Regulators rightly want to remedy a serious flaw in the financial system: The complexity of bank capital requirements has made them vulnerable to manipulation. In the rush to embrace simplicity, however, policy makers could inadvertently make safe investments unattractive for banks.

At issue is the risk-weighted capital ratio, a measure regulators have long used to assess banks’ soundness. Instead of simply dividing equity by total assets, it assigns each asset a weight that is supposed to correspond to its risk. The idea is that $10 in capital might be too little to absorb potential losses on $200 in subprime mortgage loans, but more than enough for the same amount in U.S. government bonds. The safer a bank’s assets are judged to be, the higher its ratio.

The approach hasn’t worked well, because the risks of some assets have been badly underestimated. That’s not surprising, given that regulators have often relied on banks to do the measurement using their own internal models. Bank executives typically prefer lower capital levels than regulators would judge sufficient, and thus are motivated to understate risks.

Regulators have incentive problems, too. It is politically incorrect to announce that the sovereign debts of some nations, especially their own, are riskier than others. Hence, regulators assign relatively undifferentiated and unrealistically low risk weights to government debt. This is a particular issue in the euro area, where banks have been encouraged by capital rules to load up on bonds issued by riskier countries such as Greece, Italy and Spain.

To mitigate the shortfalls of risk weighting, regulators are working on new leverage rules that would set a floor on capital as a percentage of assets, regardless of their riskiness. A proposed rule in the U.S. would set the minimum at 5 percent for large banking holding companies. Global regulators, under the auspices of the Basel Committee on Banking Supervision, are also considering a new international minimum.

The U.S. leverage rule would require some banks to raise more capital -- meaning that the new rule would replace the risk-weighted ratio as the binding constraint. Raising bank capital levels is a good idea, but doing it this way could have an unintended effect: Banks would be able to take on more risk for the same amount of capital merely by shifting to riskier assets.

It is strange for a bank to be told by regulators, “It doesn’t matter if you move out of U.S. Treasury bills into sketchy real estate loans, we will require you to have the same amount of equity to buffer your risk of loss.” This distorts incentives for investment and lending, and could lead to excessive risk-taking. This, in turn, could prompt regulators to react with further tightening of the leverage requirement, which would serve only to exacerbate the problem.

How, then, can capital requirements be reformed, in light of their complexity and past abuses? Should regulators throw up their hands and say they can’t distinguish risky assets from safe ones? Surely not. Instead, the risk-weighting process should be made more conservative.

An improved approach would recognize that, other things being equal, banks are likely to invest more heavily in assets with lower risk weights. This “piling on” can cause even a safe asset class to endanger a bank. Research I did with the mathematicians Amir Dembo and Jean-Dominique Deuschel shows that it is relatively difficult for an adequately capitalized bank to fail from many small, high-risk loans unless they have a tendency to go bad at the same time. Given a failure, the culprit is relatively likely to be losses on very large loans to borrowers that had been judged safe. During the subprime crisis and the euro area’s sovereign-debt crisis, large loans with very low risk weights quickly became life-threatening.

The lowest risk weights shouldn’t be as low as they are today. Also, as a bank’s investments become more concentrated in a given asset class, the risk weights for that asset class should go up. The same principle applies to the system as a whole. As investments by banks, in aggregate, become more concentrated in a given asset class, risk weights for that asset class should rise.

Assets whose risks are difficult to judge should be assigned higher risk weights. If a bank is relying on mathematical models to assess a potential loss, and if it doesn’t know which model gives the right answer, it should apply the one that is most relevant in extreme scenarios instead of averaging across them all. This is complicated and requires more effort on the part of regulators, but it can be done. When in doubt, regulators should err on the side of caution, and allow banks less leeway in the designs of their models.

Tossing out risk-weighted capital ratios in an effort to make regulation simpler would be a big mistake. It’s possible to lift capital in the banking system to safer levels without depending on rules that are blind to the types of risk that banks take.

(Darrell Duffie is the Dean Witter distinguished professor of finance at Stanford University. He is a member of the board of directors of Moody’s Corp., and has done paid consulting and speaking engagements for banks and other financial institutions.)

To contact the writer of this article: Darrell Duffie at duffie@stanford.edu.

To contact the editor responsible for this article: Mark Whitehouse at mwhitehouse1@bloomberg.net.