Model Risk and XVAs: Are Banks Underestimating Their True Exposure?
"A deep dive into how model inaccuracies impact valuation adjustments and capital reserves in financial institutions, revealing a potential need for more conservative risk management."
In the complex world of finance, banks rely on models to make critical decisions, from pricing derivatives to managing risk. But what happens when these models are wrong? Model risk, the potential for inaccuracies in these tools, can have a significant impact on a bank's financial health, particularly when it comes to valuation adjustments (XVAs).
A new research paper is shedding light on this often-underestimated threat. The paper revisits the concept of hedging valuation adjustment (HVA), originally designed to address dynamic hedging frictions like transaction costs. The authors argue that HVA should also account for model risk, ensuring that banks hold adequate reserves to cover potential losses stemming from inaccurate models.
The stakes are high, especially in the wake of financial crises that have exposed the weaknesses in risk management frameworks. As regulatory bodies increase scrutiny and capital requirements, banks need to take a closer look at how model risk is impacting their XVAs and overall capital reserves. Ignoring this critical factor could lead to underestimation of risk and potential financial instability.
Exposure and Uncertainty
Banks’ true exposure to model risk and XVAs is difficult to quantify consistently because it depends on assumptions about markets, counterparties, collateral, funding, and future behavior. Published assessments may therefore understate uncertainty when they rely on stable conditions or simplified inputs. The scale of the risk can vary materially as market conditions change, so reported figures should be interpreted cautiously.
Conventional Valuation Limits
Standard approaches typically combine valuation models with assumptions about default, funding, collateral, and market movements. These methods can provide a structured basis for estimating XVAs, but their results remain sensitive to data quality, calibration choices, and model scope. A method that appears robust under ordinary conditions may provide less reliable estimates during stressed or rapidly changing markets.
Models as Formal Frameworks
A foundational treatment of model theory defines a model, or structure, as an ordered pair consisting of a nonempty set and an interpretation function for the symbols of a language. This formal perspective highlights that models connect defined objects with rules for interpreting their relationships. Later institutional frameworks use the same broad idea of structured definitions and requirements: the CMMC Model incorporates FAR 52.204-21 and NIST SP 800-171 Rev. 2, while SAMHSA’s Model Definitions document seeks clarity, quality, and consistency across behavioral-health services. The CMS WISeR Model similarly tests technology-enhanced prior authorization and pre-payment medical review in limited geographic regions.
Understanding Cross Valuation Adjustments (XVAs)
Cross Valuation Adjustments (XVAs) are a suite of calculations that banks use to account for various risks associated with financial transactions. These adjustments ensure that the true economic value of a deal is reflected on the balance sheet, considering factors like counterparty credit risk, funding costs, and capital requirements. Major components of XVAs include:
- Credit Valuation Adjustment (CVA): This adjusts for the potential loss due to a counterparty's default. It considers the probability of default and the expected exposure to the counterparty.
- Funding Valuation Adjustment (FVA): This reflects the cost of funding uncollateralized trades. It accounts for the difference between the interest earned on assets and the interest paid on liabilities.
- Capital Valuation Adjustment (KVA): This represents the cost of holding capital against potential future losses. It acknowledges the economic impact of regulatory capital requirements.
Insurance Data-Security Governance
The NAIC Insurance Data Security Model Law provides a model framework for insurance data-security obligations. Its applicability language refers to a licensee’s state of domicile for an insurer and home state for a producer. This illustrates how model laws can translate broad governance objectives into jurisdiction-specific responsibilities. The framework is therefore relevant as an example of formalized controls, although it is not itself an analysis of bank XVA exposure.
Limits of Model Confidence
Models can improve consistency without eliminating uncertainty. Their outputs may diverge when institutions use different assumptions, data, or calibration methods. A model that performs adequately in ordinary conditions may also be less reliable when market behavior changes sharply. These limitations support treating model results as estimates rather than exact measures of exposure.
Comparing Model Results
Comparisons between banks should distinguish differences in methodology from differences in underlying exposure. Two institutions may report different XVA results because they use different scenarios, inputs, controls, or treatment of uncertainty. Without comparable assumptions and validation standards, apparent performance differences can be misleading. Cross-bank comparisons should therefore be read as indicative rather than definitive.
The Path Forward: Banning Bad Models
The research suggests a bold solution: instead of trying to manage model risk through complex reserves and adjustments, banks should focus on improving the quality of their models in the first place. According to research, poorly designed models, when adjusted for, require far more capital than properly designed models. Additionally, risk-adjusting poorly designed models can be very computationally exhaustive, so it makes the most sense to ban these models. While this may require more upfront investment in sophisticated modeling techniques, the long-term benefits of reduced risk and greater financial stability outweigh the costs. By prioritizing accuracy and reliability, banks can reduce their exposure to model risk and build a more resilient financial system.
A Cautious Interpretation
The central concern is not that models are unnecessary, but that their precision can be overstated. XVA estimates should be considered alongside assumptions, validation evidence, stress results, and expert judgment. A disciplined interpretation recognizes that model outputs are conditional representations of exposure. This is particularly important when institutions use them to support capital, pricing, or risk decisions.
Improving Model Resilience
Future progress is likely to depend on stronger data controls, more responsive validation, and better testing under adverse conditions. Institutions may also place greater emphasis on monitoring model drift as markets, products, and counterparty behavior change. Technology can support these activities, but automation does not remove the need for governance. The most credible advances will combine analytical sophistication with transparent oversight.
From Firm Risk to System Risk
Model risk can become systemic when many institutions rely on similar assumptions, data, or market conventions. In stressed conditions, that common dependence may produce correlated errors or synchronized adjustments. Differences in governance and validation can further complicate interpretation across firms. Managing the broader challenge therefore requires attention to both individual models and the wider financial system.
Judgment Behind the Numbers
People determine how models are designed, calibrated, approved, and used. Expert judgment remains important when data are sparse, markets are stressed, or model results conflict with observed conditions. Weak communication can cause decision-makers to mistake conditional estimates for objective facts. Clear escalation, challenge, and accountability are therefore essential to limiting the real-world consequences of model error.