Expectile Risk Forecasts: New Backtests & Evaluation

The world of financial risk modeling is quietly undergoing a shift. While Value-at-Risk (VaR) and Expected Shortfall (ES) have long been the industry standards, their limitations are increasingly apparent – and regulators are paying attention. A new study published in Risk Sciences proposes a more robust alternative, focusing on ‘expectiles,’ and, crucially, provides the backtesting tools needed to validate their use. This isn’t just an academic exercise; it signals a potential move towards more reliable risk assessments, particularly as financial markets become more volatile and complex.

For years, financial institutions have relied on VaR and ES to quantify potential losses. However, VaR fails to account for the severity of losses beyond a certain threshold (it’s not ‘coherent’), while ES, though more comprehensive, lacks a clear statistical foundation for reliable elicitation. Expectiles, in contrast, offer a statistically sound and logically consistent framework. The problem? Until now, verifying the accuracy of expectile forecasts – ‘backtesting’ – has been challenging. Existing methods suffered from inaccuracies and a lack of statistical power.

The research team, spanning Canada and the UK, tackled this problem head-on. They developed new backtests that dissect the performance of expectile forecasts into two key components: ensuring the forecasts are generally accurate (unconditional coverage) and confirming that forecast errors aren’t predictably correlated over time (independence). Their approach cleverly combines Wald-testing with Box–Pierce-style autocorrelation testing, improving the reliability of the results. The image accompanying the release shows QQ plots comparing empirical distributions to expected chi-square distributions, visually demonstrating the tests’ performance.

The Forward Look

This study isn’t likely to trigger an immediate overhaul of risk management systems. However, it represents a significant step forward. Expectiles, backed by robust backtesting methodologies, are poised to gain traction, particularly among institutions seeking to enhance their risk modeling capabilities and satisfy increasingly stringent regulatory requirements. The caveat, as noted by the authors, is the reliance on a ‘location-scale’ framework for independence testing. This means the tests may be less effective in environments characterized by stochastic volatility – a common feature of modern financial markets. Therefore, the next phase of research will likely focus on extending these backtesting methods to accommodate more complex data-generating processes. We can anticipate further refinement of these techniques, potentially leading to the development of standardized expectile-based risk reporting within the next 3-5 years. The pressure from regulators, combined with the inherent advantages of expectiles, makes this a trend worth watching closely.