Multi-horizon time series forecasting poses fundamental challenges to machine learning and statistics with applications in many domains. In direct multi-horizon forecasting, the standard approach of neural nets is to either ...