A Randomized Missing Data Approach to Robust Filtering and Forecasting

TL;DR

This paper introduces a randomized missing data approach for robust filtering and forecasting in state-space models, which enhances robustness to outliers and model misspecification by selectively using a subset of measurements.

Contribution

The paper develops a novel randomized missing data framework with endogenous and exogenous implementations, extending bootstrap aggregation to time-series filtering and forecasting.

Findings

Consistent improvement in inflation trend extraction despite outliers.

Robust filtering performance across various state-space models.

Effective trade-off control between robustness and efficiency.

Abstract

We put forward a simple new randomized missing data (RMD) approach to robust filtering of state-space models, motivated by the idea that the inclusion of only a small fraction of available highly precise measurements can still extract most of the attainable efficiency gains for filtering latent states, estimating model parameters, and producing out-of-sample forecasts. In our general RMD framework we develop two alternative implementations: endogenous (RMD-N) and exogenous (RMD-X) randomization of missing data. A degree of robustness to outliers and model misspecification is achieved by purposely randomizing over the utilized subset of data measurements in their original time series order, while treating the rest as if missing. The arising robustness-efficiency trade-off is controlled by varying the fraction of randomly utilized measurements. Our RMD framework thus relates to but is different from a wide range of machine learning methods trading off bias against variance. It also provides a time-series extension of bootstrap aggregation (bagging). As an empirical illustration, we show consistently attractive performance of RMD filtering and forecasting in popular state space models for extracting inflation trends known to be hindered by measurement outliers.