Abstract
Stock prices should respond only to unpredictable components of economic news (‘innovations’) in efficient markets. While innovations used in empirical investigations of the economic underpinnings of stock market risk should at least satisfy this basic requirement, this may not guarantee satisfactory research results. Three methods of generating innovations are evaluated for a variety of economic variables. First differencing produces unsatisfactory, serially correlated innovations in general. Both ARIMA and Kalman Filter innovations are unpredictable, but in a further evaluation the component scores from Principal Components Analysis are regressed against economic innovations using PcGets. The results are far less noisy when Kalman Filter innovations are used.