Value at risk (VaR) metrics are designed to manage and measure the risk of extreme losses. VaR can be calculated based on either a historical track record or a model with such assumptions as asset volatility and correlation. The Basel Accord rules permit banks to develop their own internal VaR models, and there is no industry standard method, which allows banks to use their own discretion in setting up a model and choosing assumptions.

Regulators are aware of the risk of banks’ understating their true risk exposure. Therefore, regulators use backtesting to compare reported numbers with actual performance to detect exceptions. A high number of exceptions can result in punishment, and banks face a trade-off between saving capital by underreporting risk on the one hand and opening themselves up to the possibility of regulatory punishment on the other.

Regulators classify banks by the number of exceptions per quarter: green (0–4 exceptions), yellow (5–9 exceptions), and red (10 or more exceptions). Higher capital charges and more oversight are applied to banks with greater numbers of exceptions. The authors show that the risk of underreporting risk is highest for banks with large trading books and with poor last-quarter stock returns.

Previous researchers have highlighted the fact that banks underreport their credit risk and that regulators need to set appropriate levels of punishment in an effort to limit this risk. The authors’ research is novel because it focuses on both equity capital and the trading book, and it considers the effects of regulatory punishment.

This research is especially interesting for investment professionals working in regulations and those in investment analysis. The former will become more aware of the shortcomings of self-reported metrics and more aware that other metrics and information will be required to avoid future crises. When analyzing company stocks and other instruments, investment analysts will be more aware that the results of VaR models depend greatly on the assumptions used when making comparisons between outcomes.

The authors use a dataset from 2002 to 2013 consisting of large financial institutions from the United States, Canada, and Europe. They select a list of 16 sample banks from an initial list of 70 banks. VaR is reported at a 99% level. The total sample thus consists of 497 quarterly data points. The authors also define an expanded sample set of 638 data points, including three broker/dealers and banks with a 95% VaR level. Marginal expected shortfall is used to measure systematic stress levels.

Sample statistics show that the dataset is heterogeneous, consisting of banks with different levels of equity capital and VaR exceptions. Hence, the sample should provide a good dataset for further analysis. The authors also adjust bank 1-day VaR levels to regulatory 10-day levels.

The authors show that banks in the green zone have an incentive to stay there and report conservatively, whereas banks in the yellow zone have less to lose and hence a higher tendency to underreport risks than green-zone banks. The difference in misreporting between these two groups is significant. The results of the cross-sectional analysis show that banks with lower equity capital and lower stock returns have more future exceptions. According to the authors’ time-series analysis, the self-reported risk measures are least accurate when they are most relevant for regulators: when systemic risk levels are high and bank capital levels are low. Lastly, the authors perform some additional robustness checks and include the 95% VaR data; their results remain the same.

The importance of risk management has increased significantly since the 2008 global financial crisis. The focus has shifted from average or media scenario projections to the limitation of extreme losses because such losses pose the greatest threat to the overall financial system’s stability. VaR has been a central risk measure in both the banking and insurance industry (the Solvency II Directive). Because limited data are available for extreme events, however, modeling such scenarios is challenging; regulators’ assumptions for these kinds of events are based on only a few data points. Comparing institutions is also ever-more challenging because the advanced models differ by company.

This research sheds some doubt on the idea that more-advanced models give better insight into an institution’s risk behavior. If banks and other financial institutions can set assumptions based on short-term capital relief goals instead of true risk behavior, reporting metrics will not provide regulators with tools to monitor and punish risky behavior before the event.

The authors’ findings are highly relevant now that some time has passed since the most recent economic crisis. Professionals in the financial sector should be aware that self-reported metrics might not provide true insight into real exposure and that other public metrics might provide more insight into risk behavior. Regulators might want to consider different levels of punishment depending on economic cycles, promoting true reporting numbers when they are needed the most.