Bayesian Imputation with Optimal Look-Ahead-Bias and Variance Tradeoff

TL;DR

This paper introduces a Bayesian imputation method that balances look-ahead bias and variance in time-series data, improving downstream task performance such as portfolio allocation.

Contribution

It proposes a novel Bayesian consensus posterior framework that optimally fuses multiple posteriors to manage bias-variance trade-offs in imputation.

Findings

Enhanced portfolio allocation accuracy with missing data.

Effective bias-variance trade-off optimization in imputation.

Demonstrated improvements in empirical and simulation studies.

Abstract

Missing time-series data is a prevalent problem in many prescriptive analytics models in operations management, healthcare and finance. Imputation methods for time-series data are usually applied to the full panel data with the purpose of training a prescriptive model for a downstream out-of-sample task. For example, the imputation of missing asset returns may be applied before estimating an optimal portfolio allocation. However, this practice can result in a look-ahead-bias in the future performance of the downstream task, and there is an inherent trade-off between the look-ahead-bias of using the entire data set for imputation and the larger variance of using only the training portion of the data set for imputation. By connecting layers of information revealed in time, we propose a Bayesian consensus posterior that fuses an arbitrary number of posteriors to optimize the variance and look-ahead-bias trade-off in the imputation. We derive tractable two-step optimization procedures for finding the optimal consensus posterior, with Kullback-Leibler divergence and Wasserstein distance as the dissimilarity measure between posterior distributions. We demonstrate in simulations and in an empirical study the benefit of our imputation mechanism for portfolio allocation with missing returns.