Inference in $\alpha$-Brownian bridge based on Karhunen-Lo\`eve expansions

TL;DR

This paper develops a method to perform hypothesis testing on the scaling parameter of the alpha-Brownian bridge using Karhunen-Loève expansions generalized to finite measures, enabling distribution calculations of likelihood ratios.

Contribution

It generalizes the Karhunen-Loève theorem to finite measures and applies it to derive the distribution of a quadratic form for inference in alpha-Brownian bridges.

Findings

Derived the Karhunen-Loève expansion for alpha-Brownian bridge under finite measures.

Provided a method to compute the distribution of the likelihood ratio statistic.

Enabled hypothesis testing for the scaling parameter in alpha-Brownian bridges.

Abstract

We study a simple decision problem on the scaling parameter in the $\alpha$-Brownian bridge $X^{(\alpha)}$ on the interval $[0,1]$: given two values $\alpha_0, \alpha_1 \geq 0$ with $\alpha_0 + \alpha_1 \geq 1$ and some time $0 \leq T \leq 1$ we want to test $H_0: \alpha = \alpha_0$ vs. $H_1: \alpha = \alpha_1$ based on the observation of $X^{(\alpha)}$ until time $T$. The likelihood ratio can be written as a functional of a quadratic form $\psi(X^{(\alpha)})$ of $X^{(\alpha)}$. In order to calculate the distribution of $\psi(X^{(\alpha)})$ under the null hypothesis, we generalize the Karhunen-Lo\`eve Theorem to positive finite measures on $[0,1]$ and compute the Karhunen-Lo\`eve expansion of $X^{(\alpha)}$ under such a measure. Based on this expansion, the distribution of $\psi(X^{(\alpha)})$ follows by Smirnov's formula.