Network Estimation for Time Series

Matteo Barigozzi and Christian Brownlees have a fascinating new paper, "Network Estimation for Time Series" that connects the econometric time series literature and the statistical graphical modeling (network) literature. It's not only useful, but also elegant: they get a beautiful decomposition into contemporaneous and dynamic aspects of network connectedness. Granger causality and "long-run covariance matrices" (spectra at frequency zero), centerpieces of modern time-series econometrics, feature prominently. It also incorporates sparsity, allowing analysis of very high-dimensional networks.

If I could figure out how get LaTeX/Mathjax running inside Blogger, I could show you some details, but no luck after five minutes of fiddling last week, and I haven't yet gotten a chance to return to it. (Anyone know? Maybe Daughter 1 is right and I should switch to WordPress?) For now you'll just have to click on the Barigozzi-Brownlees paper above, and see for yourself.

It's interesting to see that Granger causality is alive and well after all these years, still contributing to new research advances. And Barigozzi-Brownlees is hardly alone in that regard, as the recent biomedical imaging literature illustrates. Some of Vic Solo's recent work is a great example.

Finally, it's also interesting to note that both the Barigozzi-Brownlees and Diebold-Yilmaz approaches to network connectedness work in vector-autoregressive frameworks, yet they proceed in very different, complementary, ways.