Financial institutions today face mounting pressure to accurately forecast market risk, a necessity underscored by stringent regulatory mandates that require forecasts to be rigorously backtested against real-world outcomes. Traditional risk metrics, particularly Value-at-Risk (VaR) and Expected Shortfall (ES), dominate the landscape, but both come with inherent limitations that challenge their efficacy. VaR, while widely used, is criticized for lacking coherence as a risk measure, and ES, despite addressing some of VaR’s weaknesses, suffers from a lack of independent elicitability. This has spurred growing interest in expectiles, a class of risk measures that uniquely combines coherence with elicitability, yet their practical backtesting methods remain underdeveloped and comparatively immature.
In a pioneering study published in the journal Risk Sciences, an international team of researchers from Canada and the United Kingdom tackled this gap by proposing novel approaches to backtesting expectile forecasts. Their techniques aim to reconcile major shortcomings that plague traditional methods, including size distortion—where test statistics may yield misleading rejection rates—and low statistical power, which reduces confidence in test outcomes. Central to their innovation is a fresh conceptual framework that disentangles two core characteristics often evaluated separately in VaR and ES backtesting: unconditional coverage, which assesses whether the forecasts are accurate on average, and independence, which evaluates the absence of problematic time dependencies or serial correlation within forecast errors.
The team led by corresponding author Yang Lu introduced a simplified Wald-style test to assess unconditional coverage for expectiles. By focusing on a single unconditional expectation condition, this test mitigates distortion problems typical of more complex, multifaceted testing regimes. This singular focus improves the test’s reliability in finite samples, offering practitioners a powerful yet computationally tractable tool to judge whether expectile forecasts accurately represent risk without being clouded by noisy statistical artifacts.
Addressing the independence property required additional ingenuity. The researchers connected the general Wald-testing framework with the well-established Box–Pierce test, a method traditionally employed to detect autocorrelation in time series data. Under the assumption of a location-scale model specification for asset returns—a framework that describes returns as having a variable location (mean) and scale (volatility)—they constructed candidate sequences of identically and independently distributed (i.i.d.) variables. This construction allows the Box–Pierce test to be correctly applied to examine the independence of expectile forecast errors, even in the nuanced context of financial return modeling.
Extensive simulation studies conducted by the team evidenced that the proposed backtests offer promising finite-sample properties. These simulations demonstrate that the tests maintain correct sizes, meaning the probability of incorrectly rejecting accurate models is controlled, and they exhibit improved power, effectively identifying models where expectile forecasts fail either in coverage or independence. Moreover, the researchers applied their methodology empirically using S&P 500 return data to showcase practical utility and to emphasize the relevance of these methods in real-world financial markets, where accurate risk management is paramount.
Despite these advancements, the authors caution users regarding the assumption-laden nature of the independence test construction. The reliance on a location-scale framework may limit applicability in data environments characterized by alternative generating mechanisms, such as stochastic volatility models, where the return distribution parameters evolve unpredictably over time. In such settings, the independence tests may yield misleading inferences, necessitating further methodological refinement or alternative modeling assumptions to widen practical applicability.
This research marks a critical step toward advancing risk management practices by refining expectile backtesting, an area long overshadowed by VaR and ES despite expectiles’ theoretical appeal. By directly addressing the intertwined but distinct challenges of unconditional coverage and independence, the study lays the groundwork for more robust, interpretable, and actionable risk assessment tools. As financial institutions continue to grapple with increasingly complex and volatile markets, expectile-based backtests enhanced with these new methods could pave the way for more resilient and regulatory-compliant risk forecasting.
Yang Lu and colleagues emphasize the importance of integrating these novel tests into routine risk assessment protocols, noting that improved backtesting accuracy not only bolsters regulatory compliance but also enhances institutions’ internal risk controls. The technology promises to detect subtle deficiencies in forecasting models that might otherwise go unnoticed, thereby enabling preemptive adjustments that safeguard against unexpected market shocks.
Furthermore, this methodology invites future research avenues exploring how to adapt and extend these backtesting strategies for more general classes of stochastic models, especially those incorporating time-varying volatilities and nonlinear dependencies. Given the dynamic and multifaceted nature of modern financial markets, the refinement and expansion of expectile backtesting frameworks represent an urgent and fertile area for quantitative finance research.
In conclusion, the innovative approaches developed for expectile backtesting disentangle critical statistical properties in a manner that improves accuracy, power, and interpretability. This work is poised to influence not only academic research but also practical risk management, offering a valuable toolset for institutions aiming to meet and exceed the rigorous standards set forth by financial regulators worldwide. Expectile backtesting, long neglected, is finally gaining the attention and methodological sophistication it deserves—transforming the landscape of market risk measurement.
Subject of Research: Not applicable
Article Title: Backtesting expectile: Disentangling unconditional coverage and independence properties
Web References: http://dx.doi.org/10.1016/j.risk.2026.100051
Image Credits: de Ita Solis, J.A.; Lu, Y.; Mailhot, M.; Meng, X.
Keywords: Mathematics, Applied mathematics, Statistics, Algorithms
