Bank of England Model Finds a 32% Offset to Capital Relief

Cutting bank capital requirements may lower funding costs without delivering the full benefit that a conventional model would predict. A new Bank of England staff study finds that the same relief could weaken investment in cloud services, leaving more legacy technology in place and absorbing about 32% of the direct gain to depositors.

That is a modelled trade-off, not an observed result of a UK capital-policy change. The study simulates a one-percentage-point reduction in effective capital requirements for every institution in its deposit-market model. Its result is also not an official Bank position: staff working papers represent their authors, not the Bank or its policy committees.

For policymakers and bank technology chiefs, the point is narrower. Capital affects more than funding costs. By changing the value placed on future profits, it may influence whether a difficult modernisation programme is funded now or deferred.

What the 32% measures

The Bank staff working paper, published on 14 August 2026 and written by Alvaro Contreras, Peter Eccles and Paolo Siciliani, combines confidential outsourcing contracts with regulatory balance-sheet data.

Its starting dataset has 3,443 material outsourcing contracts involving 90 UK domestic banks over two decades. The authors identify 457 external cloud-service contracts, worth £1.57 billion a year across 72 banks, then merge contract start dates with the Historical Banking Regulatory Database. The main bank-quarter panel runs from the first quarter of 2010 to the first quarter of 2024.

There is a selection limit. The researchers observe contracts that were still in place in 2024, not those that had ended or were not renewed earlier. The study therefore cannot offer a complete history of failed migrations or abandoned suppliers.

In the reduced-form results, a one-percentage-point increase in capital requirements is associated with about 4.2% more cloud-service-provider spending, but that pooled coefficient is significant only at the 10% level. In the interacted specification, the capital response is highly significant for large institutions and statistically insignificant for small and medium banks and building societies. Separately, a 10% increase in cloud spending is associated with roughly 1.2% lower non-interest expenses, 1% lower staff expenses and 0.3% more deposits. Those cost and deposit associations are also concentrated among large institutions. None of these associations proves causation.

The paper interprets the pattern through a charter-value channel. A better-capitalised bank may be more likely to survive and realise future returns, giving it a stronger incentive to absorb the upfront disruption of replacing legacy systems. Other explanations remain possible. Supervisory attention, existing modernisation plans or a bank's financial condition could affect both its capital requirement and its technology budget.

The 32% figure comes from the next stage, a structural model of deposit competition. Across 56 quarters from 2010Q1 to 2023Q4, the direct funding-cost effect of a one-point capital reduction raises the model's depositor-welfare measure by 0.57 basis points. The predicted fall in cloud investment subtracts 0.18 basis points, leaving a positive net effect of about 0.38 basis points. In the model's aggregate equilibrium, average quarterly depositor welfare still rises by £0.24 billion, or 2.53%. The exercise maps the interacted reduced-form estimates into predicted cloud spending. The paper does not report alternative supplier-pricing, switching-cost, outage or capital-response sensitivities for this counterfactual.

Capital relief still wins in that exercise. It just wins by less.

Large-bank evidence, smaller-bank benefit

The distribution is as important as the total. The observed links between cloud spending, costs and deposits are strongest at large institutions. In the baseline structural demand model, the cloud effect is small and statistically insignificant for large institutions, but positive and significant for small and medium banks and building societies. With standard errors clustered by bank, the small and medium bank coefficient remains positive and significant at the 10% level. The building-society coefficient stays positive but becomes statistically imprecise.

The authors reconcile the split by arguing that large banks use cloud chiefly to reorganise internal operations, while smaller lenders use it to close service-quality gaps. When the model freezes cloud spending at 2015 levels, average inside-market concentration rises by 132 points on the Herfindahl-Hirschman Index and depositor welfare falls by 11.2%. The proportional losses are greatest at smaller banks and building societies.

That result should interest a building-society strategist, but it is not a promise that buying more cloud capacity will win deposits. Small and medium building societies had adopted cloud less intensively in the descriptive data, even while the model assigns them the largest marginal demand benefit. Governance, execution and the choice of workload still stand between potential and outcome.

The cloud market is not a neutral input

The model treats cloud spending as a bank characteristic that can improve depositor demand. It does not model the full dynamic investment decision or place internal cost savings from legacy replacement directly into the deposit-franchise profit calculation.

It also stops short of pricing the structure of the cloud-supplier market. The Competition and Markets Authority's final decision of 31 July 2025 found that Microsoft and Amazon Web Services each held 30% to 40% of UK and European Economic Area infrastructure-as-a-service supply in 2024. Fewer than 1% of customers switched provider each year. The CMA found that egress fees and technical differences impeded switching and use of multiple providers, particularly for smaller customers. It said committed-spend agreements were widespread, although it did not find that they harmed competition in their current form.

The position has since moved. On 31 March 2026, the CMA said Microsoft and Amazon had set out material steps to lower egress fees and improve interoperability. It also said further work was needed on multi-cloud and switching, and planned to review progress after six months.

Those frictions do not overturn the Bank paper's result. They narrow what it can establish. A migration may lower a bank's fixed technology costs while increasing dependence on one supplier, adding exit expense or concentrating outage exposure. A depositor-welfare measure based on rates and perceived service quality is not a complete measure of those operational risks.

For a prudential policy lead, the next test is a sensitivity exercise: vary the capital-to-cloud response, supplier pricing, switching costs and disruption assumptions before using 32% in a policy case. For a bank chief information officer, the practical task is to separate programmes that remove legacy cost from feature spending that can be paused, then map both against capital scenarios and supplier-exit plans.

The paper does not show that lower capital requirements are a mistake. It shows why funding relief cannot be evaluated on funding costs alone. If lighter capital makes long-horizon technology investment easier to defer, part of today's gain may arrive with tomorrow's legacy system still attached.