On quantifying the role of exogenous macro-economic information in machine learning for modelling financial data