Arbitrage of Forecasting Experts

Forecasting is an important task across several domains. Its generalised interest is related to the uncertainty and evolving structure of time series. Forecasting methods are typically designed to cope with temporal dependencies among observations. In this paper we propose a novel approach where several forecasting models are dynamically combined to make a predictions. We propose a dynamic ensemble method based on arbitrating. Arbitrating is a metalearning approach that combines the output of experts according to predictions of the loss that they will incur. Within the arbitrage method we present an approach for retrieving out-of-bag predictions that significantly improves its data efficiency. Finally, we propose a method for explicitly handling the interdependence between experts when aggregating their predictions. We present extensive empirical evidence for the competitiveness of the proposed method relative to state of the art approaches. Our method is publicly available as a software package.