臺灣博碩士論文加值系統

為了要了解金融市場的動態，尋找市場當中的潛在訊息是非常重要的，我們在這 從不同的面向切入來挖掘市場中的潛在資訊。本研究選取美國43 家公司的高頻財務資料，分別隸屬於 8 個不同的產業別。我們首先對每家公司的高頻財務資料生成變數，利用 Mixed Graphical Model 以及非負矩陣分解對變數進行重疊分群的探討。從中我們得到了數個影響力較大且可能帶有多種性質的變數。再者，我們討論公司之間的網絡關係，利用動態光譜聚類(dynamic covariate-assisted spectral clustering) 改善網絡原先的結構，並且尋找網絡中扮演重要角色的媒介，有趣的是，滿足這個條件的公司數量並不多。我們也使用兩種度量說明了在不同的交易日均有較優良的群體關係。此外，我們也觀察網絡結構變動的情形，並與真實事件做連結。

In order to understand the movement of the financial market, it is very important to look for the potential information in the market. Here, we explore the potential information in the market from different aspects. In this study, we select high-frequency financial data of 43 stock companies in the United States, which belong to 8 different industries. We first generate high frequency features for each stock company’s high frequency financial data, and use Mixed Graphical Model and non-negative matrix factorization to discuss overlap grouping for features. We have obtained several features that have great influence and may have various properties. Furthermore, we discuss the network relationship between stock companies, using dynamic covariate-assisted spectral clustering to improve the original structure of the network and find the medium that plays an important role in the network. Interestingly, there are not many stock companies that meet this condition. We also use
two metrics to show that there is a better group structure during the different trading days. In addition, we also observe the variant in the network structure and connect with real events.

論文審定書i
論文公開授權書ii
誌謝iii
摘要iv
Abstract v
1 Introduction 1
2 Methodology 2
2.1 High Frequency Feature Network Construction Method . . . . . . . . . . 2
2.1.1 Mixed Graphical Models . . . . . . . . . . . . . . . . . . . . . . 2
2.1.2 Non-negative Matrix Factorization for Overlap Grouping . . . . . 4
2.2 Stock Company Network Construction Method . . . . . . . . . . . . . . 5
2.2.1 OGA+HDIC+Trim VAR estimation . . . . . . . . . . . . . . . . 5
2.2.2 Similarity Matrix . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2.3 Dynamic CASC . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.4 Adapted DI-SIM. . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Measures of Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3.1 Density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3.2 Modularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3.3 Rand index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3 Data Description and Preprocessing 12
3.1 High Frequency Trading Data . . . . . . . . . . . . . . . . . . . . . . . 12
3.2 High Frequency Features . . . . . . . . . . . . . . . . . . . . . . . . . . 14
4 Empirical Studies 17
4.1 Network Analysis for High Frequency Features . . . . . . . . . . . . . . 17
4.2 Network Analysis for Stock Companies . . . . . . . . . . . . . . . . . . 21
4.2.1 Dynamic Network Construction . . . . . . . . . . . . . . . . . . 22
4.2.2 Clustering Analysis and Community Detection . . . . . . . . . . 24
4.2.3 The Interrelation of the Groups . . . . . . . . . . . . . . . . . . . 26
4.2.4 Bottleneck Communicator in the Director Network . . . . . . . . 28
5 Discussion and Conclusion 31
6 References 32

[1] Binkiewicz, N., Vogelstein, J. T. and K. Rohe (2017). Covariate-assisted Spectral Clustering. Biometrika, 104, 361377.
[2] Guo, L., Tao, Y. and Härdle, W.K. (2019). A Dynamic Network Perspective on Cryptocurrencies. arXiv preprint arXiv:1802.03708v4.
[3] Haslbeck, J. M. B. and Waldrop, L. J. (2019). mgm: Estimating Time-Varying Mixed Graphical Models in HighDimensional Data. arXiv:1510.06871v6.
[4] Hubert, L. and Arabie, P. (1985). Comparing partitions. Journal of Classification, 2:193–218.
[5] Ing, C. K. and Lai, T. L. (2011). A stepwise regression method and consistent model selection for high-dimensional sparse linear models. Statistica Sinica, 21, 14731513.
[6] Newman, M. E. (2006). Modularity and community structure in networks. Proceedings of the national academy of sciences, 103(23), 85778582.
[7] Rohe, K., Qin, T. and Yu, B. (2016). Coclustering directed graphs to discover asymmetries and directional communities. Proceedings of the National Academy of Sciences, 113(45), 1267912684.
[8] Wang, F., Li, T., Wang, X., Zhu, S. H. and Ding, C. (2011). Community discovery using nonnegative matrix factorization. Data Mining and Knowledge Discovery, pages 493–521.
[9] Wu, C. M. (2020). Dynamic Network and Clustering Analysis in Stock Market. Master thesis, National Sun Yat-Sen University.
[10] Yang, E., Baker, Y., Ravikumar, P., Allen, G. and Liu, Z. (2014). Mixed Graphical Models via Exponential Families. Proceedings of the Seventeenth International Conference on Artificial Intelligence and Statistics, PMLR, 33, 10421050.