在很多實務分析中變數之間都時常有分組的效應，例如生物、金融計量經濟學。在此研究，我們的目標是在Granger因果模型下，從時間面板數據構建網絡結構，並假設其節點之間有分組結構。我們使用正規化分組的Lasso回歸，並使用調整參數去避免分組錯誤。我們通過模擬探討網絡結構的調整參數效應。 最後，應用我們的方法去研究金融機構真實數據。

The group effect between variables arises frequently in the real data analysis (e.g., biological, financial econometric). In this study, we aim to build network structure from temporal panel data under the framework of Granger causal models with inherent grouping structure among its nodes. We use a group lasso regression regularization framework and use tuning parameters to avoid group misspecifications. We investigate the tuning parameter effect on estimation of the network structure via simulation study. Finally, we apply the method to study the network structure of financial institutes.

論文審定書 i
誌謝 ii
Abstract iv
1 Introduction 1
2 Literature Review 3
2.1 Granger causality 3
2.2 Network Granger causal (NGC) estimates with group sparsity 5
3 Compare the package in R 9
3.1 Lasso package 9
3.2 Group Lasso package 13
4 Group effect 16
4.1 Performance metrics 16
4.2 Parameter choosing 17
4.3 Simulation study 17
5 Empirical study 28
6 Conclusion 34

Basu, S., Shojaie, A. and Michailidis, G. (2015). Network granger causality with inherent grouping structure. Journal of Machine Learning Research, 5, 417--453.

Breheny, P., and Huang, J. (2009), Penalized methods for bi-level variable selection, Stat. Interface, 2, 369--380.

Granger, J. (1969). Investigating causal relations by econometric models and cross-spectral
methods. Econometrica,37, 424--438.

Huang, J. and Zhang, T. (2010), The benefit of group sparsity. tAnn. Statist, 38, 1978--2004.
L"{u}tkepohl, H. (2005). New introduction to multiple time series analysis. Springer.

FDIC. http://www.fdic.gov