國立高雄大學圖資館 |

A Dynamic Network Model With an Application to Interbank Market.

Record Type:	Electronic resources : Monograph/item
Title/Author:	A Dynamic Network Model With an Application to Interbank Market.
Author:	Zhang, Haici.
Published:	Ann Arbor : ProQuest Dissertations & Theses, 2023
Description:	189 p.
Notes:	Source: Dissertations Abstracts International, Volume: 85-03, Section: B.
Notes:	Advisor: Laux, Paul;Qian, Wei.
Contained By:	Dissertations Abstracts International85-03B.
Subject:	Finance.
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30574663
ISBN:	9798380375429

A Dynamic Network Model With an Application to Interbank Market.
Zhang, Haici.

A Dynamic Network Model With an Application to Interbank Market. - Ann Arbor : ProQuest Dissertations & Theses, 2023 - 189 p.

Source: Dissertations Abstracts International, Volume: 85-03, Section: B.

Thesis (Ph.D.)--University of Delaware, 2023.

This item must not be sold to any third party vendors.

Among central bankers and the wider academic community, interbank network analysis has received an increasing amount of attention over the past few years. In view of the high degree of interconnectedness of the financial interbank network, network theory provides a natural framework for the study of the robustness of the interbank system as a whole and its resilience to risk contagion. A good understanding of the nature of the interconnections among the interbank network and of how these interconnections respond to changes in market conditions is also vital for policymakers when they are designing regulations and monitoring systemic risk. Aiming at understanding the dynamic interconnectedness in the banking system, this dissertation proposes three dynamic network models for the e-MID electronic interbank network.In the first part of this dissertation, we propose a Bayesian dynamic interbank network model where three mechanisms control the likelihood of a link between two banks: (i) a time-series interbank activity index that expresses overall confidence of interbank across time, (ii) bank-specific latent variables describing banks’ tendency to be borrowers or lenders, and (iii) covariates characterizes pairwise past relationships. A large fraction of previous research on interconnectedness studies static and aggregated interbank networks which only reveal information about long-term connectedness (such as core-periphery structure) inside a network, and lack of studies exploring the dynamics of interbank networks. To address this research gap, we formulate a flexible dynamic interbank network model and design its novel sampling method for computational efficiency. We then demonstrate the superiority of our proposed method in a variety of areas. First, we use the proposed cross-sectional latent variables instead of general network topology statistics to explore whether these variables are representative of changes in network topology and can be used to predict macroeconomic variables that reflect the health of the banking system. Secondly, we evaluate the goodness-of-fit of the model with interbank link predictions and evaluate the results using AUC and different prediction errors. In addition, we propose two proxies for relationship lending based on information sharing, using the latent variables in our proposed model. Based on the regression results, the two proxies are significant during both the pre-crisis and crisis periods. To make the model to be more interpretive from the view of probability theory, rather than linking the binary data of the adjacency matrix to a latent variable with a probit link, we assume that the binary data has a Bernoulli distribution with a probability of success rate which is a logiit transformation of the three underlying trading mechanisms discussed in the first study. Though it makes the results more interpretive, more constraints and auxiliary variables are applied to make the estimation process converge.In the second part, we apply a GC-LSTM ( GCN units that are embedded into the LSTM framework) deep learning approach to model dynamic interbank networks. The deep learning model has to use in transportation (Lei et al. (2019)) and social networks (Martinez-Jaramillo et al. (2014)) but it has not been applied to any financial interbank transaction data. To see whether the deep learning model is good at capturing the spatial-temporal information of interbank network topology and making a good prediction, we apply this method to analyze the e-MID dataset. Compared with the statistical dynamic network methods, the deep learning method requires fewer assumptions and easier parameter estimation strategies. The results validate that our model outperforms the benchmarks in terms of AUC and PRAUC. Meanwhile, we also compare the results for crisis and pre-crisis periods, we find that the deep learning model is better than the traditional models in both crisis time and pre-crisis periods. In addition, the GC-LSTM model is better at predicting future links in the crisis period than the traditional statistical models which indicates that the model without statistical underlying assumptions is better at capturing structure change.

ISBN: 9798380375429Subjects--Topical Terms:

183252
Finance.
Subjects--Index Terms:

Bayesian sampling

A Dynamic Network Model With an Application to Interbank Market.
LDR:05410nmm a2200385 4500 001 655819
005 20240414211939.5
006 m o d
007 cr#unu||||||||
008 240620s2023 ||||||||||||||||| ||eng d
020 $a 9798380375429
035 $a (MiAaPQ)AAI30574663
035 $a AAI30574663
040 $a MiAaPQ $c MiAaPQ
100 1 $a Zhang, Haici. $3 966937
245 1 2 $a A Dynamic Network Model With an Application to Interbank Market.
260 1 $a Ann Arbor : $b ProQuest Dissertations & Theses, $c 2023
300 $a 189 p.
500 $a Source: Dissertations Abstracts International, Volume: 85-03, Section: B.
500 $a Advisor: Laux, Paul;Qian, Wei.
502 $a Thesis (Ph.D.)--University of Delaware, 2023.
506 $a This item must not be sold to any third party vendors.
520 $a Among central bankers and the wider academic community, interbank network analysis has received an increasing amount of attention over the past few years. In view of the high degree of interconnectedness of the financial interbank network, network theory provides a natural framework for the study of the robustness of the interbank system as a whole and its resilience to risk contagion. A good understanding of the nature of the interconnections among the interbank network and of how these interconnections respond to changes in market conditions is also vital for policymakers when they are designing regulations and monitoring systemic risk. Aiming at understanding the dynamic interconnectedness in the banking system, this dissertation proposes three dynamic network models for the e-MID electronic interbank network.In the first part of this dissertation, we propose a Bayesian dynamic interbank network model where three mechanisms control the likelihood of a link between two banks: (i) a time-series interbank activity index that expresses overall confidence of interbank across time, (ii) bank-specific latent variables describing banks’ tendency to be borrowers or lenders, and (iii) covariates characterizes pairwise past relationships. A large fraction of previous research on interconnectedness studies static and aggregated interbank networks which only reveal information about long-term connectedness (such as core-periphery structure) inside a network, and lack of studies exploring the dynamics of interbank networks. To address this research gap, we formulate a flexible dynamic interbank network model and design its novel sampling method for computational efficiency. We then demonstrate the superiority of our proposed method in a variety of areas. First, we use the proposed cross-sectional latent variables instead of general network topology statistics to explore whether these variables are representative of changes in network topology and can be used to predict macroeconomic variables that reflect the health of the banking system. Secondly, we evaluate the goodness-of-fit of the model with interbank link predictions and evaluate the results using AUC and different prediction errors. In addition, we propose two proxies for relationship lending based on information sharing, using the latent variables in our proposed model. Based on the regression results, the two proxies are significant during both the pre-crisis and crisis periods. To make the model to be more interpretive from the view of probability theory, rather than linking the binary data of the adjacency matrix to a latent variable with a probit link, we assume that the binary data has a Bernoulli distribution with a probability of success rate which is a logiit transformation of the three underlying trading mechanisms discussed in the first study. Though it makes the results more interpretive, more constraints and auxiliary variables are applied to make the estimation process converge.In the second part, we apply a GC-LSTM ( GCN units that are embedded into the LSTM framework) deep learning approach to model dynamic interbank networks. The deep learning model has to use in transportation (Lei et al. (2019)) and social networks (Martinez-Jaramillo et al. (2014)) but it has not been applied to any financial interbank transaction data. To see whether the deep learning model is good at capturing the spatial-temporal information of interbank network topology and making a good prediction, we apply this method to analyze the e-MID dataset. Compared with the statistical dynamic network methods, the deep learning method requires fewer assumptions and easier parameter estimation strategies. The results validate that our model outperforms the benchmarks in terms of AUC and PRAUC. Meanwhile, we also compare the results for crisis and pre-crisis periods, we find that the deep learning model is better than the traditional models in both crisis time and pre-crisis periods. In addition, the GC-LSTM model is better at predicting future links in the crisis period than the traditional statistical models which indicates that the model without statistical underlying assumptions is better at capturing structure change.
590 $a School code: 0060.
650 4 $a Finance. $3 183252
650 4 $a Statistics. $3 182057
653 $a Bayesian sampling
653 $a Deep learning
653 $a Interbank network
653 $a Link prediction
690 $a 0508
690 $a 0463
690 $a 0770
710 2 $a University of Delaware. $b Financial Services Analytics. $3 966938
773 0 $t Dissertations Abstracts International $g 85-03B.
790 $a 0060
791 $a Ph.D.
792 $a 2023
793 $a English
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=30574663