The 10th Annual Conference on Bank Regulation, co-hosted by the Bank Policy Institute and Columbia University School of International and Public Affairs, took place on Feb. 20, 2026. Each year, the conference brings together academics, regulators, and practitioners to discuss the latest research on banking and bank regulation. The overarching themes of this year’s conference were (1) innovation and automation in banking, (2) banking supervision and (3) digitalization and banking. The conference concluded with a panel discussion on the outlook for digital innovation in banking, and what experts believe comes next.
The first session, “Innovation and Automation in Banking,” focused on how technology-driven decisioning and infrastructure choices are reshaping core banking outcomes – credit supply, pricing and competitive dynamics – while raising new consumer-protection and prudential tradeoffs. It paired evidence on automated credit card account management with analysis of banks’ cloud outsourcing to major service providers, highlighting that “automation” increasingly operates both at the product level and in the underlying operational stack.
Against that backdrop, the first paper examined a largely understudied feature of U.S. credit card markets: proactive, lender-initiated credit limit increases that are increasingly determined by proprietary (often model-driven) decision rules. Using account-level regulatory data from FR Y-14M, they document that limit increases are a meaningful channel of credit expansion: roughly one-third of new available credit comes from limit increases – particularly for lower credit score “low-and-grow” accounts – and that 80 percent of increases are initiated by banks, reflecting proactive and algorithmic credit management rather than as requested by consumers. They further show that the likelihood of receiving an increase is closely related to revolving behavior (with the strongest relationship at moderate revolving utilization), and that balances tend to rise following increases, consistent with the concern that additional available credit can translate into higher borrowing even absent an explicit credit request. Motivated by policy approaches discussed internationally (e.g., restrictions for persistent revolvers or consent requirements), the authors then use a life-cycle framework with heterogeneous self-control considerations to quantify welfare tradeoffs, finding that targeted limits on unsolicited increases to revolving borrowers can reduce indebtedness and improve welfare on net, with relatively modest costs for households that primarily use credit for consumption smoothing.
The session’s second paper examined how banks’ outsourcing to cloud service providers (CSPs) affects cost structures and competition in UK lending markets. Using a proprietary dataset of 3,443 material outsourcing contracts (importantly, including 457 cloud contracts) matched to regulatory bank data, the authors find that greater CSP spending is associated with lower operating costs (notably salary and occupancy expenses), alongside higher loan rates and greater lending to riskier segments such as SMEs. They then estimate a structural model in which CSP spending shifts both marginal costs and borrower demand, and use counterfactuals to highlight policy tradeoffs: restricting cloud outsourcing before its widespread adoption would have tended to shift market share away from large institutions – suggesting that CSP adoption disproportionately benefits large banks and reinforces market position – while reductions in capital requirements can raise overall lending but dampen incentives to invest in CSPs.
The discussants emphasized that both papers address highly policy-relevant questions and go beyond reduced-form correlations by using structural frameworks to evaluate concrete regulatory counterfactuals. On automated credit limit increases, they pointed out the paper’s careful framing of the access to credit versus over indebtedness tradeoff, highlighted the surprising empirical facts on the large share of credit coming from limit increases and their predominantly proactive (bank-initiated) nature while encouraging additional evidence to strengthen the interpretation that the targeted margin reflects “behavioral” borrowing rather than (or in addition to) screening and risk-management. They also noted that the implied welfare losses for fully rational consumers under tighter limits appear surprisingly small given the amount of credit affected. On cloud outsourcing, they praised the novelty of the contract-level approach but urged greater clarity on whether the object of interest is cloud adoption itself or cloud as a proxy for broader innovation/cost restructuring. The discussants urged caution interpreting fixed-effects panel results as causal without stronger timing assumptions, controls or alternative identification. They also requested more direct evidence on the specific mechanisms – cost restructuring beyond salaries and occupancy, demand-side convenience and risk assessment and operational resilience – and tighter integration of the capital requirements channel within the model and counterfactuals. Lastly, they sought further clarification on the interpretation of the relationship between capital requirements, cloud investment, and the associated policy implications.
The second session, “Supervision,” examined how supervisory information and discretion translate into real outcomes – both in the extreme case of distressed institutions approaching failure, and more broadly through examiner judgments embedded in routine ratings. In the first paper, the authors use confidential supervisory data on U.S. commercial bank examinations and CAMELS ratings from 2000-2023 to study how supervisors anticipate, monitor and ultimately discipline failing banks. They document that supervisory ratings predict bank failures at short horizons and played a role in loss recognition, enforcement actions, and the timing of closure. To move beyond description, they exploit quasi-exogenous variation in supervisory strictness during the onset of the Global Financial Crisis, leveraging the rotating assignment of some state-chartered banks between relatively stricter federal exam teams and state supervisors. Stricter supervision leads banks to recognize losses more quickly, reduce dividend payouts, and face a higher likelihood and faster pace of closure – lowering the FDIC’s resolution costs – while also producing a clear tradeoff through reduced lending and contraction in banking inputs during the crisis period.
The session’s second paper framed bank supervision as an information-production process and asked how much supervisory ratings add beyond what could be inferred from publicly available bank fundamentals and market signals. Using confidential Federal Reserve supervision records for U.S. bank holding companies from 2005-2023, the authors show that a benchmark model built on public financial and market data explains roughly 70-80 percent of the variation in ratings, but that the residual “examiner-specific effect” is economically meaningful and systematic. They find these deviations are larger precisely when examiners have greater access to soft, non-public information – such as through on-site work, geographic proximity and greater bank opacity – consistent with supervision adding value beyond public disclosure rather than reflecting arbitrary discretion. The paper also documents persistent heterogeneity across examiners and cyclical variation in supervisory tone – more lenient during expansions and more stringent following periods of stress, including around the 2023 banking turmoil – highlighting potential procyclical dynamics in supervisory intensity. These examiner-specific effects appear to have real consequences: unexpectedly stringent assessments are followed by contraction in assets and lending alongside higher capital ratios, while leniency is associated with subsequent risk accumulation (e.g., higher nonperforming loans), indicating a core tradeoff between prudential discipline and credit provision.
The discussants acknowledged that both papers point to meaningful examiner discretion embedded in supervisory ratings, with clear benefits for discipline and resolution outcomes but also costs through tighter credit supply, leaving the “right” degree of strictness and discretion an open calibration question, and emphasizing that outcomes can depend on which examiner is assigned to a bank rather than on fundamentals. On the first paper, they urged caution in interpreting short-horizon failure predictability as uniquely supervisory foresight, noting potential endogeneity in downgrades and enforcement actions once distress is externally apparent, and suggested additional diagnostics (e.g., which CAMELS components drive predictability) and survival-style approaches to model failure timing more directly. The conversation also briefly touched upon the notion of situating the short-horizon predictability result in the context of the broader distress-prediction literature. For the second paper, they flagged that its “examiner-specific effect” residual can mechanically bundle soft information with omitted variables and estimation noise and encouraged one-step specifications (and generated-regressor corrections) that integrate soft-information proxies more directly. Finally, they linked these findings to ongoing supervision reforms – greater emphasis on measurable metrics, fewer/narrower horizontals, higher escalation thresholds and leaner teams – arguing that the evidence suggests judgment, comparative insight and human-capital intensity are core to what makes supervision effective.
In the keynote address, Lorie K. Logan, President and Chief Executive Officer of the Federal Reserve Bank of Dallas, noted that the near-term macro outlook remains shaped by uncertainty – most notably around how quickly AI-driven investment will translate into durable productivity gains – and described monetary policy to be “well positioned” to adjust as risks shift across the dual mandate. She described inflation as moving gradually back toward target, pointing to recent progress in the data alongside signs of softening in housing services, while noting that tariff-related price pressures could linger, leaving open the possibility that the disinflation path may not be smooth. Logan framed current strength in business investment, particularly tied to AI and the buildout of data, energy, and water infrastructure, as supportive over the medium run, even if demand can outpace supply in the short run. On AI and labor markets, she noted little evidence of broad displacement so far, but flagged concerns that effects may be more visible “under the surface,” particularly for newer labor-market entrants, emphasizing the importance of training and workforce preparation. A recurring theme in the Q&A was financial stability in an increasingly digitalized environment: Logan stressed that banks should plan for rapid deposit shifts by maintaining diversified depositor bases and contingency funding sources, routinely testing access and operational readiness, and improving monitoring of high-frequency channels – particularly in an era when social media can accelerate runs. She also touched on ongoing efforts to modernize liquidity backstops, including reducing frictions and standardizing access to the discount window and encouraging greater familiarity with standing facilities, while arguing that debates over deposit insurance and central bank liquidity should be approached with a “level playing field” perspective that recognizes the role of small banks as well as systemically important firms. Finally, she highlighted how innovations such as 24/7 payments and evolving liquidity regulation interact with the Fed’s operating framework, noting that clarifying what an ample reserves regime means, and ensuring that liquidity buffers are usable rather than layered into “buffers on buffers,” will be increasingly important as market structure and policy implementation continue to evolve, including in coordination with international counterparts during stress events.
The third session examined how digitalization in banking – from household crypto investing to new payment technologies – can reshape bank funding conditions and, through that channel, affect credit supply and broader macro/financial outcomes. The first paper combines IRS tax-record measures of local crypto participation with banks’ geographic deposit footprints and regulatory balance-sheet data to study the time frame of 2020-2022. They show that when cryptocurrency returns are high, retail deposits fall more in markets with higher crypto participation, and banks with greater exposure to those markets experience weaker deposit growth – concentrated in savings-type (non-transaction deposits), consistent with crypto competing as a savings/investment vehicle rather than a payments substitute. These funding outflows translate into slower loan growth at more exposed banks, with instrumental-variable estimates indicating economically meaningful pass-through from crypto-driven deposit shocks to credit contraction. Finally, the paper links these intermediation effects to the real economy: counties where high-exposure banks hold a larger deposit share see slower establishment and employment growth, particularly in bank-dependent industries, underscoring an indirect channel through which crypto market cycles can weaken traditional financial intermediation.
The session’s second paper studied how the rapid adoption of digital payments can reshape monetary policy transmission by changing banks’ deposit-market power. Leveraging administrative data on Brazil’s Pix system (introduced in late 2020), the authors combine municipality-level Pix usage with branch-level bank balance sheets and bank-level interest rates to show that, as Pix adoption rises, deposits become more rate-sensitive and banks’ local market power declines – so banks adjust deposit pricing more aggressively when policy rates move. Consistent with that mechanism, they find stronger pass-through and tighter credit supply in high-Pix areas: following a 1-percentage-point policy-rate increase, banks in municipalities with higher Pix usage exhibit smaller increases in deposit spreads (i.e., deposit rates respond more), weaker deposit growth and reallocation away from bank deposits, and larger contractions in lending. They interpret these patterns through a dynamic banking model in which digital payments increase deposit demand elasticity, and their counterfactuals imply that digital payments amplify the effects of monetary tightening primarily through the deposit-competition channel.
The discussants agreed that both papers are tackling first-order questions about how digital innovation can alter bank funding and, in turn, credit supply and policy transmission, and praised the creativity of the data construction and institutional settings. At the same time, they urged caution on identification and interpretation. For the first paper, they addressed the strength of the county-level “crypto-interest” mapping to bank deposit footprints, but also acknowledged that the 2020-2022 window was unusually noisy for deposits and small-business credit (e.g. stimulus-era flows), and that correlated shocks and local compositional effects could complicate the exclusion restriction (including the overlap of crypto participation with high-growth direct banks and fintech/NBFI lending, as well as other bank management actions like pricing promotions, marketing, and branch changes). They also raised the question of whether observed lending changes reflect true funding constraints, noting low loan-to-deposit ratios during the period as a reason to probe banks’ marginal funding conditions more directly. For the second paper, they viewed the rollout as a compelling natural experiment but also stated that the “before/after” comparison coincides with broader regime changes – only easing pre-Pix and only tightening post-Pix, and potential mechanical compression effects near the lower bound – alongside the broader rise of mobile/digital banking, making it important to separate the effect of fast payments from contemporaneous shifts in competition and technology adoption. They encouraged stronger controls and robustness around fintech-bank growth, pricing differentials, and other local factors (e.g., marketing/awareness and smartphone access), and tighter mapping from the deposit-channel mechanism to lending outcomes given the multitude of moving parts in the institutional setting.
The closing panel, “Digital Innovation and Banking, What Happens Next?”, began with discussing a central tension: banks are often perceived as slow to innovate, yet their mandate to safeguard other people’s money and operate within tight supervisory expectations naturally biases them toward technologies that are proven, scalable and resilient. Panelists posited that this “high-accuracy” requirement is less about institutional reluctance and more about the fact that even small operational failures can have outsized consequences in regulated finance – so adoption tends to occur only once governance, controls, and end-to-end reliability are demonstrably mature.
A second theme was that digital innovation is increasingly shifting the architecture of finance from account-based models toward wallet-based models, with tokenization expanding what a single wallet can hold and move – cash-like instruments, tokenized deposits and potentially real and financial assets side by side. In that framework, the bank’s role can be compressed if users no longer need a traditional deposit account to store value and transact, particularly as on-chain rails enable global transfers and new intermediaries sit between consumers and the legacy banking stack. Panelists noted, however, that banks are actively adapting – often through partnerships – especially where smaller institutions may rely on “tokenization-as-a-service” arrangements to remain relevant in a tokenized ecosystem.
The panel also noted that the most immediate competitive pressure is playing out in payments, where banks have long excelled in large-value institutional rails and securities-related flows but have left friction and cost in cross-border remittances and small-business payments. That gap has created room for new entrants offering faster, cheaper and more user-centric services – and, in parallel, has increased demand for more direct access to payment systems as de-risking and compliance burdens have reshaped correspondent relationships. At the same time, panelists stressed that as the ecosystem digitizes, risk, compliance and resiliency become non-negotiable regardless of charter status – KYC/AML, fraud controls and especially cyber risk – stating that innovation that outpaces controls can amplify systemic vulnerabilities rather than reduce them.
Finally, discussion of crypto and stablecoins emphasized a set of ironies: digital-asset markets can promise democratization, but scaled, trusted finance still depends on the costly, longstanding infrastructure – secure rails, operational resilience and market plumbing – that banks have built and maintained. Panelists noted that crypto and stablecoins are unlikely to function as standalone systems at scale without being “permissioned” into the rules-based financial system and connected to real-world payments and banking functions, and that the boundary between banks and fintechs is blurring as stablecoin issuance and similar activities begin to resemble bank-like maturity and liquidity transformation. The policy implication is that the next phase of innovation will hinge on rules-based frameworks and a level playing field that prevent fragmentation into “walled gardens,” so that networks remain interoperable and the broader cost of doing business does not rise as each platform builds its own siloed rails.
