Inference for the difference of two independent KS Sharpe ratios under lognormal returns (Q826359)

In this very short paper the authors study the problem of comparing the performance of two different investments. In order to do so they use as a tool an extension of the Sharpe ratio (SR). This ratio is, by definition, \(SR=(E(r_t)-r_f)/\sqrt{var(r_t)}\) where \(r_t=\log(g_t)\) and the net return is, by definition, \(g_t-1\). Instead the authors use what is known as the Knight and Sathell extension of the Sharpe ratio (KS) where, by definition, \(KS=E(g_t)/\sqrt{var(g_t)}\). Under log-normal assumption for the returns it is very simple to show that \(KS=1/\sqrt{e^{\sigma^2}-1}\) and they focus on the problem when there are only two investments. In this case the inferential procedure is not trivial hence they do apply the Fraser and Reid's method which gives for the modified signed log-likelihood ratio statistics an asymptotically normally distributed third order accuracy. While the paper offers a very clear presentation of the result it is not very generous about details and some are omitted. The paper offers also some simulations and only a selection of these simulations are reported in the paper but, for the interested reader, more are available from the correspondent author. Based on the simulations, they do advocate the use of this third-order likelihood-based method for the difference between two Knight and Sathell extensions of the Sharpe ratio (KS). My only very minor criticism of this paper is related to some of the presentation choices made by the authors. Maybe it would be helpful to provide more details and to present a more self-contained but longer paper. While this does not change the value of the paper in my opinion it would help the reader quite a bit. I wish to add that the paper is clear and a good read.