Results don't match with other packages

[Deleted user]

Ruben, PerfomanceAnalytics in R gives different results than yours.

https://rdrr.io/github/braverock/PerformanceAnalytics/man/ProbSharpeRatio.html

I think the difference lies in this part:

psr = scipy_stats.norm.cdf((sr - sr_benchmark) / sr_std)

The R code seems to more accurately apply the formula.

sr_prob = pnorm(((sr - refSR)*((n-1)^(0.5)))/(1-sr*sk+(sr^2)*(kr-1)/4)^(0.5))

As per your article, those should be equivalent but there are differences in practice, and not sure whether this is because of numerical error approximations or something else