A streaming algorithm for bivariate empirical copulas

TL;DR

This paper introduces a space-efficient streaming algorithm that approximates the bivariate empirical copula function with guaranteed error bounds, enabling scalable dependence modeling in multivariate data streams.

Contribution

It adapts the Greenwald and Khanna algorithm to efficiently approximate bivariate empirical copulas in streaming data with theoretical error guarantees.

Findings

Algorithm provides accurate approximations with bounded error.

Computational efficiency is demonstrated both theoretically and numerically.

Method enables scalable dependence analysis in high-dimensional data streams.

Abstract

Empirical copula functions can be used to model the dependence structure of multivariate data. The Greenwald and Khanna algorithm is adapted in order to provide a space-memory efficient approximation to the empirical copula function of a bivariate stream of data. A succinct space-memory efficient summary of values seen in the stream up to a certain time is maintained and can be queried at any point to return an approximation to the empirical bivariate copula function with guaranteed error bounds. An example then illustrates how these summaries can be used as a tool to compute approximations to higher dimensional copula decompositions containing bivariate copulas. The computational benefits and approximation error of the algorithm is theoretically and numerically assessed.