Demand forecasting, predictive maintenance for IoT networks or stock price prediction are examples of systems that require a joint modelling of multiple time-dependent variables. Fitting of multivariate time series models with onventional training methods may result in suboptimal models and forecasts.