Staff Report, Federal Reserve Bank of New York 288
This paper introduces a generalized approach to canonical regression, in which a set of jointly dependent variables enters the left-hand side of the equation as a linear combination, formally like the linear combination of regressors in the right-hand side of the equation. Natural applications occur when the dependent variable is the sum of components that may optimally receive unequal weights or in time series models in which the appropriate timing of the dependent variable is not known a priori. The paper derives a quasi-maximum likelihood estimator as well as its asymptotic distribution and provides illustrative applications.
linear regression, time series, canonical correlations