Empirical Performance of Backtesting Methods for Expected Shortfall

Three approaches have been recently proposed in the literature for backtesting Expected Shortfall, each with the advantage of being easy to implement using information readily available from a typical risk management system. The practitioner, however, is left with little information about their relative performance, weaknesses or strengths. This study sets a common basis for comparing all three approaches, five methods overall. One year, or T = 252 days of information is shown adequate to compare the methods using idealized portfolios. The comparison is extended to two market portfolios, two distinct economic periods, and two forecast calculation methods. The methods produce majority decisions about acceptance or rejection of Expected Shortfall coverage in all cases, with the Costanzino & Curran and Acerbi & Szekely Z2 methods always in that majority. Performance of the three approaches is similar, lending support for the practical decision to select a method on the basis of ease of implementation and computational cost