Talk Title: Structured Regularization for Large Vector Autoregressions
The vector autoregression (VAR) is a natural multivariate extension of the Box/Jenkins autoregressive framework for univariate time series. Popularized by Sims (1980), the VAR has proven very successful in macroeconomic forecasting and inference. However, due to its heavy parameterization, it is ill-suited for use with high-dimensional time series. Using regularization techniques influenced by Tibshirani (1996), I will present several methods which take into account both the time dependence and the structure of the VAR to allow for the efficient estimation of high dimensional time series. Joint work with David Matteson and Jacob Bien.