Financial institutions routinely forecast market risk, and regulators require those forecasts to be backtested—checked against what actually happens. Two widely used risk measures, Value-at-Risk (VaR) and Expected Shortfall (ES), come with well-known drawbacks: VaR is not coherent, and ES is not independently elicitable. Expectiles have attracted growing interest because they are “the only law-invariant risk measure that is both coherent and elicitable”; however, their backtesting has received less attention.
In a new study published in Risk Sciences , a team of researchers from Canada and the UK developed new backtesting approaches for expectile forecasts designed to address problems in existing methods, including size distortion and low test power. The central idea is to disentangle two properties that are often assessed separately for VaR/ES models: unconditional coverage (whether the forecasts are right on average) and independence (whether forecast errors show problematic time dependence).
“We wanted to introduce novel expectile backtests with better size and power properties,” shares corresponding author Yang Lu. “For unconditional coverage, we used a simplified Wald-style test focusing on a single unconditional expectation condition to reduce distortion.”
For independence, the team linked the general Wald-testing framework to the popular Box–Pierce lack-of-autocorrelation test, and—under a location-scale specification of returns—construct candidate i.i.d. sequences that allow Box–Pierce-style tests to be applied appropriately.
“Simulation studies indicated promising finite-sample performance for the proposed tests, and an empirical application illustrates how the approach can be used on S&P 500 return data,” adds Lu. “Nonetheless, we noted that the independence-testing construction relies on a location-scale framework and may be less suitable under certain alternative data-generating mechanisms (e.g., stochastic volatility settings).”
###
Contact the author:
Yang Lu, Department of Mathematics and Statistics, Concordia University (Canada)
yang.lu@concordia.ca
The publisher KeAi was established by Elsevier and China Science Publishing & Media Ltd to unfold quality research globally. In 2013, our focus shifted to open access publishing. We now proudly publish more than 200 world-class, open access, English language journals, spanning all scientific disciplines. Many of these are titles we publish in partnership with prestigious societies and academic institutions, such as the National Natural Science Foundation of China (NSFC).
Risk Sciences
Computational simulation/modeling
Not applicable
Backtesting expectile: Disentangling unconditional coverage and independence properties
“The authors declare that they have no conflict of interest.”