Factor-augmented tree ensembles

TL;DR

This paper introduces factor-augmented time-series regression trees that incorporate latent factors to handle complex predictor issues and domain knowledge, demonstrating their effectiveness in macro-finance applications like analyzing the lead-lag effect between equity volatility and the business cycle.

Contribution

It extends time-series regression trees by integrating latent stationary factors, enabling better handling of measurement errors, non-stationarity, and domain-informed modeling.

Findings

Ensembles of these trees reliably model macro-finance phenomena.

The approach effectively captures lead-lag relationships in economic data.

Method improves robustness to irregularities like missing data.

Abstract

This manuscript proposes to extend the information set of time-series regression trees with latent stationary factors extracted via state-space methods. In doing so, this approach generalises time-series regression trees on two dimensions. First, it allows to handle predictors that exhibit measurement error, non-stationary trends, seasonality and/or irregularities such as missing observations. Second, it gives a transparent way for using domain-specific theory to inform time-series regression trees. Empirically, ensembles of these factor-augmented trees provide a reliable approach for macro-finance problems. This article highlights it focussing on the lead-lag effect between equity volatility and the business cycle in the United States.