A0896
Title: CP factorization for tensor-variate time series
Authors: Long Yu - Shanghai University of Finance and Economics (China) [presenting]
Abstract: Factor structure is used to model tensor-variate time series based on CP decomposition. The target is to identify the factor loadings up to sign change. The proposed procedure relies on eigen-analysis with a normalized and truncated auto-cross covariance. The estimator's accuracy is studied under general conditions, which allows sparse or dense loading vectors and strong or weak factors. It is also shown how to de-bias the estimator so that limiting representation is available, which will be useful in related inference problems. To further reduce the estimation error, an iterative algorithm is provided, the convergence is theoretically justified, and the accuracy of the iterative estimator is improved.
