Modelling and forecasting with robust canonical analysis: Method and application

Multivariate methods often serve as an intelligent way to study the relations between two data sets. When the number of variables in one or both data sets is large, which is usually the case, the correlation matrices of the data sets may be singular or ill-conditioned. When this happens the weights obtained by multivariate methods that require the inversion of the correlation matrices are not unique, or highly unreliable. Here we present and apply a robust estimation and forecasting method that does not require us to invert the correlation matrices. This method, which we call robust canonical analysis (RCA), is a straightforward extension of the simple covariance of two variables to two data sets. As an example we use the RCA method to estimate the relations between a set of measures that describe how the firm manages its relations with its customers, and a set of variables that describe the utility of information systems applications to the firm's operations.