disputed. What is settled is the precondition: whatever creditors believed, the equity standing ahead of them roughly doubled on the measured ratio, which is a fact about balance sheets rather than about beliefs.

There is a final irony in the capital story that belongs in layer one rather than in the commentary around it. The firms whose capital rose the most were the survivors, and some of them survived by absorbing firms that did not. The crisis-acquisition artifact flagged earlier means the measured system got safer partly by getting more concentrated, which trades one risk for another. A system of fewer, better-capitalized giants is more resilient to the failure of any single mid-sized firm and more exposed to the failure of a giant, which is precisely why the untested resolution regime matters so much. Layer one and layer five are not separate stories; they are the two halves of a single bargain, and the bargain’s second half has never been tested.

The stress test regime as an enforcement mechanism

The numbers above did not rise by themselves, and the mechanism that raised them deserves its own treatment, because it became one of the most consequential innovations in American bank supervision. The statute required large bank holding companies to undergo supervisory stress tests and to run their own company tests, and the Federal Reserve built the Comprehensive Capital Analysis and Review into the annual supervisory cycle. Each year, firms projected their losses, revenues, and capital ratios under hypothetical adverse scenarios spanning roughly nine quarters, and supervisors used the results to approve or restrict dividends and share repurchases. A firm whose projected capital fell below required minimums under stress faced limits on the cash it could return to shareholders, which turned the test into a binding constraint rather than an academic exercise.

The design had a deliberate forward-looking character that distinguished it from traditional supervision. Old-style examinations looked backward at loans already made and asked whether reserves were adequate. Stress tests looked forward at losses not yet incurred and asked whether capital would survive them. That shift in temporal orientation was the regime’s intellectual contribution, and it is why the tests constrained behavior even in years when the scenarios seemed mild: the constraint operated through the planning process itself, forcing firms to maintain the capital planning infrastructure the tests required.

The regime also had limits that the 2023 failures exposed. The scenarios were built around credit losses and macroeconomic downturns, the risks that had destroyed capital in 2008, and they were less attuned to the risks that destroyed Silicon Valley Bank and Signature Bank: concentrated uninsured deposits, unrealized losses on securities portfolios as interest rates rose, and the speed of a digitally accelerated depositor run. A stress test measures resilience against the shocks its designers imagined, and the 2023 episode was a reminder that the next shock rarely repeats the last one’s shape. This is not an indictment of the regime so much as a statement of its nature. The tests made the system more resilient against the risks they modeled, which is a genuine achievement, and they did not make it resilient against every risk, which was never a realistic standard.

What the new liquidity requirements demanded

The liquidity framework attacked the two distinct ways that funding stress kills a bank. The short-horizon requirement forced large firms to hold a buffer of high-quality liquid assets, cash, central bank reserves, and the highest-quality securities, sized against net cash outflows over thirty days of prescribed stress. The logic was survival time: in a panic, the question is not whether the bank is solvent in the long run but whether it can meet the coming month’s obligations without selling assets at collapsing prices. The longer-horizon funding metric addressed the structural mismatch directly, requiring that long-dated assets be funded with liabilities of matching stability rather than with overnight borrowing that could vanish. Together the two metrics retired the business model that had made the pre-crisis system fragile: borrow short, lend long, and assume the short borrowing will always be there.

The introduction counts as a finding because the pre-crisis regime had nothing comparable in force. Several of the institutions that required rescue in 2008 were solvent by the capital measures of the day and nevertheless died of illiquidity, funding long-dated assets with short-dated wholesale borrowing that lenders refused to roll over. The regulatory framework had treated liquidity as a matter for internal risk management rather than for binding rules, and internal risk management assumed funding markets would remain open. The crisis falsified that assumption in days. Against that background, quantitative liquidity requirements were a regime change rather than a refinement: supervisors could point to a number and require that it be maintained, rather than asking whether a funding plan seemed prudent. The requirements phased in across the decade after the statute, which is why this layer reports the direction and the mechanism rather than a ratio series for the early years.

Why the starting point matters

The choice to measure the capital build from the end of 2008 rather than from before the crisis is not a technical footnote; it determines what the numbers mean, and the verifier’s record is explicit that no 2007-to-2013 ratio comparison should be used. The reason is attribution. By the end of 2008, the crisis had already forced emergency capital raising, government injections, and the first round of private rebuilding, so a measurement starting there captures the deliberate post-crisis reconstruction rather than mixing the collapse with the repair. A comparison reaching back to 2007 would blend the destruction of capital during the panic with its rebuilding afterward, producing a number that no single source documents and that would confuse two different phenomena.

The starting point also affects how the crisis-acquisition artifact should be read. The Center for Financial Stability figure dated January 13, 2014, showing the four largest banks adding 306 billion dollars in tier 1 capital since December 31, 2007, uses the pre-crisis starting point and therefore includes the capital that arrived through acquisitions of failing firms. When a large bank absorbs a failing competitor, its balance sheet grows and its capital grows with it, but the growth reflects consolidation rather than the organic strengthening of a constant set of firms. Flagging the artifact does not invalidate the figure; it clarifies what the figure measures, which is the capitalization of the surviving giants after the shakeout rather than the recapitalization of the pre-crisis system. Both measurements are legitimate. They answer different questions, and the article keeps them separate so the reader knows which question each answers.

A final note on periods: the Federal Reserve’s March 2020 dated later development, reporting 4.9 percent in the first quarter of 2009 against 12.2 percent in the fourth quarter of 2019, uses yet another starting point, the trough quarter rather than year-end. The three series together, end-2008 to fourth-quarter 2012, December 2007 to January 2014 for the big four, and first-quarter 2009 to fourth-quarter 2019, all point the same direction despite their different windows. Convergence across sources with different periods and different firm coverage is what makes the capital finding settled rather than merely reported.

Layer two: industry structure

The second layer concerns the shape of the banking industry itself: how many banks existed, and whether new ones kept being born. Here the trend is dramatic and the cause is the subject of a genuine research dispute, which makes this layer the best illustration of the article’s method. The numbers are not in doubt. The explanation is.

The starting point is a decline that began long before 2010. At the end of 1984, the United States had 14,477 insured commercial banks, and by March 1991 the count had fallen to 12,246, with the Federal Deposit Insurance Corporation’s Quarterly Banking Profile for the first quarter of 1991 describing the shrinkage as the twenty-fifth consecutive quarter of decline. The forces behind that long contraction, interstate branching deregulation, mergers, technology, and the consolidation of back-office functions, were already well established before Dodd-Frank existed. Any honest account of post-2010 consolidation has to begin by acknowledging that the statute arrived in the middle of a decades-long trend, not at its beginning.

The trend continued after the statute’s passage. The FDIC’s Quarterly Banking Profile Graph Book reported 7,658 banks in total at the end of 2010 and 6,812 at the end of 2013. That is a decline of 846 institutions, about 11 percent, in three years, and it is documented by the deposit insurer’s own statistical series. But the direction of the change tells us nothing by itself about the statute’s role in it, because the count had been falling at a comparable pace for years. Continuation, not a break, is the right word for the total count. The more interesting question is what happened to the flow of new charters.

Did new bank formation really stop?

De novo charters fell from 147 in 2006 and 140 in 2007 to 72 in 2008, 38 in 2009, 7 in 2010, 3 in 2011, zero in 2012, and 1 in 2013, according to McCord and Prescott in the Richmond Fed’s Economic Quarterly for the first quarter of 2014, a dated later development.

The sequence above is the fact that launched a thousand arguments. The fall from more than 100 new banks a year to fewer than one a year is real, and it happened on Dodd-Frank’s watch. The dispute is about why, and the two leading explanations deserve equal billing because each has serious research behind it.

The first explanation comes from economists at the central bank. Adams and Gramlich, in a Finance and Economics Discussion Series paper from the Federal Reserve Board dated December 2014 and revised in July 2016, a dated later development, concluded that the decline in new charters reflected weak economic conditions far more than regulatory change. Their estimate was that at least 75 percent of the decline in entry would have occurred even without any change in regulation, driven by the low interest rate environment, depressed bank profitability, and weak expected returns for a new institution trying to raise capital and attract deposits. On this account, the de novo drought was mostly a symptom of a miserable environment for starting a bank, and the statute’s compliance costs were a secondary factor layered on top of economics that were already hostile.

The second explanation emphasizes the regulatory burden. Bordo and Duca, in National Bureau of Economic Research Working Paper 24501 dated April 2018, a dated later development, found that the share of small loans in commercial and industrial lending fell by 9 percentage points from 2010, an effect twice as large at small banks, and attributed nearly all of that decline to the post-2010 regulatory regime change after controlling for the business cycle and bank size. On this account, the compliance architecture of the new regime, capital rules, examination intensity, and the fixed costs of new consumer and reporting requirements, materially shifted the economics of small-scale lending and discouraged the formation of the small institutions that had historically done it.

The honest presentation gives both, because the methods genuinely differ and neither can be dismissed. The central-bank economists use entry models that load heavily on profitability and interest rates; the regulatory-burden researchers use lending-share regressions that load heavily on the regime change. Each approach controls for what the other emphasizes, and each finds what its design is best equipped to find. What can be said with confidence is narrower than either camp’s headline: the formation of new banks nearly stopped after the crisis, the halt coincided with both terrible economics for new banks and a major new regulatory regime, and the apportionment of blame between those two forces is contested. Readers who want a single number for how much of the halt was caused by Dodd-Frank will not find one in the serious literature, because the serious literature does not offer one.

A related error needs correction because it is common in public discussion: attributing all industry consolidation to the statute. The FDIC series shows the count of banks falling for decades before 2010, driven by mergers that the statute neither ordered nor prevented. Dodd-Frank added compliance costs that weigh more heavily on small institutions, and it is reasonable to argue that those costs accelerated the exit of marginal banks at the margin. But the consolidation trend is older than the law, and presenting the law as its author is a misreading of the data.

The long consolidation in context

The decline in the number of banks is one of the longest-running trends in American finance, and placing the post-2010 years inside it changes how the statute’s role should be read. The United States entered the 1980s with a fragmented banking system shaped by Depression-era restrictions on branching, and the subsequent decades dismantled those restrictions piece by piece. Interstate branching deregulation allowed banks to operate across state lines, which made regional and national franchises possible and made thousands of small single-market charters redundant. The savings and loan crisis of the late 1980s and early 1990s then failed or forced the merger of hundreds of thrifts, with the federal cleanup machinery resolving institutions at a pace that permanently reduced the count. Technology added a third force: ATMs, electronic payments, and later online banking raised the efficient scale of retail banking, rewarding institutions that could spread technology investments over large customer bases.

Against that background, the FDIC’s figures trace a single continuous story. The 14,477 insured commercial banks at the end of 1984 became 12,246 by March 1991, with the Quarterly Banking Profile noting the twenty-fifth consecutive quarter of shrinkage, and the total kept falling through the 1990s and 2000s to 7,658 at the end of 2010 and 6,812 at the end of 2013. The statute arrived when the count was already less than half its 1984 level, and the post-2010 decline continued the slope rather than steepening it dramatically. A reader who encounters claims that Dodd-Frank halved the number of banks should check the dates: most of the halving happened before the law existed.

That does not exonerate the statute entirely, and the article does not claim otherwise. Consolidation has causes at the margin as well as in the trend, and a new layer of fixed compliance costs plausibly pushed some marginal institutions toward merger a few years earlier than they would otherwise have gone. The honest statement is proportional: the trend is decades old and driven by forces larger than any single statute, while the statute likely added a modest accelerant at the margin, concentrated among the smallest institutions for which compliance is a fixed cost. The de novo halt is the sharper and more statute-adjacent fact, because the flow of new charters is more sensitive to entry economics than the stock of existing banks, and it is there that the contested explanations of layer two properly belong.

Layer three: the community bank burden

The third layer is where the article’s neutrality discipline is tested most directly, because the question of whether Dodd-Frank hurt community banks has generated survey evidence and econometric evidence that point in genuinely different directions. The brief for this article requires showing the range rather than picking a number, and the verifier’s memo requires that every estimate carry its method. This section does both.

Start with the survey evidence, because it is the most quoted and the most easily misused. Peirce, Robinson, and Stratmann, in Mercatus Center Working Paper 14-05 dated February 2014, a dated later development published just after this article’s window, surveyed 200 of 500 banks with under 10 billion dollars in assets. About 83 percent of respondents reported compliance-cost increases of at least 5 percent, and the median bank’s compliance staff rose from one to two full-time equivalents. Those are striking figures, and they have been quoted as though they settled the question. They do not settle it, and the reason comes from the Government Accountability Office. GAO-18-213 warned that the Mercatus survey rested on a nonrandom sample with a low response rate and self-reported cost figures, which means the banks most motivated to report high costs were the most likely to respond. The survey measures the experience of its respondents, not the population of small banks, and any use of its numbers without the GAO caveats is a misuse of the study.

The econometric evidence tells a more measured but still serious story. The Federal Reserve Bank of St. Louis surveyed 469 banks in 2014 and found that compliance costs absorbed 8.7 percent of noninterest expense at banks under 100 million dollars in assets, compared with 2.9 percent at banks between 1 and 10 billion dollars, as reported in the Regional Economist in July 2016, a dated later development. A follow-up survey of 1,091 banks covering 2015 through 2017 found an average compliance share of 7.2 percent of noninterest expense, with 9.8 percent at banks under 100 million dollars versus 5.3 percent at banks between 1 and 10 billion dollars, in research by Dahl, Fuchs, Meyer, and Neely dated April 2018 and summarized through CRS R46648, again a dated later development. The pattern across both St. Louis surveys is consistent: compliance costs are regressive, falling most heavily as a share of expenses on the smallest institutions, which is exactly what one would expect from fixed regulatory costs spread over a small revenue base.

How heavy was the compliance burden on small banks?

A Mercatus survey dated February 2014 found about 83 percent of responding banks under 10 billion dollars reporting compliance-cost increases of at least 5 percent, while St. Louis Fed surveys dated 2016 and 2018 found compliance absorbing 2.9 to 9.8 percent of noninterest expense depending on size.

The Government Accountability Office’s own assessment added a further data point with a different emphasis. GAO-16-169, dated December 30, 2015, a dated later development, concluded that the act had produced moderate to minimal initial reductions in the availability of credit, and that regulatory data had not confirmed a negative impact on mortgage lending. The GAO’s language is worth quoting precisely because it is carefully bounded: moderate to minimal, initial, and not confirmed by the regulatory data. That is not a finding that small banks were unaffected; it is a finding that the aggregate lending data available at that point did not show the damage that the most alarmed surveys implied.

Pulling the three strands together, the community bank picture is genuinely contested in the way the brief requires the article to show. The survey strand, led by the Mercatus work, finds large reported cost increases but rests on a respondent pool the GAO judged methodologically weak. The econometric strand, led by the St. Louis Fed surveys, finds a consistent regressive pattern, with the smallest banks devoting the largest share of expenses to compliance, but the absolute levels are far below the most dramatic survey claims. The GAO strand finds limited measurable effects on credit availability in the aggregate data. A reader who quotes only the Mercatus 83 percent figure is telling the truth about the survey and a falsehood about the literature. A reader who quotes only the GAO’s moderate to minimal language is telling the truth about the aggregate data and omitting the regressive burden the St. Louis numbers document. The range is the finding, and the method is the explanation for the range.

One more caution belongs in this layer. Precise dollar-per-bank compliance-cost totals circulate in advocacy literature, and the verifier’s memo directs that they be dropped. They are dropped here. The estimates that survive attribution are shares of expenses and reported percentage increases, each with its source, its period, and its methodological limits attached.

The examination channel

The surveys measure costs, but they measure them as experienced, and part of the experience was a change in how supervision felt at small institutions. Bankers who responded to the Mercatus survey and the St. Louis Fed surveys consistently described examinations that had become longer, more document-intensive, and more focused on compliance management systems as distinct from credit quality. An examination that once centered on the loan portfolio now also probed the bank’s procedures for mortgage disclosures, its vendor management, its complaint handling, and its Bank Secrecy Act controls, each with its own workpapers and each generating Matters Requiring Attention that the board had to answer in writing. For a bank with a handful of officers, the fixed time cost of hosting examiners and remediating findings was itself a burden, separate from the dollars spent on compliance staff.

This channel matters for the evaluation because it explains part of the regressivity the surveys document. A compliance management system has a minimum viable size: the policies, the audits, and the board reporting exist whether the bank has 50 million dollars in assets or 5 billion. The St. Louis Fed’s finding that compliance absorbed 8.7 percent of noninterest expense at banks under 100 million dollars in 2014, against 2.9 percent at banks of 1 to 10 billion dollars, is consistent with a fixed-cost story in which examination intensity is one of the fixed costs. The article presents this as a mechanism consistent with the survey evidence rather than as a separately measured finding, because the surveys document the costs and the examination channel is the bankers’ own explanation for part of them.

Why the methods disagree

The three strands of community bank evidence disagree for reasons that are themselves informative, and understanding the reasons is more useful than averaging the numbers. Surveys ask bankers what compliance costs them, and bankers who feel burdened answer surveys about burden. The Mercatus survey of 200 banks, drawn from 500 solicited, is a textbook case of the resulting selection problem: the 300 banks that did not respond may have had lower costs, less frustration, or simply less time, and there is no way to know from the responses alone. Self-reporting adds a second distortion, because compliance costs are partly a matter of accounting judgment about which staff hours count as compliance, and a banker who opposes the regulatory regime may classify generously. None of this means the respondents were dishonest; it means the instrument measures perception and experience among the motivated, which is a real thing but not a population estimate. The GAO’s caveats in GAO-18-213, nonrandom sample, low response rate, self-reported figures, are the formal statement of these limits.

The St. Louis Fed surveys used larger samples and a more structured instrument, which is why their numbers carry more weight as population estimates. Even they, however, measure costs as a share of noninterest expense, a denominator that itself moves with the business cycle and with each bank’s business model. A bank with high noninterest expense from an expensive branch network will show a lower compliance share than an equally compliant bank with lean operations, even if both spend the same dollars on compliance. The 8.7 percent figure for banks under 100 million dollars in the 2014 survey and the 9.8 percent figure for the same size group in the 2015 through 2017 research are therefore best read as indicators of regressivity, the consistent finding that the smallest banks devote the largest share, rather than as precise national totals.

The GAO’s aggregate approach asks a different question entirely: not what compliance costs, but whether lending contracted. GAO-16-169, dated December 30, 2015, found moderate to minimal initial reductions in credit availability and no confirmed negative impact on mortgage lending in the regulatory data. That finding can coexist with genuine burden at individual banks, because burdened banks can maintain lending by accepting lower margins, cutting other expenses, or merging into larger institutions that absorb the fixed costs. The aggregate data measures the system’s output; the surveys measure the producers’ pain. Both can be true at once, and the article’s contested rating reflects that both are.

There is a final methodological point that applies to all three strands, which is the absence of a pre-registered baseline. Nobody measured community bank compliance costs with the same instruments before 2010, so the surveys document levels and self-reported changes rather than clean before-and-after comparisons. The 83 percent figure from Mercatus is a reported increase, not a measured one, and the St. Louis figures are snapshots, not trends. A reader who wants the definitive time series of small-bank compliance costs will not find it in the literature, because the literature never built it. The range is the finding, and the reasons for the range are now on the table.

Layer four: mortgage credit

The fourth layer concerns the market where the statute’s consumer protections met borrowers most directly: mortgages. The ability-to-repay and qualified mortgage framework, the ATR/QM rules, was the most consequential change to mortgage underwriting standards in the post-crisis period, and its effects are documented by the regulator that wrote it. Here again the direction of the finding is clearer than the attribution.

The rule itself was finalized on January 30, 2013 and took effect on January 10, 2014, dates that sit at the very edge of this article’s publication window. Its core requirements were straightforward. Lenders had to verify a borrower’s ability to repay, and loans that met the qualified mortgage definition, including a 43 percent cap on the debt-to-income ratio and limits on risky features, received legal protections: a safe harbor from ability-to-repay liability for first-lien loans priced within 1.5 percentage points of the average prime offer rate, and a rebuttable presumption of compliance for loans priced above that spread. The consumer side of the statute’s record, including the bureau that administered these rules, is described in the account of the Consumer Financial Protection Bureau’s creation and powers.

Did mortgage credit tighten after the qualified mortgage rule?

The CFPB’s assessment of the ATR/QM rule, dated January 2019 and summarized by CRS in February 2021, found approval rates for non-qualified high debt-to-income loans fell across all credit tiers and income groups, with refinancing borrowers above 43 percent debt-to-income facing reduced access.

The assessment’s findings deserve careful reading, because they are frequently summarized in ways that go beyond what the document supports. What the CFPB found was a decline in approval rates for the specific category of loans the rule disfavored: non-QM loans to borrowers with high debt-to-income ratios. That decline appeared across credit tiers and income groups, which suggests the rule’s binding constraint operated broadly rather than only at the margins. Refinancing borrowers above the 43 percent threshold faced reduced access, which is consistent with a rule that made the threshold a hard underwriting boundary. The article attributes these findings generically to the CFPB’s assessment rather than labeling them with any particular dataset, because the verifier’s memo requires that generic attribution and forbids presenting the findings as direct loan-level data analysis.

The contested question is how much of the broader tightening of mortgage credit after the crisis reflects the rules and how much reflects post-crisis lender risk appetite. This is the same structure as the de novo dispute in layer two, and it resists clean separation for the same reason. Lenders emerging from the crisis faced putback risk on loans sold to the government-sponsored enterprises, litigation exposure, and a general aversion to the products and borrowers associated with the bust. Those forces would have tightened underwriting even without the ATR/QM rule. At the same time, the rule’s 43 percent threshold and its liability framework gave lenders a bright line that many treated as a ceiling rather than a guideline, which amplified the tightening beyond what caution alone would have produced. The evidence supports both mechanisms operating at once, and no study in the record reviewed here cleanly decomposes the two. The honest statement is that measured credit availability for borrowers with weaker credit profiles fell, that the rules contributed, that lender caution contributed, and that the proportions are disputed.

There is a distributional point that the evidence does support. The tightening was not uniform. Borrowers with strong credit profiles and low debt-to-income ratios experienced little change in access, while borrowers with weaker profiles, precisely the borrowers the pre-crisis market had served most recklessly, faced the sharpest contraction. Whether that pattern represents the rule working as intended, protecting borrowers from unaffordable loans, or the rule overshooting, excluding creditworthy borrowers who would have repaid, is a normative judgment the evidence alone cannot deliver. The article records the pattern and leaves the judgment to the reader.

The bright line problem

The 43 percent debt-to-income threshold illustrates a general property of bright-line regulation: a boundary drawn for legal clarity becomes an economic cliff. The rule’s drafters intended the threshold as one element of a broader ability-to-repay determination, with the safe harbor protecting lenders who stayed inside it. In practice, many lenders treated 43 percent as a ceiling rather than a guideline, declining loans just above the line that a holistic underwriting judgment might have approved. The CFPB’s assessment documented the consequence: refinancing borrowers above the threshold faced reduced access, and the decline in non-qualified high debt-to-income approvals spanned credit tiers and income groups. A rule meant to sort affordable loans from unaffordable ones also sorted some repayable loans into the rejected pile, because no threshold can track individual repayment capacity perfectly.

The bright line interacted with a second force that amplified its effect. Lenders emerging from the crisis faced putback risk, the danger that the government-sponsored enterprises would force them to repurchase defaulted loans, and they responded with overlays, internal underwriting standards stricter than any rule required. An overlay is caution expressed as policy: a lender that the rule would have permitted to make a loan at 44 percent debt-to-income might still refuse it at 42 percent, because the institution’s own risk committee had decided the crisis proved the old models wrong. The assessment’s findings capture the combined effect of the rule and the overlays, and the literature reviewed here does not separate them. A reader who attributes every rejected application above the threshold to the statute is ignoring the overlays; a reader who attributes them all to caution is ignoring the threshold’s gravitational pull on industry practice.

The distributional consequences deserve emphasis because they cut against simple narratives in both directions. Borrowers with strong credit and low debt-to-income ratios experienced a mortgage market that, by most measures, functioned well after the rules took effect: rates were low, underwriting was transparent, and access was broad. Borrowers with weaker profiles faced a market that had essentially two doors, the qualified mortgage door with its protections and the non-qualified door with its higher costs and scarcer supply, and the second door narrowed. Defenders of the framework argue this is the rule working as intended, because the pre-crisis market’s service to weak-profile borrowers was predatory as often as it was generous, and excluding marginal borrowers from unaffordable loans is a benefit, not a cost. Critics argue the framework overshot, excluding creditworthy borrowers whose repayment capacity did not fit the threshold’s arithmetic. The evidence establishes the narrowing; it does not establish whether the narrowing was wise.

The resolution question that never got answered

The untested core: Dodd-Frank’s central bargain was that a large firm could fail without a bailout and without contagion, and that mechanism has never been used, which means the statute’s most important claim about itself remains an assertion rather than a finding.

The fifth layer is the reason this article exists in its present form, and it is the layer where the statute’s central bargain meets the hardest evidentiary discipline. Title II of Dodd-Frank created the orderly liquidation authority, a mechanism designed to let a large financial firm fail without a taxpayer bailout and without contagion spreading through the financial system. That was the promise that distinguished the statute from a mere capital-raising exercise: not just safer banks, but a credible way to let an unsafe one die. The mechanism has never been used. That single fact dominates everything else in this layer, because it means the statute’s most important claim about itself remains an assertion rather than a finding.

The Congressional Research Service reports that the orderly liquidation authority has never been used, and the Federal Deposit Insurance Corporation’s Office of Inspector General stated in December 2023, a dated later development, that a Title II resolution has not yet occurred. Those are dated official statements, and the article presents them as such rather than as timeless facts. The consequence is straightforward: more than a decade after the statute’s passage, there is no empirical record of how the authority would perform under the conditions it was built for. Models, simulations, and living wills exist, but none of them is a test. A resolution regime that has never resolved anything is a theory with a statute attached.

Has the resolution authority ever been used?

The Congressional Research Service states that the orderly liquidation authority has never been used, and the FDIC Office of Inspector General reported in December 2023, a dated later development, that no Title II resolution has yet occurred. The three large bank failures of 2023 were handled through deposit insurance receiverships, not through Title II.

The failures of 2023 are the closest the system has come to a test, and they are instructive precisely because they bypassed the mechanism. Silicon Valley Bank was closed by the California Department of Financial Protection and Innovation on March 10, 2023. Signature Bank was closed by the New York State Department of Financial Services on March 12, 2023. First Republic Bank was closed on May 1, 2023, not in March, a dating correction the record requires. Each held between 100 billion and 250 billion dollars in assets, making them the second, third, and fourth largest bank failures in United States history. In all three cases the FDIC was appointed receiver under the Federal Deposit Insurance Act, the traditional deposit insurance receivership process, not under Title II of Dodd-Frank. The resolution authority created for systemically important failures was not invoked for the largest bank failures since the crisis.

What was invoked, for two of the three, was the systemic risk exception. On March 12, 2023, the systemic risk determination was made for Silicon Valley Bank and Signature Bank, allowing the FDIC to guarantee uninsured deposits and contain the panic. First Republic was resolved without emergency powers, through a sale announced in FDIC press release PR-23-019 dated May 1, 2023. The distinction matters for the evaluation. The systemic risk exception is a deposit insurance tool with roots in the 1991 legislation, not the Title II machinery. Its use in March 2023 demonstrated that the authorities would act aggressively to stop contagion, but it demonstrated nothing about whether orderly liquidation authority would have worked, because that authority was never triggered.

The 2023 episode must be described through dated official reviews rather than commentary, and the record provides them. The Federal Reserve’s review of the Silicon Valley Bank failure, led by Vice Chair for Supervision Michael Barr and dated April 28, 2023, a dated later development, examined supervisory and regulatory failures. The FDIC’s own reviews and the Congressional Research Service’s account, CRS R47658, a dated later development, provide the official chronology. What these reviews establish is the sequence of supervisory misses and the mechanics of the response. What they do not establish, and what no review can establish, is how Title II would have performed, because it was not used.

One further question from 2023 requires the contested label. The 2018 tailoring legislation, described in the account of the 2018 Dodd-Frank rollback, changed which firms faced the law’s enhanced prudential requirements, and the three failed banks sat in the size range affected by those changes. Whether the tailoring contributed to the failures is contested: some assessments argue that reduced supervisory intensity left the firms’ interest rate and liquidity risks under-monitored, while others argue that the failures reflected classic asset-liability mismanagement that the pre-2018 regime would not necessarily have prevented. The article does not weight these assessments, because the evidence does not support a weighting. Contested is the rating, and the rating is the point.

Step back and the shape of the fifth layer is stark. The statute’s central bargain was that a large firm could fail without a bailout and without contagion. The mechanism built to honor that bargain has never been used. The failures that did occur were handled through older tools plus an emergency determination. The most important question went effectively untested for more than a decade, and the article refuses to deliver a verdict the evidence does not support.

What a real test would require

It is worth specifying what would actually count as a test of the resolution regime, because the word is used loosely in public debate and the article’s untested rating depends on a strict meaning. A genuine test would require the failure, or the imminent failure, of a financial company large and interconnected enough to trigger the Title II determination process: the recommendations from the Federal Reserve and the FDIC, the Treasury Secretary’s determination, and the appointment of the FDIC as receiver under Title II rather than under the deposit insurance law. The receivership would then have to wind the firm down, allocate losses to shareholders and creditors according to the statute’s priority scheme, and do so without the contagion the mechanism was built to prevent and without taxpayer losses beyond the statute’s industry-funded backstop. Only that sequence would generate evidence about whether the authority works.

None of the post-2010 failures has produced that sequence. The failures that did occur were commercial banks resolved under the deposit insurance law, a process the FDIC had operated for decades before Dodd-Frank and one whose mechanics were well understood. The systemic risk exception of March 2023, invoked for Silicon Valley Bank and Signature Bank, demonstrated that the government would protect uninsured depositors to stop a panic, but it answered a different question from the one Title II poses. Title II asks whether a giant can be dismantled. The 2023 episode asked whether depositors could be reassured. The first question remains open, and the article’s discipline requires saying so even though the distinction is unsatisfying to readers who want a verdict.

There is a related point about living wills, the resolution plans the statute required large firms to file with regulators. The plans were meant to map how each firm could be resolved under bankruptcy, giving supervisors a blueprint in advance of failure. Their value as evidence is limited in the same way as the rest of the resolution architecture: a plan is a document, and documents do not fail. Supervisors reviewed the plans, demanded revisions, and in some cycles found particular submissions not credible, which demonstrated that the review process had teeth. But credibility judgments about hypothetical resolutions are not resolutions, and the article rates the regime on outcomes rather than on paperwork about outcomes.

How the wind-down was supposed to be funded

A wind-down needs liquidity, and the statute’s designers knew that private lenders will not fund a firm in receivership. Title II created an Orderly Liquidation Fund, administered by the FDIC, that could borrow from the Treasury to supply operating liquidity to a bridge institution while the failed firm’s assets were sold. The borrowing would be repaid from the failed firm’s recoveries and, if those proved insufficient, from assessments levied on the largest financial firms. The statute’s drafters insisted this was not a bailout mechanism: the law barred the fund from rescuing the failed firm or benefiting its shareholders and creditors beyond the statutory priority scheme, and any shortfall would be recovered from the industry rather than the taxpayer.

Whether that distinction holds in practice is one of the questions the authority’s non-use leaves unanswered. Temporary public liquidity with an industry backstop may be meaningfully different from a bailout when the alternative is contagion, or it may be a bailout by another name, and no test exists to decide between the readings. The design also concentrated extraordinary discretion in the executive branch: invoking the authority requires findings of systemic risk, consultation among regulators, and presidential involvement, which means the decision to use it would be one of the most politically fraught calls a Treasury Secretary could make. A mechanism whose activation is uncertain shapes behavior in its shadow, and that shadow effect is itself part of the regime, even if it cannot be measured.

The too-big-to-fail subsidy debate

Running alongside the five layers is a debate the article must address without pretending to resolve: whether the largest banks enjoyed a funding advantage from the market’s belief that they would be rescued, and whether Dodd-Frank reduced it. Before the crisis, the largest institutions could borrow more cheaply than their standalone creditworthiness justified, because creditors believed the government would not let them fail. That belief was a subsidy, paid by taxpayers in the form of risk they did not choose to bear, and one of the statute’s aims was to eliminate it by making failure credible.

Measuring the subsidy is notoriously difficult, because it requires estimating what the largest banks would have paid for funding without the implicit guarantee, a counterfactual no market directly reveals. Researchers have tried, using credit rating uplifts, bond spread comparisons, and options-based models, and their estimates vary widely depending on method and period. The direction of the finding after 2010 is that the measured advantage shrank as capital rose and the resolution framework took shape, which is consistent with the statute having made the largest firms safer and their failure more imaginable. But the estimates remain sensitive to their assumptions, and the verifier’s record does not contain the study that would let this article put a number on the change. The honest treatment is therefore qualitative: the subsidy debate is contested, the direction of travel favored the statute’s aims, and the magnitude is disputed.

The 2023 failures added a wrinkle to the debate that neither side had fully anticipated. The systemic risk exception’s use to protect uninsured depositors at Silicon Valley Bank and Signature Bank was, on one reading, exactly the kind of intervention the statute was meant to make unnecessary: the government stepping in to prevent contagion from failing banks. On another reading, it was proof that the deposit insurance framework, not Title II, remains the operative backstop for commercial bank failures, and that the too-big-to-fail problem for the largest firms is a separate question the 2023 episode did not address, since none of the failed banks was among the very largest. Both readings are defended in the official reviews and the commentary around them, and the article records the disagreement rather than choosing between the readings.

Supervision, living wills, and the plumbing of reform

Beyond the five measured layers sits a quieter set of changes that resist quantification but shaped how the statute worked in practice. The law reorganized supervision of the largest firms, concentrating enhanced prudential oversight at the Federal Reserve and requiring regular stress tests, resolution planning, and risk committee structures that had no pre-crisis equivalent. For the firms subject to these requirements, supervision became a continuous process rather than a periodic examination: capital plans were reviewed annually, liquidity positions were monitored against the new requirements, and boards were expected to demonstrate that risk management reached the holding company level rather than stopping at the bank subsidiary.

The resolution plans, known as living wills, deserve a final note as the most paperwork-like element of a statute critics call paperwork. Each large firm had to describe how it could be resolved under bankruptcy without systemic disruption, and supervisors graded the submissions, demanded improvements, and in some review cycles publicly found particular plans not credible. The process forced firms to simplify legal structures, pre-position loss-absorbing resources, and think through failure mechanics in a way none had done before the crisis. Whether that preparation would survive contact with a real failure is the untested question of layer five, and the article does not upgrade the rating on the strength of the preparation. But dismissing the planning as meaningless confuses the absence of a test with the absence of an effect: the exercise changed firm behavior in observable ways, even if the ultimate question remains open.

The plumbing metaphor is useful for the whole statute. Much of what Dodd-Frank did was infrastructural: reporting systems, data collection, interagency councils, examination procedures. Infrastructure is invisible when it works and blamed when it fails, and its effects show up in the five layers only indirectly. The article’s evidence-first method necessarily underweights these changes, because they are inputs rather than outcomes, and the reader should understand that limitation. A full history of the statute would give the plumbing its own volume. This article measures what the plumbing produced.

The small-business lending channel

One of the most policy-relevant findings in the contested layers concerns small-business lending, because small firms depend disproportionately on bank credit and cannot easily substitute bond markets or commercial paper. Bordo and Duca’s working paper for the National Bureau of Economic Research, numbered 24501 and dated April 2018, a dated later development, examined the share of small loans in commercial and industrial lending and found it had fallen by 9 percentage points from 2010, with the decline twice as large at small banks. After controlling for the business cycle and for bank size, the authors attributed nearly all of the decline to the post-2010 regulatory regime change. On its face, this is the strongest quantitative statement in the literature for the proposition that the statute materially shifted credit away from small borrowers.

The mechanism the authors propose is the fixed-cost story in its lending form. Small commercial loans carry underwriting and monitoring costs that do not scale down with loan size, and a regulatory regime that adds fixed compliance costs to every loan makes the smallest loans the first to become uneconomical. A bank deciding whether to maintain a small-business lending operation weighs the revenue from a portfolio of modest loans against the cost of the compliance infrastructure those loans require, and the arithmetic favors either larger loans or exit from the segment. That the effect was twice as large at small banks is consistent with the mechanism, because small banks do more small-business lending and have smaller revenue bases over which to spread the fixed costs.

The finding must be read alongside its limits, which the article states plainly. It is a single study, and its attribution of nearly all of the decline to the regime change depends on the controls fully capturing the cyclical forces that also depressed small-business borrowing after the crisis. Small firms were hit hard by the recession itself, demand for credit fell, and the slow recovery meant fewer creditworthy borrowers sought loans, all of which could reduce small-loan shares without any regulatory cause. The authors’ controls are designed to address this, and the article does not second-guess their econometrics, but the honest presentation notes that the result sits on one side of the contested ledger while Adams and Gramlich’s entry analysis sits on the other. A reader who cites the 9 percentage point figure as proof that regulation crushed small-business lending is overclaiming; a reader who dismisses it because demand also fell is under-reading a serious attempt to separate the two.

What the aggregate lending data showed

The GAO’s December 2015 assessment provides the aggregate counterweight to the survey and lending-share evidence, and its careful language repays close attention. GAO-16-169 concluded that the statute had produced moderate to minimal initial reductions in the availability of credit, and that the regulatory data had not confirmed a negative impact on mortgage lending. Three words in that sentence do the heavy lifting, and each deserves unpacking.

Moderate to minimal is a deliberately bounded characterization. It says the measurable contraction in credit availability attributable to the statute was small in the aggregate data, not zero and not large. The GAO was not claiming that no borrower was affected; the mortgage layer of this article documents affected borrowers in detail. It was claiming that when the lending of the system as a whole was measured, the statute’s footprint was modest. That is compatible with real pain at the margin, because aggregate series average the unaffected majority with the affected minority.

Initial is a time stamp. The assessment was dated December 30, 2015, five years after the statute’s passage but early in the phase-in of several major rules, including the mortgage rules that took effect in January 2014. An initial finding is not a final one, and the article treats it as a dated observation rather than a permanent verdict. Later research, including the CFPB’s 2019 assessment of the mortgage rules and the 2018 compliance cost studies, added texture the GAO could not have had, which is why the article’s contested ratings draw on the full timeline rather than stopping at 2015.

Not confirmed is an evidentiary standard, not a finding of absence. The GAO was saying that the regulatory lending data available to it did not show the negative impact on mortgage lending that the most alarmed predictions had expected. Regulatory data measures what was reported through supervisory channels, and it can miss effects that operate through channels the data does not capture, such as borrowers who never applied because they assumed rejection. The article therefore reads the GAO finding as what it is: evidence that the aggregate mortgage market continued to function, not evidence that every borrower was unaffected.

Taken together, the aggregate data and the distributional evidence tell a coherent story rather than a contradictory one. The system lent, the averages held, and the margins tightened for identifiable groups: small banks facing regressive costs, new charters facing hostile entry economics, weaker-profile borrowers facing the 43 percent boundary, and small businesses facing a lending mix that shifted against them. An evidence review that reported only the aggregates would miss the margins, and one that reported only the margins would miss the system. The five layers exist to hold both in view.

A timeline of the evidence

The sources this article relies on arrived over three decades, and laying them in chronological order clarifies which findings were available when and why the later developments matter. The earliest is the FDIC’s Quarterly Banking Profile for the first quarter of 1991, which recorded the fall from 14,477 insured commercial banks at the end of 1984 to 12,246 in March 1991 and described the twenty-fifth consecutive quarter of shrinkage. That document establishes the pre-statute trend and is the reason the article treats post-2010 consolidation as a continuation. Next comes the statute itself, Public Law 111-203, signed July 21, 2010, with its requirements phasing in across the following decade.

The evidentiary core for the early implementation period clusters in 2013 and 2014. The Federal Reserve’s March 2013 methodology and results publication gave the capital figures for the end of 2008 through the fourth quarter of 2012. The ATR/QM rule was finalized January 30, 2013 and took effect January 10, 2014. The FDIC’s Graph Book gave the bank counts for the end of 2010 through the end of 2013. The Center for Financial Stability published its big-four capital figures on January 13, 2014, and the Mercatus survey appeared in February 2014, both just after this article’s January 2014 date and therefore treated as dated later developments. McCord and Prescott’s de novo series appeared in the first quarter of 2014, also just after the window.

The second wave of evidence, the dated later developments from 2015 through 2021, refined the contested layers. The GAO’s December 2015 assessment of credit availability, the St. Louis Fed’s July 2016 survey results, Adams and Gramlich’s December 2014 paper on bank entry, Dahl and coauthors’ April 2018 compliance cost research, Bordo and Duca’s April 2018 working paper on small-business lending, the CFPB’s January 2019 assessment of the mortgage rules, the Federal Reserve’s March 2020 capital update, and the CRS summary of February 2021 each added a dated data point. None of these studies was available when the article’s window closed, and the article marks all of them accordingly, but together they are what allow the contested ratings to be drawn with confidence rather than asserted.

The final wave is the 2023 failure record: the Barr review dated April 28, 2023, the FDIC’s May 2023 materials including press release PR-23-019, the CRS account R47658, and the FDIC Office of Inspector General’s December 2023 statement on Title II. These are the dated official reviews through which the article describes the failures, and they close the evidentiary timeline with the finding that dominates layer five: the resolution authority had still never been used.

What the 2023 autopsy showed

The official reviews of the 2023 failures deserve a closer reading, because they are the most detailed post-statute examination of how large-bank distress actually unfolds, and they reveal which risks the reform framework caught and which it missed. The Federal Reserve’s review led by Vice Chair for Supervision Michael Barr, dated April 28, 2023, examined the failure of Silicon Valley Bank through both a supervisory and a regulatory lens, asking what examiners missed and what the rules failed to require. The FDIC’s reviews and the CRS account R47658, both dated later developments, provide the chronology of the three failures and the official response. Read together, the reviews tell a consistent story about a classic banking vulnerability wearing unfamiliar clothes.

The vulnerability was the combination of concentrated uninsured deposits and unrealized losses on securities portfolios as interest rates rose. The failed banks funded themselves heavily with deposits above the insurance limit, held by customer bases that were unusually concentrated and unusually quick to move, and they held large portfolios of longer-term securities whose market value fell as rates climbed. When depositors grasped the unrealized losses, the run was swift, accelerated by digital banking and instant communication in ways the pre-crisis framework had never modeled. The mechanics were old, borrow short and lend long, mismanage the interest rate risk, lose the confidence of uninsured funders. The speed was new.

The supervisory autopsy found that examiners had identified weaknesses at the failed firms and had not acted forcefully enough on them, a finding that implicates supervision rather than the statute’s capital rules. The regulatory autopsy found that the 2018 tailoring had reduced the intensity of requirements for firms in the failed banks’ size range, which is the factual predicate for the contested causal question the article labels rather than resolves. What the autopsies did not find was a capital shortfall of the 2008 variety: the failures were liquidity and confidence failures, not credit-loss failures, which is why they blindsided a supervisory apparatus built to stress-test loan books.

For the evaluation, the autopsy cuts in two directions at once, which is why the article treats 2023 as evidence for humility rather than for either camp. The failures show that the post-crisis framework did not prevent large-bank runs, which weakens the strongest version of the statute’s safety claim. They also show that the failures were contained without a systemic collapse and without taxpayer losses on the scale of 2008, which weakens the strongest version of the nothing-changed claim. And they show, above all, that the resolution regime built for exactly such moments was bypassed in favor of older tools, which is the untested finding of layer five restated in the language of events rather than of ratings.

The five-layer evidence table

The table below compresses the five layers into the artifact this article promised: each outcome with the direction of the finding, the period, the principal sources, and the rating of settled, contested, or untested. It is meant to be read alongside the sections above, not instead of them, because the ratings summarize judgments the text defends at length.

Outcome	Direction of finding	Period	Principal sources	Settled, contested, or untested
Capital and liquidity at the largest banks	Higher loss-absorbing capacity; new liquidity requirements introduced	End-2008 to Q4 2012; dated later developments to Q4 2019	Federal Reserve, March 2013 (Methodology and Results, Box 1); Center for Financial Stability, January 13, 2014; Federal Reserve, March 2020	Settled on the direction and scale of the capital build; contested on whether the new levels are sufficient
Industry structure and new bank formation	Bank count continued its long decline; new charters nearly stopped	1984 to 2013; studies dated 2014 to 2018	FDIC Quarterly Banking Profile (Q1 1991; Graph Book); McCord and Prescott, Richmond Fed Economic Quarterly, Q1 2014; Adams and Gramlich, Fed FEDS, December 2014; Bordo and Duca, NBER WP 24501, April 2018	Settled on the trend; contested on the cause of the de novo halt
Community bank compliance burden	Costs rose; magnitude and aggregate significance disputed by method	Surveys dated 2014 to 2018	Mercatus WP 14-05, February 2014 (with GAO-18-213 caveats); St. Louis Fed surveys, July 2016 and April 2018; GAO-16-169, December 2015	Contested
Mortgage credit availability	Tightened underwriting, concentrated on weaker credit profiles	Rule finalized January 2013, effective January 2014; assessment dated 2019 to 2021	CFPB ATR/QM assessment, January 2019; CRS IF11761, February 16, 2021	Settled on the tightening; contested on rules versus lender risk appetite
Resolution regime	Authority never used; 2023 failures handled under deposit insurance law	Title II since 2010; failures dated March and May 2023	CRS R45162; FDIC OIG, December 2023; Fed Barr review, April 28, 2023; CRS R47658; FDIC PR-23-019, May 1, 2023	Untested

How to use the ratings

The settled, contested, and untested ratings are the article’s load-bearing structure, and a short guide to reading them will help the reader carry the findings into arguments elsewhere. A settled rating means the direction of the finding is established across multiple sources with consistent periods. It does not mean every researcher agrees on the interpretation, and it does not mean the finding is complete. The capital build is settled on direction and scale, and contested on sufficiency, which shows how a single layer can carry both ratings at once: what happened is established, what it means for the future is not.

A contested rating means competent research reaches genuinely different conclusions, or the direction is clear while the cause is not. Contested is not a euphemism for unknown; it is a positive finding that the literature contains a real disagreement, and the article’s job in those layers is to map the disagreement rather than to vote in it. The de novo halt is contested on cause, the community bank burden is contested on magnitude, the mortgage tightening is contested on attribution, and the 2018 tailoring’s role in 2023 is contested on causal weighting. In each case the article names the studies, the methods, and the periods, so the reader can see exactly where the disagreement lives.

An untested rating means the mechanism has not operated under the conditions it was built for. Untested is the strongest form of epistemic humility in the article’s vocabulary, stronger than contested, because it says not that researchers disagree but that there is nothing yet to disagree about. Only one layer carries it, and it is the layer the statute’s defenders would most like to claim and its critics would most like to dismiss. The rating refuses both moves. A resolution authority that has never been used cannot be credited with working and cannot be charged with failing; it can only be described, dated, and watched.

Readers applying these ratings should also remember the standing caveats. Post-2013 findings are marked as dated later developments. Every number carries its source and period. No finding is extended to any current proposal. And the asymmetry of prevention, that avoided crises leave no trace, means the ratings systematically understate benefits that consist of things not happening. The article accepts that limitation rather than correcting for it, because the alternative is speculation.

Both directions of overreach

With the five layers established, the article can now address the complication the brief requires: the two directions of overreach that distort public discussion of the statute. Each direction contains a grain of truth and an inference the evidence does not support, and the fair treatment is to grant the grain and withhold the inference.

The first overreach holds that the act ended systemic risk. Its grain of truth is layer one: the capital build is real, the liquidity requirements are real, and the stress testing regime forced the largest firms to plan for losses in a way nothing before 2010 required. A banking system with twice the equity cushion is genuinely more resilient than the one that entered the crisis, and dismissing that achievement requires ignoring the best-documented facts in the record. The inference the evidence does not support is the leap from thicker cushions to the end of systemic risk. Capital does not prevent asset bubbles, it does not prevent correlated losses across firms holding similar exposures, and it does not prevent panics driven by uninsured deposit flight, as the 2023 failures demonstrated with institutions that were well capitalized on paper against credit risk but undone by interest rate risk and depositor runs. The 2023 episode is particularly damaging to the ended-risk claim, because it showed a fast-moving classic bank run occurring more than a decade after the statute’s passage, handled through emergency deposit insurance powers rather than the new resolution machinery. A reader who concludes from layer one that the system is safer is reading the evidence correctly. A reader who concludes that systemic risk has been ended is reading beyond it.

The second overreach holds that the act accomplished nothing but paperwork. Its grain of truth is the compliance burden documented in layer three: the fixed costs of the new regime fell regressively on small institutions, the paperwork was real, and the thousands of pages of rulemaking imposed genuine costs that the aggregate lending data only partly captures. The inference the evidence does not support is that paperwork was all there was. The 393-billion-dollar increase in tier 1 common capital across the 18 assessed firms from the end of 2008 to the fourth quarter of 2012, documented by the Federal Reserve in March 2013, is not paperwork. The near-total halt in de novo formation, the tightening of mortgage underwriting at the 43 percent debt-to-income boundary, and the introduction of liquidity requirements are not paperwork. They are measurable changes in the structure and behavior of American banking, whether one judges them beneficial or not. A reader who concludes that the law imposed real costs is reading the evidence correctly. A reader who concludes that costs were all it imposed is ignoring the balance sheet.

The balanced position, and the one the evidence compels, is the three-sentence summary stated at the outset. Higher capital is real and measurable. Compliance burden is real and concentrated. The resolution question is open. Both overreach camps fail because each takes one true sentence and discards the other two.

The case each camp cannot answer

Each direction of overreach has a question it cannot answer, and posing those questions is the fastest way to see why the balanced position is not a compromise but a conclusion. To the camp that holds the act ended systemic risk, the question is the spring of 2023. If systemic risk had been ended, then the second, third, and fourth largest bank failures in American history would not have occurred within weeks of each other, depositors would not have run, and the systemic risk exception would not have been invoked on March 12, 2023 to guarantee uninsured deposits. The defenders have replies available: the failures were commercial banks, not the trading giants of 2008; the system absorbed them without a broader panic; capital at the largest firms held. Those replies have force, and the article does not dismiss them. But they concede the essential point, which is that the claim being defended has retreated from the end of systemic risk to the containment of particular failures, and containment is a weaker claim than the slogan promised.

To the camp that holds the act accomplished nothing but paperwork, the question is the capital build. If the statute changed nothing, then the tier 1 common ratio of the 18 assessed firms would not have risen from 5.6 percent at the end of 2008 to 11.3 percent in the fourth quarter of 2012, aggregate tier 1 common capital would not have grown by 393 billion dollars to 792 billion dollars over that period, and the Federal Reserve’s March 2020 dated later development would not have shown 12.2 percent in the fourth quarter of 2019. The critics have replies available: some of the build reflected market pressure rather than regulation; risk weights can be gamed; capital does not prevent every kind of failure. Those replies have force too. But they concede the essential point, which is that the balance sheets of the largest banks were transformed on the statute’s watch, and transformation is more than paperwork.

Notice what the two unanswered questions have in common: each camp’s strongest evidence comes from a different layer, and each camp’s weakest point is the layer it ignores. The ended-risk camp lives in layer one and cannot explain layer five. The paperwork camp lives in layer three and cannot explain layer one. The five-layer structure exists precisely to prevent this kind of selective reading, by forcing every claim about the statute to survive contact with all five kinds of evidence. A verdict that cannot survive all five layers is not a verdict; it is a preference wearing a verdict’s clothes.

Where both camps agree

For all their disagreement, the two overreach camps share more common ground than either admits, and naming it shows how much of the debate is about emphasis rather than facts. Both camps accept that capital at the largest banks rose substantially after the crisis; they differ only on whether the rise was sufficient and whether the statute deserves the credit. Both camps accept that compliance costs fell regressively on small institutions; they differ only on whether the burden was a reasonable price for stability or a deadweight loss. Both camps accept that the orderly liquidation authority has never been used; they differ only on whether that fact indicts the statute’s design or merely reflects the good fortune of not having needed it. A debate in which the facts are largely shared and the inferences differ is a healthier debate than the slogans suggest, and the article’s contribution is to keep the shared facts visible so the inferences can be argued honestly.

The shared facts also point to the questions on which future evidence could actually move minds, which is the most productive use of an evidence review. If the resolution authority were used successfully under stress, the ended-risk camp would gain its strongest possible evidence and the paperwork camp would lose its strongest talking point. If a future crisis revealed the capital buffers to be inadequate, the positions would reverse. If continued research narrowed the contested ranges on community bank costs and de novo formation, both camps would have to update. An evidence review earns its keep not by ending arguments but by specifying what would end them, and the five layers above are that specification.

The baseline and the structural alternative

Two of the required cross-links belong to the framing of the evaluation itself: the baseline against which the act is measured, and the structural alternative its critics preferred. Each deserves a short treatment, because the choice of comparison determines what counts as success.

The baseline is the crisis legislation of 2008, surveyed in the account of the 2008 financial crisis legislation. Dodd-Frank was written as the permanent reform following the emergency interventions of the panic, and its aims only make sense against that background. The 2008 measures were emergency stabilizations: capital injections, guarantees, and liquidity facilities designed to stop a collapse already underway. Dodd-Frank was the attempt to build a system that would not need them again. Measuring the statute against the 2008 baseline means asking whether the structural vulnerabilities the emergency measures papered over, thin capital, runnable funding, unresolvable giants, were actually repaired. Layers one through four answer parts of that question with evidence. Layer five answers the unresolvable-giants part with a silence that is itself the finding.

The structural alternative is the Glass-Steagall comparison, explored in the comparison of Glass-Steagall’s repeal with Dodd-Frank. A persistent line of criticism holds that the statute regulated large complex firms instead of breaking them up or restoring the separation of commercial and investment banking, and that no amount of capital or supervision can make a too-big-to-fail institution safe. This article does not adjudicate that debate, because the evidence reviewed here cannot settle a counterfactual about a law that was never passed. What the evidence does establish is relevant to it: the consolidation trend predated Dodd-Frank by decades, the largest firms got larger partly through crisis-era acquisitions, and the resolution regime meant to handle their failure remains untested. Readers who favor the structural alternative will find in layer five their strongest argument, which is that the statute’s answer to bigness has never been demonstrated. Readers who favor the statute’s approach will find in layer one theirs, which is that the largest firms are far better capitalized than the ones that failed in 2008. The article’s contribution is to keep both of those facts on the table at once.

What the evidence cannot yet tell us

The series thesis for this article is that assessing a statute means measuring it against its own stated aims and naming what the evidence cannot yet tell us, and the closing section honors that thesis by listing the open questions explicitly rather than burying them in qualifications.

First, the resolution regime. The orderly liquidation authority has never been used, the 2023 failures bypassed it, and the FDIC Office of Inspector General confirmed as of December 2023 that no Title II resolution had yet occurred. Until the authority operates under stress, every claim about whether it ends bailouts or contains contagion is a prediction, not a finding. This is the largest known unknown in the post-crisis regulatory architecture, and no honest evidence review can shrink it.

Second, the sufficiency of capital. The build is settled; the right level is not. Whether the ratios reached by the end of the 2010s would absorb the losses of a severe crisis depends on the severity of the crisis, and the historical record offers crises of many severities. Researchers disagree, and the disagreement is legitimate because it concerns the future rather than the past.

Third, the decomposition problems. The de novo halt cannot be cleanly apportioned between economics and regulation. The mortgage tightening cannot be cleanly apportioned between the ATR/QM rules and lender risk appetite. The community bank burden cannot be reduced to a single number because the methods disagree. In each case the article has shown the range and named the methods, which is what the evidence supports, and it has declined to manufacture precision the literature does not contain.

Fourth, the long-run growth question. Whether the statute’s costs to financial intermediation outweighed its benefits to stability in aggregate economic terms is the question most often asked in public debate and the least well answered in the record reviewed here. Macroeconomic estimates of the law’s effect on growth vary widely and depend on assumptions about crises that did not happen, which makes them sensitive to their assumptions in ways the capital measurements are not. The article does not present a growth verdict, because the verified record does not contain one.

A fifth open question concerns the interaction between the statute and monetary policy, which the literature has only begun to untangle. The entire post-2010 period covered by the early evidence was a period of extraordinarily low interest rates, and low rates compress bank profitability, discourage new entry, and push investors toward riskier assets in ways that confound every layer of this review. The Adams and Gramlich finding that weak economics explained most of the de novo collapse is one instance of a general problem: the statute’s effects were measured in an environment that itself moved the outcomes. A future evaluation conducted across a full interest rate cycle would be better positioned to separate the two, and until such an evaluation exists, the contested ratings carry an asterisk that the article makes explicit here.

A sixth concerns the durability of the measured gains. Capital ratios can fall as well as rise: firms can increase distributions, supervisors can ease requirements, and later legislation can narrow the perimeter, as the 2018 tailoring did for mid-sized firms. The March 2020 dated later development showed the build continuing through the fourth quarter of 2019, but a trend is not a guarantee, and the history of financial regulation is a history of reforms eroding under pressure from the regulated. Whether the capital and liquidity gains documented in layer one survive the next cycle of deregulation pressure is a question about politics as much as about economics, and the evidence reviewed here, which ends with the documented periods, cannot answer it.

None of these open questions is extended to any current proposal. The article’s findings concern the statute as enacted and as implemented through the periods documented, and they imply nothing about legislation that came later or might come next. That boundary is part of the neutrality discipline, and it is stated here explicitly so it cannot be missed.

What would change the ratings

A rating scheme earns its keep only if the ratings can change, so the article should say what evidence would move each one. The settled rating on capital would weaken if later supervisory data showed ratios eroding toward pre-crisis levels, or if research demonstrated at scale that risk weights were systematically gamed, so that measured buffers did not correspond to genuine loss absorption. The dated later evidence through the fourth quarter of 2019 pointed the other way, which is why the rating held across two vintages, but settled is a judgment about the record as it stands, not a permanent certification.

The contested attribution of the de novo halt could narrow if chartering recovered and researchers could study the recovery’s correlates: a return of entry alongside rising interest rates and improving expected profitability would favor the economic account, while a return following specific regulatory relief would favor the regulatory account. The compliance-cost range could narrow with better measurement rather than better argument: what the record lacks is administrative data on compliance spending, collected on a consistent basis across a large representative sample of small institutions over a long window, audited rather than self-reported. The mortgage attribution dispute has a natural experiment built into the framework’s own design, because researchers can compare lending at the affected margins as later assessments accumulate. The untested rating on the resolution authority can change in only one way: use. No simulation, no living-will credibility determination, and no supervisory dialogue can substitute for an actual Title II resolution, because the questions the authority raises, whether the determination process can move fast enough, whether the funding mechanism works under stress, whether the priority scheme holds when creditors litigate, can only be answered by performance under pressure.

The brief for this article defined its success by a single test: a reader who finishes it should be able to state what measurably changed in American banking after 2010, distinguish the findings that are close to settled from the ones that remain genuinely disputed, name the research behind each, and explain why the resolution question went effectively untested for more than a decade. The closing checklist restates the answer in the order the test requires. What measurably changed: capital ratios at the largest firms roughly doubled on the Federal Reserve’s headline measure between the end of 2008 and the fourth quarter of 2012, new liquidity requirements were introduced, the bank count continued its decades-long decline from 7,658 at the end of 2010 to 6,812 at the end of 2013, new charters collapsed to near zero, mortgage underwriting tightened at the 43 percent debt-to-income boundary, and compliance costs rose regressively on small institutions. What is close to settled: the direction and scale of the capital build, the continuation of the consolidation trend, and the tightening of mortgage credit for weaker profiles. What remains genuinely disputed: the causes of the de novo halt, the magnitude of the community bank burden, the attribution of the mortgage tightening between rules and lender caution, and the causal weighting of the 2018 tailoring in the 2023 failures. The research behind each: the Federal Reserve’s March 2013 and March 2020 publications, the FDIC’s Quarterly Banking Profile series, McCord and Prescott, Adams and Gramlich, Bordo and Duca, the Mercatus and St. Louis Fed surveys, the GAO assessments, the CFPB’s ATR/QM assessment, and the official 2023 reviews. And why the resolution question went untested: because the orderly liquidation authority has never been used, the 2023 failures were handled through deposit insurance receiverships and a systemic risk exception, and a mechanism that has never operated cannot be evaluated.

Seventh, the denominator question. The capital finding rests on risk-weighted ratios, and risk weights are regulatory constructs that banks can influence through asset choices and internal models. If firms improved their measured ratios partly by migrating toward lightly weighted assets rather than by reducing economic risk, then the headline doubling overstates the genuine strengthening. Supervisors added leverage constraints to guard against exactly this, and the direction of travel was upward on both families of measures, but the literature reviewed here does not quantify how much of the risk-weighted improvement was denominator engineering. Until it does, the settled rating on the capital build carries a footnote that the article states here rather than hiding.

The companion VaultBook legislation study notebook provides a structured worksheet for readers who want to test these findings against the statute’s text and trace each layer back to the provisions that produced it.

Glossary of the reform’s vocabulary

The article uses a specialized vocabulary, and a short glossary will help readers who encounter these terms in the research cited above. Tier 1 common capital is the highest-quality equity a bank holds, common stock and retained earnings, the funds wiped out first when losses arrive. Risk-weighted assets are the denominator of the headline capital ratio, a regulatory measure that weights exposures by their presumed riskiness so that safer assets count for less. The tier 1 common ratio expresses the first as a percentage of the second, and its rise from 5.6 percent at the end of 2008 to 11.3 percent in the fourth quarter of 2012 for the 18 assessed firms, documented by the Federal Reserve in March 2013, is the article’s central settled finding.

Stress tests are the supervisory exercises that project a firm’s losses, revenues, and capital under hypothetical adverse scenarios, used to constrain dividends and share repurchases at firms whose projected capital falls short. A de novo bank is a newly chartered institution, and the de novo series, 147 charters in 2006 falling to zero in 2012 and 1 in 2013 according to McCord and Prescott, is the article’s central contested fact. The ability-to-repay rule requires lenders to verify that borrowers can repay their mortgages, and the qualified mortgage definition, including the 43 percent debt-to-income threshold, gives compliant loans legal protections in the form of a safe harbor or a rebuttable presumption.

The orderly liquidation authority is the Title II mechanism for winding down a failing financial company outside bankruptcy, with losses imposed on shareholders and creditors. The systemic risk exception is the separate deposit insurance determination, invoked on March 12, 2023 for Silicon Valley Bank and Signature Bank, that allowed protection of uninsured deposits to contain contagion. Living wills are the resolution plans large firms were required to file describing how they could be resolved under bankruptcy. Enhanced prudential standards are the tougher supervision, testing, and planning requirements the statute imposed on the largest firms, whose coverage the 2018 tailoring later altered.

Two measurement terms recur throughout the evidence and deserve definitions as well. Noninterest expense is the denominator the St. Louis Fed surveys used for compliance cost shares: a bank’s operating costs excluding interest paid on deposits and borrowings, so that compliance spending can be expressed as a share of what the bank spends to run itself. Debt-to-income ratio is the borrower’s total monthly debt payments expressed as a share of monthly income, the underwriting metric at the heart of the qualified mortgage rule, whose 43 percent threshold became the bright line the article’s fourth layer examines.

Frequently Asked Questions

Q: Did Dodd-Frank make banks safer?

On the most measurable dimension, yes. The tier 1 common capital ratio for the 18 bank holding companies in the Federal Reserve’s 2013 assessment rose from 5.6 percent at the end of 2008 to 11.3 percent in the fourth quarter of 2012, with aggregate tier 1 common capital rising by 393 billion dollars to 792 billion dollars (Federal Reserve, March 2013). New liquidity requirements were also introduced during the implementation decade. Those are settled findings. What they do not establish is that systemic risk ended: capital cushions absorb losses but do not prevent asset bubbles, correlated exposures, or depositor panics, as the 2023 failures showed. Safer is the right word. Safe is not.

Q: How much did bank capital rise after Dodd-Frank?

The cleanest measurement comes from the Federal Reserve’s March 2013 stress test documentation: across the 18 assessed bank holding companies, the tier 1 common ratio rose from 5.6 percent at the end of 2008 to 11.3 percent in the fourth quarter of 2012, and aggregate tier 1 common capital increased by 393 billion dollars to 792 billion dollars. A dated later development, the Center for Financial Stability on January 13, 2014, reported the four largest banks added 306 billion dollars in tier 1 capital since December 31, 2007, a 103 percent increase, though that figure partly reflects crisis-era acquisitions. A further dated later development, the Federal Reserve in March 2020, put the ratio at 12.2 percent in the fourth quarter of 2019 versus 4.9 percent in the first quarter of 2009.

Q: Did Dodd-Frank hurt community banks?

The evidence is genuinely mixed and depends on method, so the honest answer is a range rather than a verdict. A Mercatus Center survey dated February 2014 found about 83 percent of responding banks under 10 billion dollars reporting compliance-cost increases of at least 5 percent, but the GAO (GAO-18-213) cautioned the survey used a nonrandom sample with a low response rate and self-reported figures. Federal Reserve Bank of St. Louis surveys found compliance costs absorbed 8.7 percent of noninterest expense at the smallest banks versus 2.9 percent at larger community banks in 2014, and 9.8 percent versus 5.3 percent in 2015 through 2017 research. The GAO (GAO-16-169, December 2015) found moderate to minimal initial reductions in credit availability. Burden is real and regressive; its aggregate size is contested.

Q: Did Dodd-Frank stop new banks from forming?

New bank formation nearly stopped, but the halt began before the statute and its causes are disputed. De novo charters fell from 147 in 2006 and 140 in 2007 to 72 in 2008, 38 in 2009, 7 in 2010, 3 in 2011, zero in 2012, and 1 in 2013 (McCord and Prescott, Richmond Fed Economic Quarterly, first quarter 2014). Economists at the Federal Reserve Board (Adams and Gramlich, December 2014) estimated at least 75 percent of the decline would have occurred without any regulatory change, blaming low rates and weak profitability. Other research (Bordo and Duca, NBER Working Paper 24501, April 2018) attributed nearly all of a 9 percentage point post-2010 decline in small commercial and industrial lending shares to the regulatory regime. Both explanations have serious support; the apportionment is contested.

Q: Did Dodd-Frank make mortgages harder to get?

For borrowers with weaker credit profiles, yes, though the rules share responsibility with lender caution. The ability-to-repay and qualified mortgage rule, finalized January 30, 2013 and effective January 10, 2014, set a 43 percent debt-to-income threshold for qualified mortgages with legal protections. The Consumer Financial Protection Bureau’s assessment (January 2019, summarized by CRS in February 2021) found approval rates for non-qualified high debt-to-income loans fell across all credit tiers and income groups, and refinancing borrowers above 43 percent debt-to-income faced reduced access. Post-crisis lenders were also independently risk-averse after putback and litigation exposure. The evidence confirms tightening and confirms both mechanisms; it does not cleanly separate their shares.

Q: What do stress tests under Dodd-Frank measure?

The statute’s stress test provisions require large bank holding companies and the Federal Reserve itself to project losses, revenues, and capital ratios under hypothetical adverse economic scenarios, typically spanning nine quarters. The company-run and supervisory tests estimate how much capital a firm would retain if unemployment surged, asset prices collapsed, and funding markets seized, and supervisors use the results to constrain dividends and share repurchases at firms whose projected capital falls short. The tests measure resilience against imagined shocks, not predictions of actual outcomes. Their value depends on the severity and imagination of the scenarios, which is why the 2023 failures, driven by interest rate risk and uninsured depositor flight rather than credit losses, renewed debate about what the scenarios were designed to catch.

Q: Did Dodd-Frank cost the economy growth?

The verified record reviewed here does not contain a settled answer, and this article does not manufacture one. Macroeconomic estimates of the statute’s effect on growth depend heavily on assumptions about financial crises that did not occur: if the reforms prevented or softened a future crisis, their growth benefit is large but unobservable, while their compliance costs are observable and concentrated. Research on the aggregate growth effects reaches different conclusions depending on those assumptions, and no single estimate commands consensus. What the evidence does establish is narrower: higher capital requirements can modestly raise the cost of bank intermediation, and the compliance burden fell regressively on small institutions. Whether those costs outweighed the stability benefits in growth terms remains genuinely contested.

Q: Has Dodd-Frank been tested by a real bank failure?

The resolution regime at the heart of the statute has not been tested. The orderly liquidation authority in Title II has never been used, according to the Congressional Research Service, and the FDIC Office of Inspector General confirmed in December 2023 that no Title II resolution had yet occurred. The three large failures of 2023, Silicon Valley Bank (closed March 10), Signature Bank (closed March 12), and First Republic Bank (closed May 1), were handled through traditional deposit insurance receiverships, with a systemic risk exception invoked on March 12, 2023 for the first two only. So the system has seen large failures, but the statute’s central mechanism for handling them remains untested, which is why the article rates the resolution question as untested rather than failed or vindicated.

Q: How was the orderly liquidation authority designed to work?

The orderly liquidation authority is the resolution mechanism created in Title II of Dodd-Frank for failing financial companies whose collapse could threaten financial stability. It empowers the FDIC, with recommendations from the Federal Reserve and the Treasury and a determination by the Treasury Secretary, to take over and wind down a failing firm outside of bankruptcy, with the stated aim of imposing losses on shareholders and creditors rather than taxpayers. It was the statute’s answer to the too-big-to-fail problem: a legal pathway for letting a giant fail without a bailout and without contagion. Because it has never been invoked, in any of the failures since 2010, there is no operational record of whether its tools, bridge companies, and loss-allocation rules would work under the pressure of a real collapse.

Q: Why do economists disagree about the decline of new banks?

They disagree because the two leading explanations point to different causes and use different methods. The weak-economy explanation, advanced by Federal Reserve Board economists Adams and Gramlich in December 2014, models bank entry as a function of profitability and interest rates and concludes at least 75 percent of the post-crisis decline in new charters would have happened without any regulatory change. The regulatory-burden explanation, advanced by Bordo and Duca in April 2018, examines lending shares and concludes the post-2010 regime change explains nearly all of a 9 percentage point decline in small commercial and industrial lending. Each study controls for what the other emphasizes, and each is best equipped to find what its design targets. The disagreement is methodological as well as substantive, which is why the article presents both rather than adjudicating between them.

Q: How did capital ratios at large banks change in the early years after the crisis?

They roughly doubled on the Federal Reserve’s headline measure. For the 18 bank holding companies in the 2013 capital assessment, the tier 1 common ratio rose from 5.6 percent at the end of 2008 to 11.3 percent in the fourth quarter of 2012, while aggregate tier 1 common capital grew by 393 billion dollars to 792 billion dollars (Federal Reserve, March 2013, Methodology and Results, Box 1). The starting point matters: end-2008 already reflected emergency crisis-era capital raising, so the increase measures the rebuilding during the statute’s early implementation rather than a comparison with pre-crisis peaks. A dated later development, the Federal Reserve in March 2020, extended the series to 12.2 percent in the fourth quarter of 2019 from 4.9 percent in the first quarter of 2009, confirming the upward path continued as requirements phased in.

Q: What did surveys find about compliance costs at small banks?

Two survey programs dominate the record and they tell different stories about magnitude. The Mercatus Center’s February 2014 survey of 200 banks under 10 billion dollars found about 83 percent reporting compliance-cost increases of at least 5 percent and median compliance staff doubling from one to two, but the GAO warned the sample was nonrandom with a low response rate and self-reported costs. The Federal Reserve Bank of St. Louis surveys used larger samples and found compliance absorbing 8.7 percent of noninterest expense at banks under 100 million dollars versus 2.9 percent at banks of 1 to 10 billion dollars in 2014, and 9.8 percent versus 5.3 percent across 1,091 banks surveyed for 2015 through 2017. The consistent finding across methods is regressivity: the smallest banks devote the largest share of expenses to compliance.

Q: What is a qualified mortgage under the ability-to-repay rule?

A qualified mortgage is a home loan that meets underwriting standards defined by the Consumer Financial Protection Bureau’s ability-to-repay rule, finalized January 30, 2013 and effective January 10, 2014, and therefore carries legal protections for the lender. To qualify, a loan must satisfy limits including a 43 percent cap on the borrower’s debt-to-income ratio and prohibitions on risky features such as negative amortization and interest-only periods beyond limits. First-lien qualified mortgages priced within 1.5 percentage points of the average prime offer rate receive a safe harbor from ability-to-repay liability, meaning the lender is conclusively presumed compliant; loans priced above that spread receive a rebuttable presumption. The framework was designed to prevent the unaffordable lending of the boom years, and its binding thresholds measurably shifted underwriting behavior.

Q: How did the 2018 tailoring law change Dodd-Frank coverage?

The 2018 legislation altered which firms faced the statute’s enhanced prudential requirements, raising the asset thresholds that triggered the toughest supervision, stress testing, and planning obligations, so that mid-sized firms faced a lighter regime than the original law imposed. Whether that tailoring contributed to the 2023 failures of Silicon Valley Bank, Signature Bank, and First Republic Bank, all in the affected size range, is contested. Some assessments argue reduced supervisory intensity left interest rate and liquidity risks under-monitored; others argue the failures reflected classic asset-liability mismanagement that earlier supervision would not necessarily have caught. The official reviews dated 2023 document the supervisory misses without settling the causal weighting, which is why this article labels the question contested rather than resolved.

Q: Why was the systemic risk exception used in March 2023?

On March 12, 2023, federal authorities invoked the systemic risk exception for Silicon Valley Bank and Signature Bank, a determination that allowed the FDIC to guarantee uninsured deposits at the two failed institutions beyond the standard insurance limits. The exception exists for situations where the standard least-cost resolution would have serious adverse effects on economic conditions or financial stability, and it was used because officials feared uninsured depositor runs would spread to other banks. First Republic Bank, which failed on May 1, 2023, was resolved through a sale without emergency powers (FDIC PR-23-019). The episode demonstrated willingness to act against contagion, but it used deposit insurance tools, not the Title II orderly liquidation authority, leaving the statute’s central mechanism untested.

Q: What do researchers say about consolidation and bank mergers after the law?

Researchers agree the decline in the number of banks continued after 2010 but disagree about the statute’s role in it. The FDIC’s Quarterly Banking Profile shows 7,658 banks at the end of 2010 falling to 6,812 at the end of 2013, continuing a contraction visible as early as 1984, when 14,477 insured commercial banks shrank to 12,246 by March 1991. That long trend was driven by branching deregulation, mergers, and technology, forces that predated Dodd-Frank. The law added compliance costs that weigh more heavily on small institutions, and research such as Bordo and Duca (NBER Working Paper 24501, April 2018) attributes post-2010 shifts in small-business lending substantially to the regulatory regime. But no serious study attributes the broader consolidation trend to the statute alone, and presenting the law as the author of consolidation misreads decades of data.

Q: Did Dodd-Frank solve the problem of banks being too big to fail?

No credible reading of the evidence supports that claim, and the statute’s own central mechanism for achieving it remains unproven. Too big to fail ends when a large firm can collapse without a taxpayer rescue and without contagion, and the tool built for that task, the Title II orderly liquidation authority, has never been used (CRS; FDIC OIG, December 2023). The 2023 failures required a systemic risk exception to protect uninsured depositors at two of the three failed banks, which is evidence that authorities still feared contagion from large failures. Higher capital made the largest banks more resilient, which is a genuine achievement, but resilience is not the same as resolvability. Until the resolution authority is used under stress, the end of too big to fail is an aspiration of the statute, not an outcome of it.

Q: How does the Glass-Steagall comparison frame the debate over the law’s design?

The comparison contrasts two theories of reform: Dodd-Frank regulated large complex firms with capital, supervision, and a resolution regime, while the Glass-Steagall alternative would have structurally separated commercial and investment banking or broken up the largest institutions. Critics of the statute’s approach argue no amount of supervision makes a too-big-to-fail firm safe, and they point to layer five, the untested resolution authority, as the weak point. Defenders point to layer one, the measured capital build, as proof the regulatory approach delivered real safety gains. The evidence reviewed here cannot settle which theory is correct, because the structural alternative was never enacted and its effects are a counterfactual. The comparison remains useful as a framing device for the law’s choices rather than as a verdict on them.

Q: What role did the Consumer Financial Protection Bureau play in the law’s outcomes?

The bureau was the statute’s consumer protection arm and the author of the mortgage rules whose effects are measured in layer four. Its ability-to-repay and qualified mortgage rule, finalized January 30, 2013 and effective January 10, 2014, set the 43 percent debt-to-income threshold and the safe harbor framework that measurably tightened underwriting for weaker credit profiles. The bureau’s January 2019 assessment of that rule is the principal source for the finding that non-qualified high debt-to-income approval rates fell across credit tiers and income groups. Beyond mortgages, the bureau’s supervision and enforcement reshaped practices in credit cards, servicing, and debt collection, though this article’s evidence layers focus on the mortgage channel where measurement is strongest. The bureau represents the consumer side of the statute’s record, distinct from the safety and soundness side.

Q: What evidence would settle the open questions about the law’s effects?

Three kinds of evidence would move the open questions toward answers. First, an actual Title II resolution under stress would test whether the orderly liquidation authority works as designed; until that happens, the central bargain remains an assertion. Second, a full credit cycle with the post-reform capital and liquidity regime in place would reveal whether the thicker cushions absorb losses as intended across different kinds of shocks, including the interest rate and depositor-run risks that the 2023 failures exposed. Third, continued research separating regulatory effects from cyclical ones, using methods that neither assume the answer, would narrow the contested ranges on community bank burden, de novo formation, and mortgage credit attribution. The article’s method is to name what the evidence cannot yet tell us, and these are the findings that would change the ratings.