Why CCAR 2015’s failures were priced in dividends and paid for in data

In March 2015 the Fed passed every large bank on the numbers yet stopped two from paying out, over how their plans were built. Capital planning turned into a spending decision on data, models and controls, bought under examiners’ deadlines.

Looking Back revisits past weeks one at a time, up to the most recent, and asks what it meant for the banks, health systems and public agencies we serve. The series was written in 2026, with the benefit of hindsight, and each article says which week it looks back on.

Two-minute summaryThe dek and the first sentence of each section, quoted from the article

In March 2015 the Fed passed every large bank on the numbers yet stopped two from paying out, over how their plans were built. Capital planning turned into a spending decision on data, models and controls, bought under examiners’ deadlines.

  1. The week Two Federal Reserve releases framed the second week of March 2015 for the largest U.S. banks.
  2. The price of failing: what a qualitative objection cost A qualitative objection did not say a bank was short of capital.
  3. Where the money went: data, models, controls and people The Fed’s list of what it reviewed reads like a cost breakdown.
  4. How it was bought: build, buy or rent Every firm in CCAR faced the same sourcing choice for each of the categories above.
  5. Who else paid: the cost cascade beyond the 31 firms CCAR applied to the largest bank holding companies.
  6. Capital planning or compliance theater: the strongest objections A fair critic of this reading would raise three objections.
  7. Setting the budget for process quality today The March 2015 results are now more than a decade old.
  8. With hindsight The two firms that failed in March 2015 failed again the next year.

The week

Two Federal Reserve releases framed the second week of March 2015 for the largest U.S. banks. On Thursday, March 5, the Board published the results of its supervisory stress tests under the Dodd-Frank Act. On Wednesday, March 11, at 4:30 p.m., it published the results of the Comprehensive Capital Analysis and Review, known as CCAR [1] [2].

The March 5 release covered 31 bank holding companies. The Fed said they held more than 80 percent of domestic banking assets. It was the fifth round of Fed-led stress tests since 2009 and the third required by Dodd-Frank. The Fed ran the numbers on its own models, not the banks’ [2].

The severely adverse scenario assumed unemployment peaking at 10 percent, home prices falling 25 percent and stock prices dropping by nearly 60 percent. Under it, the Fed projected $340 billion of loan losses across the 31 firms over nine quarters. The aggregate tier 1 common ratio fell from an actual 11.9 percent in the third quarter of 2014 to a projected minimum of 8.2 percent. The Fed set that against 5.5 percent at the start of 2009. It called the results deliberately conservative hypothetical estimates, not forecasts [2].

The stress test was one input to CCAR. CCAR asked a broader question: could each firm’s board justify its planned dividends and share buybacks, given its capital under stress and the quality of the process that produced the plan? The firms had submitted their capital plans in January 2015. The projections ran from the fourth quarter of 2014 through the fourth quarter of 2016 [3].

On March 11 the Fed objected to no firm on quantitative grounds. It objected to two on qualitative grounds: Deutsche Bank Trust Corporation and Santander Holdings USA. Both were U.S. holding companies owned by European banks. An objection meant the firm could make only the capital distributions the Fed expressly permitted [1].

The Fed’s summary described Santander’s deficiencies as widespread and critical across its capital planning processes. It was Santander’s second consecutive objection; the Fed had also objected to its plan in 2014. For Deutsche Bank Trust Corporation, the Fed said it had applied the standards it uses for the largest and most complex U.S. firms, given the size and complexity of Deutsche Bank’s U.S. operations. It found numerous and significant deficiencies in risk identification, measurement and aggregation, in the approaches to projecting losses and revenues, and in internal controls [3].

Bank of America received a conditional non-objection. The Fed cited weaknesses in parts of its loss and revenue modeling and in some internal controls. It said those weaknesses did not undermine the quantitative results. The condition was a new capital plan by September 30, 2015, with an objection expected if the problems were not fixed [3].

The other 28 firms received non-objections. Three of them, Goldman Sachs, JPMorgan Chase and Morgan Stanley, met the post-stress minimums only after submitting adjusted, smaller capital actions [1].

Two details in the March 11 release matter for this article. First, the Fed published corrected stress test results for four firms that had submitted incorrect data. The corrections raised those firms’ starting and post-stress ratios. Second, the Fed listed what its qualitative review covered: governance, internal controls, risk identification and risk management, management information systems, and the assumptions and analysis behind the plan [1] [3].

The Fed also gave the aggregate picture. The common equity ratio of the tested firms had risen from 5.5 percent in the first quarter of 2009 to 12.5 percent in the fourth quarter of 2014. Common equity had grown by more than $641 billion, to $1.1 trillion. The firms held about $14 trillion of assets [1].

The executive summary added a note on direction. The Fed said it had allowed firms time to reach its capital planning standards. It said it expected steady progress from the largest firms, and that it would keep monitoring through the year [4].

The headline of March 11, 2015 was reassurance. Capital at the largest banks had more than doubled since 2009, and no firm failed on the numbers. The more instructive part of the release sat in the qualitative column. Two firms with adequate projected capital were still barred from distributing it, because the Fed did not trust how their numbers were made. A third was told to rebuild parts of its process within six months.

That outcome turned capital planning into a spending decision with a visible price on failure. A board could now compare two figures: the cost of fixing data, models and controls, and the cost of being unable to pay dividends or buy back stock. Once those figures sat side by side, the work moved from the risk function’s wish list into the budget.

We read the week through the economics of that choice. It breaks into four parts that follow the money in order:

  • The price of failing: what a qualitative objection or a conditional non-objection actually cost, and who bore it.
  • Where the money went: the categories of spending that a credible capital plan required.
  • How it was bought: the choice between building, buying from vendors and renting consultants, and the trade-offs of each.
  • Who else paid: how the costs spread to institutions outside CCAR, from mid-size banks to large credit unions to the vendors that serve them.

The first part sets the size of the incentive. The second and third describe how firms spent against it. The fourth follows the costs past the 31 firms that were tested. Two shorter sections close the argument: one tests it against the strongest objections, and one turns it into practice for a finance or risk leader today.

The price of failing: what a qualitative objection cost

A qualitative objection did not say a bank was short of capital. It said the Fed could not rely on the process that produced the bank’s capital plan. The practical result was the same as a capital shortfall for shareholders: distributions stopped unless the Fed allowed them.

Direct costs

For a listed U.S. bank holding company, the first cost was the dividend or buyback that did not happen. Shareholders who expected a payout waited a year. The stock price absorbed the signal. Management lost a lever it used to manage capital ratios and returns on equity.

For the two objecting firms in 2015, the shape of the cost was different. Deutsche Bank Trust Corporation and Santander Holdings USA were subsidiaries of foreign banks. Their capital distributions mostly moved money to a parent, not to public shareholders. An objection meant capital stayed trapped in the U.S. entity. The parent could not redeploy it to other markets or use it to support group capital ratios. For a European group under its own capital pressure, that was a balance sheet cost, not a market signal.

The conditional non-objection carried a smaller but real price. Bank of America could distribute capital, but it had to submit a new plan by September 30. That meant a second full capital planning cycle in one year, run by the same teams that were supposed to be fixing the weaknesses. The deadline also set a clock on remediation. If the fix was not credible by the fall, the Fed said it expected to object.

Indirect costs

The indirect costs were harder to see and often larger. An objection drew supervisory attention to the whole risk management program. Findings on capital planning tended to travel into other examination areas: model risk, data governance, internal audit and board oversight. Each finding needed a remediation plan, an owner, milestones and evidence. Each carried its own staffing need.

There was also an opportunity cost inside the bank. The people who understood the balance sheet best, in finance, treasury and risk, spent months on remediation. Projects that might have improved products, pricing or client service waited. Most firms did not record this cost, but every executive who lived through a remediation knew it.

The adjusted-plan option

The quantitative side had its own economics. Goldman Sachs, JPMorgan Chase and Morgan Stanley passed only after cutting their planned capital actions. The process gave firms one chance to revise after seeing the Fed’s projections. The cost of that path was smaller distributions, not an objection. It was also a cost that a better internal forecast could reduce. A firm whose own models came close to the Fed’s results could ask for the right amount the first time.

That created an incentive that ran through the whole program. The closer a firm’s internal projections were to the supervisor’s, the less capital it left idle as a cushion. Better loss and revenue models therefore had a measurable payoff in distributions. They paid back in capital, which few compliance costs do.

Why the price mattered for budgets

Before CCAR, a chief risk officer asking for money to clean up data lineage had to argue in terms of risk reduction, which is hard to price. After March 2015, the same request could cite a concrete downside: a blocked dividend, trapped capital or a second planning cycle. Chief financial officers understood that trade. Budgets for capital planning infrastructure grew because the cost of failure had become concrete and public.

The same logic cut the other way. Once the price of failure was known, a firm could also overspend. Spending more than the downside justified was possible, especially when remediation was run as a crash program under supervisory deadlines. The next two sections look at where the money went and how it was bought, because those choices decided whether the spending built lasting capability or paid for a temporary fix.

Where the money went: data, models, controls and people

The Fed’s list of what it reviewed reads like a cost breakdown. Governance, internal controls, risk identification, management information systems and the assumptions behind the projections each required a different kind of investment. We group them into four categories that most finance and risk leaders will recognize.

Data: from regulatory reports to a reconciled source

The March 11 release corrected stress test results for four firms because they had submitted incorrect data. That detail explains much of the spending that followed. The Fed’s projections started from the firms’ own regulatory reports, including the detailed FR Y-14 schedules that fed the stress test. If the inputs were wrong, the Fed’s outputs were wrong, and the firm’s plan was judged on bad numbers.

Fixing that required work that is easy to describe and slow to finish. A firm had to know where each reported figure came from, which system produced it and which transformations touched it on the way to the report. It had to reconcile the stress test submissions with the FR Y-9C and the general ledger, and explain differences. It had to prove the controls on each step, with evidence an examiner could test.

The spending landed in several places. Data warehouses or data marts dedicated to regulatory reporting were one. Data lineage tools and metadata catalogs were another. Reconciliation software and the people to run it were a third. Many firms also had to fix source systems, such as loan servicing platforms with incomplete fields. The CCAR data problem was often a loan system problem that had existed for a decade.

Deutsche Bank Trust Corporation’s cited weakness in risk identification, measurement and aggregation pointed at the same issue. Aggregation was the hard word. A firm with several legal entities, booking systems and product lines needed to produce one consistent view of its exposures. The Basel Committee had published its principles for risk data aggregation in 2013, aimed at globally systemic banks. CCAR made similar expectations concrete for U.S. holding companies of all the sizes in the program.

Models: loss, revenue and the challenge around them

Bank of America’s cited weakness was in loss and revenue modeling. Deutsche Bank Trust Corporation’s included its approaches to projecting losses and revenues. Those are the core of a capital plan: how much a portfolio would lose under stress, and how much income the firm would still earn.

Building those models cost money in three layers. The first was development: statisticians and economists who could connect macroeconomic variables to portfolio losses and revenues. The second was validation: an independent team that tested the models, their data and their limits. Supervisory guidance on model risk management, issued by the Fed and the OCC in 2011, already expected that independence. The third was the overlay process: the documented judgment that adjusted model outputs where the models were weak.

Revenue models were often the harder problem. Losses on a mortgage book could be modeled from long data histories. Fee income, trading revenue and deposit behavior under stress were less tractable. Weak revenue projections produced plans that looked too optimistic, which is exactly what a supervisor would challenge.

Firms also had to pay for challenger models and benchmarks. A capital plan that relied on a single model for a major portfolio invited the question of what happened if that model was wrong. Running a second approach, or buying an external benchmark, gave the board and the Fed a way to judge the main estimate.

Scenarios: the firm’s own view of stress

CCAR did not rest only on the Fed’s scenarios. The capital plan rule also asked each firm to design its own baseline and stress scenarios, fitted to its particular risks. A bank concentrated in commercial real estate in one region needed a stress scenario that hurt that book. A trading bank needed one that hurt its market positions and counterparties.

Designing a firm-specific scenario cost money in a less obvious way. It required economists and business heads to agree on what would actually hurt the firm, and to write that down. It required a process to translate the narrative into the variables the models used. It required the board to accept a scenario that showed the firm in a bad light. Supervisors were skeptical of firm scenarios that turned out milder than the Fed’s, so the exercise had to be honest to be credible.

The return on that work was mostly internal. A well-built firm scenario told management which risks it had underpriced. That was useful long after the submission date, if someone in the business was listening.

Controls: proving the process worked

Internal controls appeared in both the Deutsche Bank and Bank of America findings. For capital planning, controls meant more than system access. They covered the process from data to board approval: who could change a model assumption, who reviewed it, how overrides were documented and how errors were caught.

The cost of controls was mostly people and time. Internal audit had to test the capital planning process as it would test financial reporting. Process owners had to write procedures and keep evidence. Many firms borrowed from their Sarbanes-Oxley control frameworks, which already had the vocabulary of control design and operating effectiveness. The difference was that capital planning involved judgment and forecasts, not historical transactions. Controls on judgment are harder to write.

The evidence an examiner asked for gave the cost its shape. A typical request covered the inventory of models used in the plan, with their validation status and open findings. It covered the reconciliation of submitted data to the general ledger, with explanations for breaks. It covered the record of management overlays, who approved them and why. It covered board minutes showing that directors had challenged the plan rather than received it. A firm that could produce those records quickly spent less on each examination. A firm that had to assemble them by hand paid the same cost every year.

Governance and people

The Fed expected boards and senior management to oversee capital planning effectively. That required board materials that explained assumptions, limitations and alternatives in terms directors could challenge. It required senior managers who understood the models well enough to own them.

The people cost here was concentrated in scarce skills. Experienced model developers, validators and regulatory reporting specialists were in demand across the industry at once. Thirty-one firms were competing for the same talent in the same years. Salaries rose, and turnover created its own risk: a model whose developer left was harder to defend.

How the categories interacted

The four categories were not independent. Better data made models more reliable. Stronger models made controls easier to design, because fewer manual overrides were needed. Good governance decided where to spend first. Firms that funded one category in isolation tended to find the problem reappearing in another. A model built on unreconciled data failed validation. A control framework over a manual process cost more each year than automating the process.

That interaction is why the cost of a credible capital plan was larger than the sum of the visible projects. The underlying investment was an operating model: a recurring annual cycle with stable data, maintained models, tested controls and informed oversight. The question for a CFO was how to buy that operating model at a cost the firm could sustain.

How it was bought: build, buy or rent

Every firm in CCAR faced the same sourcing choice for each of the categories above. It could build capability in-house, buy software and data from vendors, or rent expertise from consultants. Most used all three. The mix decided what the firm owned when the consultants left.

Building in-house

Building meant hiring model developers, data engineers and process owners, and writing the firm’s own tools. The advantages were ownership and fit. A firm that built its own loss models understood their assumptions and could explain them to the board and the Fed. Examiners tended to favor that, because the firm could answer questions without calling a vendor.

The disadvantages were time and talent. Building took years, and CCAR ran on an annual clock. A firm that had just received an objection had months, not years, to show progress. Talent was scarce and expensive. Building also concentrated key-person risk: a small team that understood a critical model was a vulnerability.

Buying from vendors

Vendors offered several kinds of products by 2015. Some sold platforms to orchestrate the stress testing process: scenario management, model execution, aggregation and reporting. Others sold data and benchmark models, such as industry loss histories or macroeconomic scenario tools. Regulatory reporting vendors sold software to produce the FR Y-14 and related schedules.

Buying could shorten the timeline and bring in tested components. It carried two risks. The first was fit: a vendor platform designed for a generic bank needed configuration and integration with the firm’s own systems, often at a cost close to building. The second was understanding. A firm still had to own and explain any model it used, including a vendor model. Supervisory model risk guidance did not exempt purchased models from validation. A vendor that would not disclose enough about its methodology made validation harder and sometimes impossible.

Procurement terms mattered more than usual. A firm buying a stress testing component needed rights to the documentation required for validation, access to change logs, notice of model changes and support during the examination cycle. It also needed an exit plan. A firm locked into one vendor’s platform faced a large switching cost if the vendor fell behind regulatory changes. Those terms were negotiable in 2015, but only if the procurement team knew to ask.

Renting consultants

Consultants were the fastest option and the most common response to an objection. Large advisory firms had teams who had worked on capital planning at several banks. They could stand up a remediation program, write documentation, build or validate models and prepare board materials in a short time.

The economics of renting had a hidden cost. Consultants were paid to deliver a result in a cycle. Knowledge transfer was often an afterthought. When the engagement ended, the firm could be left with models and documents its own staff did not fully understand. The next cycle then required another engagement. Supervisors noticed this pattern. A plan built and explained by outsiders did not show the internal ownership the Fed’s governance expectations called for.

There was also an independence issue. A consulting firm that built a model could not credibly validate it. Firms that used one adviser for development and another for validation paid twice. Firms that skipped the second engagement risked a validation finding.

A sourcing approach that held up

The firms that got the most lasting value tended to follow a pattern. They used consultants for speed in the first cycle, with contracts that required knowledge transfer and joint working with named internal staff. They bought platforms and data where the market had mature products and where the vendor would support validation. They built in-house where the model reflected the firm’s own business: its main loan books, its revenue lines and its balance sheet behavior.

The procurement function had a real role in this, though it was often brought in late. A sourcing plan for capital planning needed several features. It needed a clear map of which capabilities were core and which were commodity. It needed contract terms that protected validation and exit. It needed a budget view that covered the recurring annual cycle as well as the remediation project. And it needed someone tracking how much of the program depended on people who did not work for the firm.

Sequencing the spend

The order of purchases mattered as much as the mix. Firms that bought a stress testing platform before fixing their data often found the platform produced fast, well-formatted versions of the same wrong numbers. Firms that rebuilt models before establishing a validation function found their new models failing review. The sequence that tended to work started with governance, which decided ownership and priorities. Data came next, because every later step depended on it. Models and validation followed, then automation of the cycle.

That sequence was hard to follow under an objection. The Fed expected visible progress within a year, and data work is slow to show. Firms often ran the steps in parallel and accepted rework as the price of speed. A finance leader approving the budget should have expected that rework and funded it, rather than treating it as a sign of failure.

Measuring value for money

Measuring the return on this spending was difficult, but not impossible. Useful measures included the number of open supervisory findings on capital planning and their age. Others were the share of reported data fields with documented lineage and the share of material models with current independent validation. The number of manual adjustments in each cycle mattered, and so did the gap between the firm’s projections and the Fed’s. A falling gap meant the firm was asking for distributions closer to what the Fed would allow, which had a direct value to shareholders.

Who else paid: the cost cascade beyond the 31 firms

CCAR applied to the largest bank holding companies. Its costs and its methods did not stop there. They spread through supervisory expectations, through parallel rules for smaller firms and through the vendor market.

Mid-size banks under company-run stress tests

Dodd-Frank also required company-run stress tests for banks and holding companies with more than $10 billion in assets. Those firms were not in CCAR and faced no capital plan objection. They did have to run their own stress tests on supervisory scenarios, report the results to their regulator and publish a summary.

For a regional bank with $15 billion in assets, the requirement raised the same questions as CCAR at a smaller scale. Its data needed lineage. Its models needed validation. Its board needed to understand the results. The bank had fewer staff and smaller budgets, so it leaned more heavily on vendors and consultants. That made the procurement terms described above even more important. A mid-size bank was more likely to depend on one vendor for its whole stress testing process, and less able to absorb a switching cost.

There was also expectation creep. Examiners of mid-size banks saw how CCAR firms ran capital planning. Practices that began as expectations for the largest firms became reference points in examinations of smaller ones. A regional bank could find that a practice described in the Fed’s CCAR publications had become, in its examiner’s view, simply what a sound bank did.

Credit unions

The National Credit Union Administration had adopted its own capital planning and stress testing rule in April 2014, for federally insured credit unions with $10 billion or more in assets. At adoption, NCUA said four credit unions were above that threshold. Those institutions faced a version of the same questions: credible loss projections, governance by a volunteer board, and data that could support both.

Credit unions had a different economics of failure. They had no public shareholders and paid no dividends in the bank sense. The pressure came through net worth and supervisory standing. The spending case therefore had to rest on safety and on the supervisory relationship, not on a blocked payout. That made it harder to win budget. The lesson from CCAR was still relevant: the cost of building capital planning capability was lower when done steadily than when done under a supervisory deadline.

Foreign banks’ U.S. operations

The objection to Deutsche Bank Trust Corporation had a particular audience. The Fed’s enhanced prudential standards, adopted in 2014, required large foreign banking organizations to place their U.S. subsidiaries under an intermediate holding company by July 2016. Those holding companies were expected to come into the capital planning program. Every foreign bank with large U.S. operations could read the March 2015 finding as a preview of the standard it would face.

For those groups, the economics were especially awkward. Capital planning infrastructure built for the U.S. entity often duplicated work done at the parent for European supervisors. Data and models had to meet U.S. expectations on U.S. timelines. The cost per dollar of U.S. assets could be higher than for a domestic bank of similar size, because the firm was building a second capital planning function inside a group that already had one.

Vendors and service providers

The vendor market carried its own costs. Software and data firms had to keep their products aligned with annual changes to scenarios, reporting forms and supervisory expectations. Each change meant development work and new releases on the regulators’ calendar. Some vendors raised prices. Others narrowed their offerings to the largest clients.

For a bank, that meant vendor risk management had to cover regulatory alignment as well as security and continuity. A stress testing vendor that missed a reporting change could leave its clients exposed. Contracts that tied the vendor to timely updates, with remedies if it fell behind, were worth negotiating.

Internal audit and the second line

Inside each firm, part of the cost landed on functions that did not own capital planning. Internal audit had to build the skill to test models and forecasts as well as transactions. Independent risk management had to review assumptions it had not designed. Both had to hire people with quantitative backgrounds, often competing with the first line for the same candidates.

That shift had a lasting budget effect. Once internal audit had a model audit team, it kept one. Once risk management had a capital planning review function, the next supervisory expectation was that it would cover other forecasts too, such as liquidity stress testing and recovery planning. The 2015 spending raised the baseline cost of the oversight functions for years.

Consumers of bank capital

The last group to pay was diffuse. Capital planning costs were operating expenses. Banks recovered them, where they could, through pricing. Some economists and industry groups argued that the combined weight of stress testing and capital rules reduced lending in certain segments. Others argued that stronger capital planning made banks safer lenders through a downturn. We do not have reliable figures that settle that debate. The direction of the cost was clear: it moved from the regulated firm to its customers and shareholders, as compliance costs generally do.

Capital planning or compliance theater: the strongest objections

A fair critic of this reading would raise three objections. Each deserves an answer.

The work was documentation, not capability

The first objection is that much of the spending produced documents, not better decisions. Firms wrote thousands of pages of model documentation and process descriptions to satisfy examiners. Some of that material had little effect on how the firm ran its balance sheet.

There is truth in that. A capital planning program can become a reporting exercise that runs alongside the real business. The test is whether the same data and models were used for internal decisions, such as limit setting, pricing and strategic planning. Firms that wired stress testing outputs into those decisions got value beyond compliance. Firms that kept a separate regulatory process paid for two views of risk and used one.

The qualitative standard was opaque

The second objection is that the qualitative assessment was hard to predict. Firms could not see exactly what the Fed would judge as adequate. That uncertainty encouraged overspending, because no firm wanted to be the one that under-invested.

That criticism was widely made at the time, and the Fed itself later narrowed the qualitative objection, as the hindsight section describes. It strengthens the economic reading rather than weakening it. Uncertainty about a standard raises the expected cost of failing it. Rational firms buy more insurance against an opaque test. The answer to opacity was to spend on things that had value whether or not the Fed asked for them: clean data, validated models and real board challenge.

Large firms could afford it and small firms could not

The third objection is that the costs fell unevenly. A global bank could spread its capital planning costs over a very large balance sheet. A firm at the bottom of the CCAR range paid a much higher cost per dollar of assets. The same was true of mid-size banks under company-run tests.

That is accurate, and it is part of why Congress and the agencies later raised thresholds. It does not change the decision each firm faced in 2015. Within the rules as they stood, the firm had to choose how to spend. A smaller firm had more reason to buy shared components and less reason to build everything in-house. Its sourcing choices mattered more, not less.

Setting the budget for process quality today

The March 2015 results are now more than a decade old. The specific objection mechanism has since been removed. The economic logic still applies whenever a supervisor can restrict a firm’s actions based on the quality of its processes.

Price the downside before the examiner does

A firm that waits for a finding pays crisis prices. Remediation under a deadline costs more per unit of progress than steady investment. Finance leaders can estimate the downside of a process failure in their own terms: blocked distributions, trapped capital, a repeat cycle, or a consent order. They can fund the fix before the finding arrives.

Buy the operating model, not the project

Capital planning is an annual cycle. Spending that builds a recurring capability is worth more than spending that clears one finding. Contracts with vendors and consultants should be written for that cycle: knowledge transfer, validation rights, update obligations and exit terms.

Count the people risk

The scarcest input in 2015 was experienced people. That is still true for model risk, regulatory data and risk governance. A program whose key knowledge sits with a consultant or one internal expert has a weakness that no software purchase fixes.

Use the same data for decisions and for regulators

The highest return on regulatory data spending comes when the business uses it too. A loss model that informs pricing and limit decisions is reviewed and challenged every day by people with an interest in getting it right. A model used only once a year for a regulatory submission gets less scrutiny and decays faster.

The Fed’s March 11 release put dollar consequences on process quality. The firms that treated that as a reason to build lasting infrastructure got more from their spending than those that treated it as a cost of passing a test. For institutions facing any supervisory regime that judges process as well as outcomes, that remains the right way to set the budget.

With hindsight

The two firms that failed in March 2015 failed again the next year. In June 2016 the Fed objected to the capital plans of Deutsche Bank Trust Corporation and Santander Holdings USA, again on qualitative grounds. It gave Morgan Stanley a conditional non-objection and objected to no firm on quantitative grounds [5]. The pattern showed how slowly process weaknesses could be fixed, even with a public price attached.

The Fed then narrowed the qualitative test in steps. In January 2017 it removed the qualitative assessment for large and noncomplex firms. In March 2019 it announced it would limit the qualitative objection to firms that had not yet passed it in their fourth year in the program. Five firms remained exposed in the 2019 cycle, including DB USA Corporation, the intermediate holding company for Deutsche Bank’s U.S. operations [6].

The March 2019 release was clear on one point. Firms no longer exposed to a qualitative objection still faced a rigorous review of capital planning. Weak practices could lead to a deficient supervisory rating and possibly an enforcement action [6]. The public objection went away. The supervisory expectation stayed, enforced through ratings and findings instead of a headline.

In March 2020 the Board approved the stress capital buffer. The rule combined stress test results with the Fed’s non-stress capital requirements into one framework. It reduced the number of capital requirements large banks had to meet from 13 to 8 [7]. A firm’s stress test results now set part of its ongoing capital requirement, rather than serving as a pass or fail gate for an annual plan. In early 2021, a further rule removed the qualitative objection from the capital plan rule entirely.

Thresholds also moved. The Economic Growth, Regulatory Relief, and Consumer Protection Act of 2018 raised the asset level at which company-run stress tests were required, and the agencies’ tailoring rules followed. Many of the mid-size banks that had built stress testing programs in 2015 were no longer required to run them. Most kept some version, because the data and models were useful for their own planning and because examiners still asked how they managed capital under stress.

NCUA moved in a similar direction. In 2018 it revised its credit union rule to reduce burden: covered credit unions under $15 billion in assets no longer had to run annual stress tests, and the largest could run their own instead of relying on NCUA.

The debate about the stress test itself continued. In December 2024 the Fed said it would seek comment on changes to make the test more transparent and its capital results less volatile. In April 2025 it proposed averaging two years of results to reduce swings in capital requirements, and Governor Barr publicly dissented. In October 2025 it proposed publishing its stress test models and its scenario design framework for comment, and asked for comment on the 2026 scenarios [8]. The Board said the model and scenario changes would not materially change capital requirements. On September 30, 2026 the Board finalized the changes. It will invite public input each year on the scenarios and on material model changes, and from 2028 the stress capital buffer will draw on a firm’s two most recent tests. The Board said it did not expect aggregate capital requirements to change materially.

What held up from the 2015 reading is the economics. The qualitative objection is gone, but the spending it prompted bought data lineage, model validation and controls that firms still use. Those investments now support supervisory ratings, internal capital decisions and the stress capital buffer calculation itself. Firms that treated 2015 as a one-time remediation have had to rebuild as expectations shifted. Firms that built an operating model have mostly adapted.

What did not hold up is the idea that the qualitative test would remain the main enforcement tool. The Fed moved toward rules that set capital mechanically from stress results, with process quality judged through ordinary supervision. For a CFO, that shifts the economics. The downside of weak capital planning is now less visible and more gradual: rating downgrades, findings and constraints on growth, rather than a blocked dividend announced on a Wednesday afternoon.

The transparency changes have a further consequence. With the Fed’s models and scenarios open to comment, firms can predict their results more precisely. That rewards those whose internal data and models are good enough to use that information. The investments of 2015 matter more under a transparent regime, not less.

Spotted an error? Tell us and we will check it and correct the article where needed.

  • CCAR
  • stress testing
  • capital planning
  • Federal Reserve
  • model risk
  • risk data