By giving final approval to Basel III capital standards this week, U.S. regulators have inadvertently exposed the financial system to massive model risk.

Implementing robust capital standards that give individual institutions sufficient buffers from extreme events and protect the system at-large has been a major challenge for regulators and the Basel Committee since the inception of risk-based capital charges years ago. However, overreliance on analytic methods that failed miserably during the crisis puts the entire system at risk while creating enormous burdens on institutions and regulators to closely oversee these models.

The Basel capital rules feature essentially two types of requirements: leverage ratios that do not make adjustments for specific risk types; and risk-based standards that do account for such differences. In the past, critics of simple leverage ratios have called out the potential for such measures to be overly simplistic. Without differentiating, risk leverage ratios may artificially lead to market distortions and misallocations of capital across sectors. Eventually, Basel got around to adding risk-based standards for credit risk and then components for market and even operational risk. The concept of imposing a set of risk-based capital standards spanning these major risk types is sensible in theory, but turns out to be extraordinarily cumbersome at best and systemically risky at worst in practice.

Tagging Basel III as overly complex is hardly a revelation. Prominent regulators such as Andrew Haldane of the Bank of England and Thomas Hoenig, a director at the FDIC, have been vocal critics of Basel III's overengineering for some time. Both have been proponents of leaning more on leverage ratios than on the risk-based capital standards driven by complex models that played a role in vastly underestimating the buildup of risks before the crisis and remain prone to a host of governance and technical weaknesses. Three areas that highlight Basel's overreliance on models are treatment of operational risk; the use of value-at-risk models for determining capital requirements; and counterparty valuation adjustment computations for derivatives.

Operational risks stemming from breakdowns attributed to people, process and technology have become increasingly apparent in banking over the years. Measuring a bank's average exposure to such risks does not easily lend itself to strict quantification due to inherent challenges in measuring events that occur very infrequently but generate large losses when they happen. However, statistically-based models are used in estimating Basel operational risk capital under the advanced measurement approach for the largest banks. Generating loss distributions, which describe the range of potential losses that a bank could experience, from inherently stable internal data sources is challenging enough. It becomes almost a form of alchemy when applied against noisy operational risk data sets. Applying the latest statistical techniques to operational risk measurement lulls us into a false sense of analytic security and desensitizes management to the important qualitative aspects of controlling these risks.

Basel also permits the application of VaR models – notorious for vastly underestimating bad scenarios during the mortgage crash – in estimating credit, market and even operational risk. Even today issues with VaR models remain, as evidenced by last year's JPMorgan Chase derivatives trading incident, where changes in VaR models and errors in measurement of trading losses contributed to misidentification of emerging risk. VaR models and extensions of these methods (referred to as stress VaR models) heavily depend on stable data and assumptions about model inputs. Moreover, there is a tendency to become overly enamored with modeled outcomes given their intellectual rigor and empirical basis – but that doesn't make them right.

Another example of misplaced emphasis on models occurs in estimating counterparty risk for derivatives activities, or CVA calculations. The financial crisis underscored the need to strengthen counterparty risk assessment. However, this is an area where models replete with mathematical elegance are notoriously sensitive to a number of critical assumptions and data. And if this weren't enough, VaR for CVA is allowed under advanced measurement methods for Basel – further subjecting the CVA results to inherent flaws in estimating extreme loss events using VaR techniques.

Unfortunately, despite impressive advances in computational and technical risk estimation, our ability to accurately measure complex risks over extreme outcomes remains elusive in many instances. By no means should we cease research and development on risk quantification. However, despite all the advances in modeling, we need to admit that it is simply not practical to cement these technologies in bank capital rules.

Rather, as Hoenig and Haldane suggest, getting back to basics until then with simple leverage ratios, as imperfect as they may be, is the most palatable solution. Standards approved by U.S. regulators and set up in part to mitigate systemic risk may actually wind up increasing it under Basel III.

Clifford Rossi is the Executive-in-Residence and Tyser Teaching Fellow at the Robert H. Smith School of Business at the University of Maryland.