We develop metrics based on Shapley values for interpreting time-series forecasting models, including “black-box” models from machine learning. Our metrics are model agnostic, so that they are applicable to any model (linear or nonlinear, parametric or nonparametric). Two of the metrics,...