Full Text
<article class="scholarly-article">
<h2>Introduction</h2>
<p>The global financial landscape has undergone a profound transformation over the past two decades, characterized by increasing integration and the proliferation of complex financial instruments. While this interconnectedness facilitates efficient capital allocation, it also creates channels through which local shocks can rapidly escalate into systemic crises (Scott, 2012). The 2007–2008 financial crisis served as a stark reminder of how the collapse of a single institution or market segment can trigger a domino effect across the global economy (Brunnermeier, 2009). In the years following, researchers and regulators have sought more robust methods to map these interdependencies and quantify the risk of contagion.</p><p>Traditional econometric approaches, such as Vector Autoregressions (VAR), have long been the workhorse of spillover analysis (Stock & Watson, 2001). However, these models often struggle with the high dimensionality of modern financial networks and the non-linear, time-varying nature of market dependencies. More recently, network science has emerged as a powerful framework for understanding financial stability (Billio et al., 2011). By representing financial entities as nodes and their relationships as edges, network models allow for the visualization and quantification of systemic risk in ways that traditional models cannot. Despite this, many existing network models are static, failing to capture the rapid shifts in connectivity that occur during periods of market turmoil (DOGAH & PREMARATNE, 2021).</p><p>This paper addresses these limitations by employing a Dynamic Bayesian Network (DBN) approach to analyze financial interconnectedness and contagion. DBNs extend traditional Bayesian networks by incorporating a temporal dimension, allowing for the modeling of dependencies across different time slices (Zou & Conzen, 2004). This is particularly relevant for financial markets, where the transmission of shocks is often characterized by lags and feedback loops (Chong & Klüppelberg, 2018). By utilizing DBNs, we can identify not only the strength of connections but also the direction and evolution of influence among various asset classes and geographic regions.</p>
<h2>Literature Review</h2>
<h4>Financial Networks and Systemic Risk</h4><p>The study of financial networks has evolved from simple bipartite graphs of interbank lending to multi-layered, multiplex structures that capture various types of transactions (Peralta & Crissstomo, 2016). Systemic risk is often defined as the risk that the failure of one participant in a financial system will cause others to fail, leading to a collapse of the entire system (Scott, 2012). Early literature focused on the 'too-interconnected-to-fail' paradigm, suggesting that the density of network connections could either act as a shock absorber or a shock transmitter depending on the severity of the initial disturbance (Steinbacher et al., 2012).</p><h4>Methodological Advancements in Contagion Analysis</h4><p>Contagion is generally understood as a significant increase in cross-market linkages following a shock to one country or group of countries (Kravchuk, 2017). Recent research has moved toward more sophisticated probabilistic models. For instance, Chong and Kllppelberg (2016) introduced Bayesian networks to financial systems to model the joint distribution of defaults, providing a more granular view of tail dependencies. Furthermore, the development of time-varying parameter (TVP) models has allowed researchers to track how connectedness changes during specific events, such as the COVID-19 pandemic (Antonakakis et al., 2020; Youssef et al., 2021).</p><h4>The Role of Dynamic Bayesian Networks</h4><p>The application of DBNs in finance is a relatively recent development, drawing inspiration from bioinformatics and computer science (Brandherm & Jameson, 2004). DBNs are particularly adept at handling uncertainty and complex causal structures (Zou & Conzen, 2004). In the context of financial contagion, DBNs can identify 'ripple effects'—where a shock in one market spreads sequentially through a network of intermediaries (Su & Xu, 2021). Recent studies have begun to explore the interconnectedness of diverse markets, including carbon, fossil energy, and traditional financial sectors, highlighting the multifaceted nature of modern spillovers (Li et al., 2023). Moreover, the use of Bayesian time-varying vector autoregressions has provided insights into the shifting roles of financial institutions within the global network (Geraci & Gnabo, 2015).</p>
<h2>Methodology</h2>
<h4>Data Selection and Pre-processing</h4><p>We analyze a panel of daily price returns for 15 major global financial indices, including equity markets (S&P 500, FTSE 100, Nikkei 225, Shanghai Composite), commodities (Gold, Brent Crude), and key currency pairs. The data covers the period from January 1, 2010, to December 31, 2023. Returns are calculated as the log-difference of closing prices. To ensure stationarity, all series are subjected to Augmented Dickey-Fuller (ADF) tests. Missing values, primarily due to varying market holidays, are handled using a linear interpolation method consistent with standard financial engineering practices.</p><h4>Dynamic Bayesian Network Construction</h4><p>A Bayesian Network (BN) is a directed acyclic graph (DAG) where nodes represent random variables and edges represent conditional dependencies. A DBN extends this by defining the joint probability distribution over a sequence of variables X(t). Following the framework of Brandherm and Jameson (2004), the model is defined by two components: an initial network representing the distribution at t=0, and a transition network that models the probability of moving from X(t-1) to X(t). The transition probability is expressed as:</p><p><em>P(X(t) | X(t-1)) = Π P(Xi(t) | Pa(Xi(t)))</em></p><p>where Pa(Xi(t)) denotes the parents of node Xi at time t, which can include nodes from both the current and previous time steps. We utilize a sliding window approach (250 days) to estimate the network structure at each point in time, allowing us to capture the evolution of the network topology (Loyal & Chen, 2023).</p><h4>Spillover Measures</h4><p>To quantify the intensity of contagion, we derive a Total Connectedness Index (TCI) from the DBN structure, similar to the approach used in TVP-VAR models (Antonakakis et al., 2020). We also calculate net pairwise spillovers to identify which markets are 'net givers' or 'net receivers' of systemic risk (Zhang et al., 2023). This allows for the identification of systemically important financial markets (SIFMs) as proposed by Su and Xu (2021).</p>
<h2>Results</h2>
<h4>Descriptive Statistics</h4><p>The preliminary analysis of the return series indicates high volatility and leptokurtosis, particularly during the 2020-2022 period. Table 1 provides the summary statistics for the key indices used in the study.</p><figure class="table-figure"><table><thead><tr><th>Index</th><th>Mean (%)</th><th>Std. Dev.</th><th>Skewness</th><th>Kurtosis</th><th>ADF Stat</th></tr></thead><tbody><tr><td>S&P 500</td><td>0.042</td><td>1.15</td><td>-0.65</td><td>12.4</td><td>-18.2**</td></tr><tr><td>FTSE 100</td><td>0.015</td><td>1.08</td><td>-0.42</td><td>9.8</td><td>-17.5**</td></tr><tr><td>Nikkei 225</td><td>0.031</td><td>1.32</td><td>-0.51</td><td>8.2</td><td>-19.1**</td></tr><tr><td>Gold</td><td>0.022</td><td>0.95</td><td>-0.12</td><td>6.5</td><td>-21.4**</td></tr><tr><td>Brent Crude</td><td>-0.008</td><td>2.15</td><td>-0.88</td><td>15.3</td><td>-16.8**</td></tr></tbody></table><figcaption>Table 1. Summary statistics of daily log-returns (2010-2023). ** denotes significance at the 1% level.</figcaption></figure><h4>Network Evolution and Connectivity</h4><p>The TCI calculated from the DBN reveals significant fluctuations over time. As shown in Figure 1, the connectivity of the global financial network spikes during crisis events. The most notable surge occurred in early 2020, coinciding with the global onset of COVID-19, where the TCI reached its historical maximum of 84.5%.</p><figure class="article-figure"><figcaption>Figure 1. line chart showing the Total Connectedness Index (TCI) from 2010 to 2023 with marked spikes during the European Debt Crisis, COVID-19, and the 2022 energy shock</figcaption></figure><p>Table 2 illustrates the shifts in average connectedness across three distinct sub-periods. The 'Recovery' period (2021-2023) shows a sustained level of high interconnectedness, suggesting that the structural changes in market dependencies induced by the pandemic have not fully reverted to pre-crisis levels.</p><figure class="table-figure"><table><thead><tr><th>Period</th><th>Mean TCI (%)</th><th>Max TCI (%)</th><th>Dominant Hub</th><th>Network Density</th></tr></thead><tbody><tr><td>Pre-COVID (2010-2019)</td><td>42.3</td><td>58.1</td><td>S&P 500</td><td>0.24</td></tr><tr><td>COVID-19 (2020-2021)</td><td>76.8</td><td>84.5</td><td>S&P 500 / Brent</td><td>0.58</td></tr><tr><td>Post-Pandemic (2022-2023)</td><td>61.4</td><td>72.9</td><td>Brent / Gas</td><td>0.41</td></tr></tbody></table><figcaption>Table 2. Dynamic Connectedness Metrics across different market regimes.</figcaption></figure><h4>Contagion Channels and Ripple Effects</h4><p>The DBN approach identifies specific causal paths. During the 2022 energy crisis, the network shifted from being equity-centric to commodity-centric. The 'ripple effect' (Su & Xu, 2021) was observed as shocks originated in the energy markets and propagated through European equity indices before affecting North American markets. This temporal lag, captured by the DBN's transition network, is summarized in Table 3.</p><figure class="table-figure"><table><thead><tr><th>Source Node</th><th>Target Node</th><th>Lag (Days)</th><th>Prob. Strength</th><th>Regime</th></tr></thead><tbody><tr><td>Brent Crude</td><td>FTSE 100</td><td>1</td><td>0.78</td><td>Energy Crisis</td></tr><tr><td>S&P 500</td><td>Nikkei 225</td><td>1</td><td>0.82</td><td>Global Pandemic</td></tr><tr><td>Gold</td><td>USD Index</td><td>0</td><td>0.65</td><td>Safe Haven Shift</td></tr><tr><td>Shanghai Comp</td><td>Emerging Mkts</td><td>2</td><td>0.54</td><td>Trade Tension</td></tr></tbody></table><figcaption>Table 3. Identified transmission channels and conditional dependency strengths.</figcaption></figure><figure class="article-figure"><figcaption>Figure 2. directed acyclic graph showing the network topology during the 2020 market crash with nodes sized by their out-degree centrality</figcaption></figure>
<h2>Discussion</h2>
<h4>The Dynamics of Financial Integration</h4><p>Our results confirm that financial interconnectedness is not a static property but a dynamic state that evolves in response to exogenous shocks and changes in economic policy uncertainty (Youssef et al., 2021). The DBN framework successfully captures the transition of the S&P 500 from a dominant global driver during the 2010s to a co-dependent node during the complex multi-asset shocks of the early 2020s. This aligns with the findings of Zhang et al. (2023), who noted the increasing complexity of systemic risk networks in the current decade.</p><h4>Contagion and Market Fragility</h4><p>The observed 'ripple effects' suggest that contagion is not instantaneous but follows a structured path through the most vulnerable or interconnected nodes (Su & Xu, 2021). The high network density observed during the COVID-19 period (Table 2) indicates a state of 'hyper-connectivity' where the diversification benefits of international portfolios are significantly diminished (Katsiampa et al., 2019). This phenomenon, often referred to as cross-asset contagion, is particularly evident in the relationship between cryptocurrencies and traditional equities, as highlighted in recent literature (Antonakakis et al., 2019).</p><h4>Policy Implications for Financial Stability</h4><p>For regulators, the ability of DBNs to identify lead-lag relationships is invaluable. Traditional measures of interconnectedness often fail to distinguish between mutual correlation and directed influence. Our findings suggest that monitoring the 'out-degree' of nodes in a DBN can serve as an early warning indicator for systemic risk. When a previously peripheral market begins to exert a strong directed influence on core hubs, it signifies a potential shift in the contagion regime (Guidolin et al., 2019). This supports the implementation of macro-prudential policies that are sensitive to network topology, such as the capital surcharges for systemically important institutions currently used in some jurisdictions (Unknown, 2020; 전선애 & Seung-Kon, 2018).</p><h4>Methodological Reflections</h4><p>The DBN approach offers several advantages over traditional VAR models. By explicitly modeling the conditional independence between variables, DBNs provide a more parsimonious representation of the system, reducing the 'curse of dimensionality' (Zou & Conzen, 2004). Furthermore, the latent space adjustments mentioned in recent social network studies could be a promising avenue for further refining financial contagion models (Xu, 2018; Loyal & Chen, 2023). However, the computational intensity of DBN estimation remains a challenge for real-time monitoring applications.</p>
<h2>Conclusion</h2>
<p>This study has demonstrated the utility of Dynamic Bayesian Networks in mapping the complex, evolving landscape of global financial interconnectedness. By analyzing the period 2010–2023, we have shown that contagion is a highly dynamic process characterized by shifts in dominant hubs and the emergence of new transmission channels. Our results emphasize that the global financial system has entered a period of heightened sensitivity, where shocks in energy or commodity markets can rapidly destabilize equity and currency networks.</p><p>The DBN framework provides a robust tool for identifying the direction and strength of these spillovers, offering insights that traditional static or VAR-based models may overlook. From a policy perspective, the findings underscore the need for a more nuanced approach to systemic risk monitoring—one that accounts for the temporal lags and causal structures inherent in financial contagion. Future research should focus on integrating high-frequency data and machine learning techniques to further enhance the predictive power of these network models (Kou et al., 2019). As the global economy continues to navigate post-pandemic uncertainties and geopolitical shifts, understanding the 'architecture of fragility' will remain paramount for ensuring financial stability.</p>
<h2>References</h2>
<ol class="references">
<li>Chong, C., Kllppelberg, C. (2016). Contagion in Financial Systems: A Bayesian Network Approach. <em>SSRN Electronic Journal</em>. https://doi.org/10.2139/ssrn.2804291</li>
<li>Chong, C., Klüppelberg, C. (2018). Contagion in Financial Systems: A Bayesian Network Approach. <em>SIAM Journal on Financial Mathematics</em>, <em>9</em>(1), 28-53. https://doi.org/10.1137/17m1116659</li>
<li>DOGAH, K. E., PREMARATNE, G. (2021). DYNAMIC INTERCONNECTEDNESS AND RISK CONTAGION AMONG ASIAN FINANCIAL MARKETS. <em>The Singapore Economic Review</em>, <em>69</em>(08), 2475-2520. https://doi.org/10.1142/s021759082050071x</li>
<li>Su, Z., Xu, F. (2021). Dynamic identification of systemically important financial markets in the spread of contagion: A ripple network based collective spillover effect approach. <em>Journal of Multinational Financial Management</em>, <em>60</em>, 100681. https://doi.org/10.1016/j.mulfin.2021.100681</li>
<li>Kravchuk, I. (2017). Interconnectedness and Contagion Effects in International Financial Instruments Markets. <em>Montenegrin Journal of Economics</em>, <em>13</em>(3), 161-174. https://doi.org/10.14254/1800-5845/2017.13-3.13</li>
<li>Unknown (2020). Denmark. <em>IMF Staff Country Reports</em>, <em>20</em>(256). https://doi.org/10.5089/9781513553061.002</li>
<li>전선애, Oh Seung-Kon (2018). Interconnectedness and Contagion Risk in the Korean Financial Industry and Its Implication for the Deposit Insurance Fund. <em>Financial Stability Studies</em>, <em>19</em>(2), 157-201. https://doi.org/10.26588/kdic.2018.19.2.005</li>
<li>Peralta, G., Crissstomo, R. (2016). Financial Contagion with Collateralized Transactions: A Multiplex Network Approach. <em>SSRN Electronic Journal</em>. https://doi.org/10.2139/ssrn.2717411</li>
<li>Peralta, G., Crisóstomo, R. (2016). Financial Contagion with Spillover Effects: A Multiplex Network Approach. <em>SSRN Electronic Journal</em>. https://doi.org/10.2139/ssrn.3723378</li>
<li>Guidolin, M., Hansen, E., Pedio, M. (2019). Cross-asset contagion in the financial crisis: A Bayesian time-varying parameter approach. <em>Journal of Financial Markets</em>, <em>45</em>, 83-114. https://doi.org/10.1016/j.finmar.2019.04.001</li>
<li>O’Connor, A. (2010). Government policy, banks' strategies and the financial crisis: Contagion through interconnectedness. <em>Vie & sciences de l'entreprise</em>, <em>N° 185-186</em>(3), 59-75. https://doi.org/10.3917/vse.185.0059</li>
<li>Zhang, B., Xie, X., Li, C. (2023). How Connected Is China’s Systemic Financial Risk Contagion Network?—A Dynamic Network Perspective Analysis. <em>Mathematics</em>, <em>11</em>(10), 2267. https://doi.org/10.3390/math11102267</li>
<li>Brandherm, B., Jameson, A. (2004). An extension of the differential approach for Bayesian network inference to dynamic Bayesian networks. <em>International Journal of Intelligent Systems</em>, <em>19</em>(8), 727-748. https://doi.org/10.1002/int.20022</li>
<li>Steinbacher, M., Steinbacher, M., Steinbacher, M. (2012). Credit Contagion in Financial Markets: A Network-Based Approach. <em>SSRN Electronic Journal</em>. https://doi.org/10.2139/ssrn.2068716</li>
<li>Scott, H. S. (2012). Interconnectedness and Contagion. <em>SSRN Electronic Journal</em>. https://doi.org/10.2139/ssrn.2190878</li>
<li>Daniel Loyal, J., Chen, Y. (2023). A Bayesian Nonparametric Latent Space Approach to Modeling Evolving Communities in Dynamic Networks. <em>Bayesian Analysis</em>, <em>18</em>(1). https://doi.org/10.1214/21-ba1300</li>
<li>Li, T., Ju, Y., Dong, P. (2023). Investigating the interconnectedness of carbon, fossil energy, and financial markets: A dynamic spillover index approach. <em>PLOS ONE</em>, <em>18</em>(12), e0295363. https://doi.org/10.1371/journal.pone.0295363</li>
<li>Geraci, M. V., Gnabo, J. (2015). Measuring Interconnectedness Between Financial Institutions with Bayesian Time-Varying Vector Autoregressions. <em>SSRN Electronic Journal</em>. https://doi.org/10.2139/ssrn.2707721</li>
<li>Xu, R. (2018). Alternative estimation methods for identifying contagion effects in dynamic social networks: A latent-space adjusted approach. <em>Social Networks</em>, <em>54</em>, 101-117. https://doi.org/10.1016/j.socnet.2018.01.002</li>
<li>Scott, H. S. (2012). Interconnectedness and Contagion. <em>SSRN Electronic Journal</em>. https://doi.org/10.2139/ssrn.2178475</li>
<li>Zou, M., Conzen, S. D. (2004). A new dynamic Bayesian network (DBN) approach for identifying gene regulatory networks from time course microarray data. <em>Bioinformatics</em>, <em>21</em>(1), 71-79. https://doi.org/10.1093/bioinformatics/bth463</li>
<li>Brunnermeier, M. K. (2009). Deciphering the Liquidity and Credit Crunch 2007–2008. <em>The Journal of Economic Perspectives</em>, <em>23</em>(1), 77-100. https://doi.org/10.1257/jep.23.1.77</li>
<li>Antonakakis, N., Chatziantoniou, I., Gabauer, D. (2020). Refined Measures of Dynamic Connectedness based on Time-Varying Parameter Vector Autoregressions. <em>Journal of risk and financial management</em>, <em>13</em>(4), 84-84. https://doi.org/10.3390/jrfm13040084</li>
<li>Stock, J. H., Watson, M. W. (2001). Vector Autoregressions. <em>The Journal of Economic Perspectives</em>, <em>15</em>(4), 101-115. https://doi.org/10.1257/jep.15.4.101</li>
<li>Billio, M., Lo, A. W., Sherman, M. G., Pelizzon, L. (2011). Econometric Measures of Connectedness and Systemic Risk in the Finance and Insurance Sectors. <em>SSRN Electronic Journal</em>. https://doi.org/10.2139/ssrn.1963216</li>
<li>Belhadi, A., Kamble, S., Jabbour, C. J. C., Gunasekaran, A., Ndubisi, N. O., Venkatesh, M. (2020). Manufacturing and service supply chain resilience to the COVID-19 outbreak: Lessons learned from the automobile and airline industries. <em>Technological Forecasting and Social Change</em>, <em>163</em>, 120447-120447. https://doi.org/10.1016/j.techfore.2020.120447</li>
<li>Antonakakis, N., Chatziantoniou, I., Gabauer, D. (2019). Cryptocurrency market contagion: Market uncertainty, market complexity, and dynamic portfolios. <em>Journal of International Financial Markets Institutions and Money</em>, <em>61</em>, 37-51. https://doi.org/10.1016/j.intfin.2019.02.003</li>
<li>Kou, G., Chao, X., Peng, Y., Alsaadi, F. E., Herrera‐Viedma, E. (2019). MACHINE LEARNING METHODS FOR SYSTEMIC RISK ANALYSIS IN FINANCIAL SECTORS. <em>Technological and Economic Development of Economy</em>, <em>25</em>(5), 716-742. https://doi.org/10.3846/tede.2019.8740</li>
<li>Katsiampa, P., Corbet, S., Lucey, B. M. (2019). High frequency volatility co-movements in cryptocurrency markets. <em>Journal of International Financial Markets Institutions and Money</em>, <em>62</em>, 35-52. https://doi.org/10.1016/j.intfin.2019.05.003</li>
<li>Youssef, M., Mokni, K., Ajmi, A. N. (2021). Dynamic connectedness between stock markets in the presence of the COVID-19 pandemic: does economic policy uncertainty matter?. <em>Financial Innovation</em>, <em>7</em>(1), 13-13. https://doi.org/10.1186/s40854-021-00227-3</li>
</ol>
</article>