I have a question regarding the statistical assumptions behind Pyth’s confidence intervals for a Hayashi-Yoshida covariance estimator I’m building.
I am modeling “surprisal bits” using the KL divergence between consecutive (μ, σ) updates to construct information bars (accumulating some given threshold). Is it safe to strictly assume the provided confidence interval maps to a Laplace distribution (at 95% confidence) for this purpose?
Would appreciate any feedback from the math/architecture team