Measuring Market Risks in Banks
Can banks measure precisely the market risks that influence their position? Not necessarily, according to Christophe Pérignon, whose empirical studies into the matter point to a general tendency for banks to overstate the risks when the markets are stable and to underestimate them in periods of financial crisis.
With the increase in incidents of colossal losses suffered by certain banks, a need has grown since the 1990s for greater transparency in the banking world. Daily risk assessment was seen as a vital initiative to support the banks’ trading positions. And yet, ten years on, while risk is now a core concept in the financial system, Christophe Pérignon has observed that little has been done to analyze the quality of risk measurement systems. He decided to correct this situation and, in 2005, carried out a detailed analysis of data drawn from the risk management systems of numerous banks around the globe.
WHY MEASURE MARKET RISKS?
Since the amendment of the Basel Accord of 1996, financial authorities require the banks to put aside sufficient capital reserves to guard against both financial and market risks. Each bank needs to be able to assess the risks it faces on a daily basis. While the regulator imposes no specific risk assessment models, the banks must show that the models they use are reliable. This way, the market risk assessment systems used by the banks satisfy the financial regulator’s monitoring requirements. Moreover, in a highly competitive business environment, any bank that can measure the risks it faces gains a competitive edge, while demonstrating the dependability of the models it uses.
ARE THERE ANY IDEAL EVALUATION TOOLS?
Value at Risk (VaR) is currently the best system for measuring market risk. It refers to the maximum loss with a given confidence level (generally at 99%) in a set period of time (for example, in one day). Ideally, this model is designed to avoid exceptions, meaning days when the actual losses are superior to the estimated losses. From a more operational point of view, VaR is used when communicating to traders the maximum positions they are authorized to take. It’s the standard market risk measurement tool.
DOES THE RISK ESTIMATED BY BANKS CORRESPONDWITH ACTUAL RISK?
One of our studies involved examining six major Canadian banks between 1999 and 2005—representing a total of 7,354 trading days—with the intention of comparing their risk estimation and actual risk. We observed, in total, that of the 74 forecast exceptions (1%of 7,354), actual losses only exceeded estimated loss in two instances. This surprising result was backed up by the observations we gathered from studies carried out on other banks around the world (including the Bank of America, Deutsche Bank, and Société Générale) as they confirmed the banks’ tendency to overstate the risks they face. Paradoxically, most of the observers were expecting these financial institutions to underestimate their risks to reduce the capital reserve levels set by their regulators!
WHAT’S THE EXPLANATION FOR THIS RISK OVERSTATEMENT?
The overstatement of risks can be explained in three ways:
1.The portfolios held by the banks are so large and diversified that even if the measurement can be applied to each risk source, aggregation is a major headache.
2. As it is difficult to calibrate the models to actual data, the banks prefer to overstate the risks out of precaution: indeed, if the models used by the banks underestimate the risks, the regulator could demand that capital reserve levels be raised.
3. The bank’s risk management department sets the VaR; but the heavy responsibility that weighs on these players may lead them to exaggerate the risks in order to cover themselves against an elevated error rate.
WHAT IMPACT DOES THIS OVERSTATEMENT HAVE?
Don’t you find it worrying that banks find it difficult, impossible even, to apply a precise system for measuring the risks they face? The 2007-2008 subprime crisis highlighted real problems in their ability to evaluate and anticipate market shifts. It takes banks several weeks to assess actual losses, so there’s no way they can measure risk on a daily basis. For example, the risk management model used by UBS, the Swiss bank, generated 29 VaR exceptions in 2007, which greatly exceeded the anticipated 2 to 3 annual exceptions. The VaR could be likened to an airbag that works well when there aren’t any incidents, but which is useless when you most need it—in calm periods the reserves are too large, and during crises they’re insufficient. This has a major impact on the entire economy, because the central banks are obliged to intervene to meet the banks’ liquidity requirements. Today, all the risk measurement tools need to be structurally reformed. It’s time to re-examine the effectiveness of VaR; maybe it’s not the right standard. Maybe amore disaggregate form of risk calculation is required, that takes the volume of losses into account (and not just their frequency) and that considers the facility for unwinding positions. All these issues remain to be solved…
Based on an interview with Christophe Pérignon and on his articles “Do Banks Overstate their Valueat- Risk?1 " (Journal of Banking & Finance, 2008) and “The Level and Quality of Value-at-Risk Disclosure by Commercial Banks2 ” (presented at the 2008 Conference of the American Finance Association).
1 . C. Pérignon et al., “Do Banks Overstate their Value-at-Risk?,” Journal of Banking & Finance , 2008.
2 . C. Pérignon and D. Smith, “The Level and Quality of Value at- Risk Disclosure by Commercial Banks,” article presented during the Conference of the American Finance Association in January 2008.
The article “Do Banks Overstate their Value-at-Risk?” is based on the daily comparison of the VaR trading revenues (P&L) of six Canadian banks between 1999 and 2005, in which 7,354 trading days were analyzed. The data comes from the bank’s annual reports.
In “The Level and Quality of Value-at-Risk Disclosure by Commercial Banks,” Christophe Pérignon assessed the quality of the VaR communicated by a sample of sixty international banks over the period 1996 to 2005. Both articles can be downloaded at: www.hec.fr/perignon