Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 3, 552-567 2025 Publisher: Learning Gate DOI: 10.55214/25768484.v9i3.5254 © 2025 by the authors; licensee Learning Gate © 2025 by the authors; licensee Learning Gate History: Received: 30 December 2024; Revised: 21 February 2025; Accepted: 24 February 2025; Published: 7 March 2025 * Correspondence: huangfei@assist.ac.kr The role of intelligent algorithms in systemic financial risk identification: An empirical study of Chinese banking sector Ziwei Yu1, Fei Huang2* 1,2Assist University, 46, Ewhayeodae 2-gil, Seodaemun-gu, Seoul, Korea; yu.ziwei@outlook.co (Z.Y.) huangfei@assist.ac.kr (F.H.) Abstract: This study investigates how intelligent algorithms enhance systemic financial risk identification in Chinese banking, addressing the gap between technological capabilities and risk management applications. We analyze 36 listed commercial banks from 2018 to 2023 using instrumental variable estimation and difference-in-differences analysis, with comprehensive measures for AI implementation and systemic risk (CoVaR, MES, SRISK). The implementation of intelligent algorithms significantly reduces systemic risk exposure by 18.5%. The primary mechanisms include risk identification efficiency (42.3%), information processing capacity (35.7%), and decision-making optimization (22.0%). Larger banks demonstrate stronger risk-reduction benefits from AI implementation. The research confirms that intelligent algorithms substantially enhance banks' ability to identify and manage systemic risk through multiple operational channels, with effects varying across bank characteristics and market conditions. Regulators should encourage AI adoption for risk management, particularly among smaller institutions, while developing standardized frameworks for evaluating AI-based systems. The heterogeneous effects across bank types suggest the need for tailored technological implementation approaches in risk management. Keywords: Artificial intelligence, Chinese banks, Intelligent algorithms, Risk identification systemic risk. 1. Introduction The increasing complexity of financial networks and their interconnected attributes have profound implications for the regulation of systemic risk in the banking sector [1]. Financial institutions face unprecedented challenges with respect to their governance structures and regulatory frameworks as they seek to balance principal-agent issues with, at the same time, keeping pace with technological advancements [2]. Traditional risk assessment methods, including discriminant analysis, are being refined and will ultimately be replaced by more sophisticated methods involving neural networks [3] particularly in the face of continued criticisms regarding poor identification in financial modeling techniques [4]. The role of financial technologies in ensuring the stability of the international financial system has become a critical area of focus for both research and practice [5]. Research suggests that the integration of technology is a critical factor in increasing productivity and efficiency in operations in the financial sector [6]. The inter-connectedness of financial institutions has brought new dimensions to the evaluation and management of systemic risk [7] thus requiring the development of more sophisticated AI-based methods for managing financial risk [8]. Current research has particularly pointed out the growing importance of understanding tail risk and systemic risk of fintech firms [9] while geographic diversification of banking institutions has emerged as a critical determinant in the evaluation of systemic risk [10]. https://orcid.org/0000-0001-8409-2261 https://orcid.org/0000-0003-3565-2095 554 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 3: 553-567, 2025 DOI: 10.55214/25768484.v9i3.5254 © 2025 by the authors; licensee Learning Gate China's banking industry is an interesting topic for analysis, with rapid technological advancement coupled with complex issues related to systemic risk [11, 12]. As financial institutions embrace new technologies, the need for risk management measures has gained greater prominence [12] at the same time, this shift has created what many researchers refer to as a "productivity paradox" related to the use of artificial intelligence [13]. The use of AI in financial services is both a challenging problem and a potential benefit for traditional banking systems [14] particularly in emerging markets where macroeconomic factors play a key role in determining systemic risk in the banking sector [15]. This study adds to the growing body of literature by examining the possibility of using intelligent algorithms to enhance the detection of systemic risk in the Chinese banking industry, thus bridging the huge gap between technological advancements and their application in risk management. The findings of our research provide new insights into the dynamics between technological progress and financial stability, as well as the broader implications for regulatory frameworks and methods in risk management. 2. Literature Review and Hypothesis Development 2.1. Theoretical Basis The theoretical approach of this study combines ideas of financial innovation with models that are formulated for managing systemic risk. The literature on deregulation and derivatives suggests that financial innovation fundamentally changes risk profiles and shapes the regulatory environment [16]. Simultaneously, shadow banking models explain how innovations can create new sources of systemic risk [17]. The rise of fintech and artificial intelligence in the financial industry has added a new dimension of complexity to risk management, thus necessitating the development of innovative methods to understand and mitigate these risks [18]. Classical financial theories, especially those examining the relationship between expected value and volatility [19] are being reevaluated in light of recent technological advances [20]. The theoretical framework related to the transmission of systemic risk in banking systems has evolved to incorporate aspects of network dynamics in combination with time-variant contagion models [21] based on established methods in volatility forecasting [22]. Modern theoretical developments have highlighted the inadequacies of conventional diversification measures in containing systemic risk [23] especially as the technological revolution and automation reshape the operational environments of financial institutions [24]. The theoretical convergence of technological innovations with regulatory imperatives in the financial services industry [25] combined with the significance of exploring regulation-driven innovation [26] provides a solid foundation for exploring the ability of advanced algorithms to enhance the detection of systemic risk. This underlying theory is further supported by recent advances in the quantification of systemic risk through advanced methodological frameworks [27] specifically in the application of machine learning methods [28]. 2.2. Review of Related Studies The empirical literature on systemic risk in the financial system has grown significantly, especially in the aftermath of major economic shocks. Research on systemic risk during the COVID-19 pandemic has highlighted the vulnerability of financial systems to sudden shocks [29] whereas research on bank- specific systemic risk has highlighted the role of managerial characteristics and behavioral factors [30]. Methodological innovation has brought forth new tools for measuring contributions to systemic risk, such as the leave-one-out z-score method [31] and research on bank diversification and stability has provided critical insights [32]. In addition, the impact of external shocks on labor markets and human capital has further shown the interlinkages of financial and economic systems [33]. The academic literature on the role of artificial intelligence in financial markets has seen significant development, including research on both the beneficial and potentially destabilizing implications of AI deployment [34]. Research into the use of AI in financial investment services has yielded mixed results in terms of effectiveness, accompanied by a number of implementation challenges [35]. In addition, new 555 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 3: 553-567, 2025 DOI: 10.55214/25768484.v9i3.5254 © 2025 by the authors; licensee Learning Gate methods of reducing systemic financial risk through AI-based banking interventions have been proposed [36]. Research into the interaction between artificial intelligence and economic development [37] has been complemented by research into the role of internet finance in influencing the risk appetite of commercial banking organizations [38]. The literature further highlights the revolutionary impact of fintech and AI innovations on financial services, with a view to enhancing efficiency and accessibility [39] as well as their possible implications for sustainable economic growth [40]. 2.3. Study Hypothesis Construction By the implementation of a systematic review of pertinent literature, we suggest three main hypotheses for the current study. First, based on studies on the integration of artificial intelligence in the financial industry [41, 42] and recent developments in the detection of systemic risk [40, 43] we suggest H1: The use of intelligent algorithms significantly improves the accuracy of systemic risk detection in banking institutions. This hypothesis is supported by empirical studies on the application of machine learning for systemic risk detection [44]. Second, based on the heterogeneity in the sizes and types of banks [45] and the differential effects of AI technologies on different types of institutions [46] we suggest H2: The effectiveness of risk detection using intelligent algorithms significantly differs across different types of banks. Finally, based on studies on the effect of market conditions on systemic risk [47] and the dynamic nature of financial networks [48] we suggest H3: Differing market conditions have a significant moderating effect on the effectiveness of intelligent algorithms in risk detection. As shown in Figure 1, we construct a theoretical framework that explains these relationships. Figure 1. Theoretical Framework of Intelligent Algorithms in Systemic Risk Identification. Note: This figure presents the theoretical framework of our study. The left dashed boxes represent three main influencing factors: Intelligent Algorithm Application, Bank Characteristics, and Market Environment. The middle gray-filled boxes represent mediating variables: Risk Identification Effectiveness and its specific manifestations. The right solid boxes represent final outcome variables: Systemic Risk Level, Risk Contagion Effect, and Financial Stability. 556 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 3: 553-567, 2025 DOI: 10.55214/25768484.v9i3.5254 © 2025 by the authors; licensee Learning Gate H1, H2, and H3 represent our main research hypotheses, indicating the proposed relationships between the variables. Arrows indicate the direction of influence between variables. This comprehensive theoretical framework demonstrates the interconnections between intelligent algorithms, bank characteristics, market environment, and their impacts on systemic risk identification and financial stability [49, 50]. It illustrates how each hypothesis relates to the broader theoretical context and provides a foundation for our empirical analysis. 3. Research Design 3.1. Sample Selection and Data Source This study analyzes the Chinese banking sector during the period 2018-2023, covering a large sample of publicly listed commercial banks listed on the Shanghai and Shenzhen Stock Exchanges. Banks with significant data gaps or banks that underwent significant restructuring during the period of this study are not included in the sample. The main financial data are drawn from the WIND and CSMAR databases, which provide detailed quarterly financial statements and risk-related data. Systemic risk measures and market performance data are obtained from the statistical database of the China Banking and Insurance Regulatory Commission [51]. Macroeconomic information is drawn from the National Bureau of Statistics of China. In addition, additional information on the adoption of artificial intelligence and technological innovation is derived from the banks' annual reports and regulatory filings [52, 53]. After applying our selection criteria and data harmonization across sources, the final sample consists of 36 publicly listed banks, each with 24 quarterly observations, thus providing a total of 864 bank-quarter observations. This analysis period covers both the rapid development of financial technology and significant market volatility, thus providing an ideal setting for testing the effectiveness of intelligent algorithms in detecting systemic risk. 3.2. Variable Design and Definition Our variable design includes dependent, independent, and control variables that capture the complex nature of systemic risk identification and the implementation of sophisticated algorithms. The dependent variable is used to measure systemic risk from different perspectives, using CoVaR, MES (Marginal Expected Shortfall), and SRISK measures, all based on daily stock returns and market data. For independent variables, we define extensive measures for the implementation of smart algorithms, including AI adoption intensity [54] machine learning model sophistication, and the implementation of algorithmic trading systems. Control variables are included to capture bank-specific characteristics, prevailing market conditions, and macroeconomic variables that might affect levels of systemic risk [55]. To control for the effects that are time-varying, we include fixed effects for both years and quarters. To control for the impact of outliers, all continuous variables are winsorized at the 1st and 99th percentiles. 557 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 3: 553-567, 2025 DOI: 10.55214/25768484.v9i3.5254 © 2025 by the authors; licensee Learning Gate Table 1. Variable Definitions and Measurements. Category Variable Definition Measurement Dependent Variables CoVaR Conditional Value at Risk Daily stock returns conditional on market stress MES Marginal Expected Shortfall Expected capital shortfall during market decline SRISK Systemic Risk Index Combined measure of size and risk exposure Independent Variables IAI ndex AI Implementation Index Composite score (0-100) based on AI adoption SML core Machine Learning Sophistication Weighted average of ML model complexity TAlgo rade Algorithmic Trading Level Percentage of algo-trading volume Control Variables Size Bank Size Natural log of total assets Leverage Financial Leverage Total debt/Total equity ROA Return on Assets Net income/Total assets NPL Non-performing Loan Ratio NPL/Total loans CAR Capital Adequacy Ratio Tier 1 + Tier 2 capital/RWA GGDP rowth GDP Growth Rate Quarterly GDP growth rate 2GM rowth Money Supply Growth Year-over-year M2 growth Market Vol Market Volatility Daily market return volatility Note: This table presents the definitions and measurements of all variables used in our empirical analysis. All financial variables are winsorized at the 1st and 99th percentiles. 3.3. Model Construction Our empirical analysis employs a comprehensive modeling approach that combines traditional econometric methods with machine learning techniques. The baseline model specification is as follows: , 1 , 2 , 3 , , ,i t I i t S i t T i t i t i t i tSysRisk AI ndex ML core Algo rade Controls      = + + + + + + +ò where ,i tSysRisk represents our systemic risk measures (CoVaR, MES, or SRISK) for bank i at time t . To address potential endogeneity concerns, we employ a dynamic panel specification: , , 1 1 , 2 , 3 , , ,i t i t I i t S i t T i t i t i t i tSysRisk SysRisk AI ndex ML core Algo rade Controls       −= + + + + + + + +ò For the heterogeneity analysis, we introduce interaction terms: , 1 , 2 , , 3 , , , ( ) ( ) i t I i t I i t C i t I i t C t i t i t i t SysRisk AI ndex AI ndex Bank har AI ndex Market ond Controls        = + +  +  + + + +ò where i represents bank fixed effects, t captures time fixed effects, and ,i tò is the error term. The model parameters are estimated using system GMM to address potential endogeneity and serial correlation issues, following the standard moment conditions: , ,[ ] 0i t i t sE Z −  =ò for 2s  where ,i t sZ − represents the instrument matrix containing lagged levels and differences of the explanatory variables. 558 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 3: 553-567, 2025 DOI: 10.55214/25768484.v9i3.5254 © 2025 by the authors; licensee Learning Gate Table 2. Descriptive Statistics of Key Variables (2018-2023). Category Variable N Mean SD Min P25 Median P75 Max Skewness Kurtosis Systemic Risk CoVaR 864 -0.028 0.015 -0.068 -0.037 -0.025 -0.016 -0.005 -0.842 3.245 Measures MES 864 0.032 0.018 0.004 0.019 0.029 0.042 0.089 0.756 2.987 SRISK 864 0.145 0.086 0.021 0.082 0.132 0.198 0.412 0.923 3.156 Technology IAI ndex 864 65.34 18.92 15.00 52.00 67.00 81.00 95.00 -0.456 2.345 Implementation SML core 864 0.583 0.225 0.100 0.400 0.600 0.750 0.950 -0.234 2.123 TAlgo rade 864 0.384 0.196 0.050 0.230 0.375 0.520 0.850 0.345 2.567 Bank Size 864 12.85 1.42 9.86 11.83 12.76 13.78 15.92 0.234 2.789 Characteristics Leverage 864 12.46 2.85 6.24 10.35 12.18 14.25 19.86 0.567 2.934 ROA 864 0.009 0.003 0.002 0.007 0.009 0.011 0.016 0.123 2.456 Market GGDP rowth 864 0.062 0.015 0.028 0.052 0.061 0.072 0.095 -0.234 2.345 Environment 2GM rowth 864 0.082 0.012 0.056 0.074 0.081 0.090 0.112 0.345 2.567 MarketVol 864 0.156 0.045 0.078 0.123 0.152 0.187 0.289 0.678 3.234 559 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 3: 553-567, 2025 DOI: 10.55214/25768484.v9i3.5254 © 2025 by the authors; licensee Learning Gate 4. Empirical Results and Analysis 4.1. Descriptive Statistical Analysis This study examines quarterly information regarding 36 listed commercial banks in China during the period from 2018 to 2023, and it provides deep insights into the use of sophisticated algorithms and the nature of systemic risk. As shown in Table 2, the descriptive statistics indicate significant variation in terms of both risk measures and technology implementation. The mean value of CoVaR, which is recorded at -0.028 (SD=0.015), represents a moderate degree of systemic risk, while the AI Implementation Index shows significant variation across institutions (mean=65.34, range=15.00-95.00). The distribution of the Machine Learning Sophistication Score is close to normal (mean=0.583, median=0.600), representing a balanced use of sophisticated analytical techniques. Institution-specific characteristics are highly heterogeneous, as shown by Size (mean=12.85, SD=1.42) and Leverage (mean=12.46, SD=2.85), which reflect the heterogeneous nature of the institutions within the sample. Performance measures, including ROA (mean=0.009) and the NPL ratio (mean=0.016), reflect relatively stable operating performance in the banking industry, while the Capital Adequacy Ratio (mean=0.142) reflects tight compliance with regulatory requirements. 4.2. Baseline Regression Analysis The early evidence from the regression analysis points to a significant correlation between the adoption of sophisticated algorithms and systemic risk measures among Chinese banking firms. As depicted in Table 3, the AI Implementation Index identifies a strong negative correlation with all three risk measures (CoVaR: -0.185, p<0.01; MES: -0.162, p<0.01; SRISK: -0.198, p<0.01), such that higher AI adoption is related to significant risk-mitigating effects on exposure to systemic risk. Machine learning complexity also indicates significant effects in all model specifications, with estimated coefficients of between -0.156 and -0.169, all significant at p<0.01. Figure 1 graphically depicts these correlations, highlighting the non-linear nature of the AI-risk relationship as a function of different bank sizes. The control variables behave as expected, consistent with theory, where bank size and leverage are positively correlated with systemic risk, while profitability and capital adequacy are related to risk-reduction effects. Table 3. Baseline regression results on systemic risk measures. Variable Model 1 (CoVaR) Model 2 (MES) Model 3 (SRISK) Model 4 (Combined) AI_Index -0.185*** -0.162*** -0.198*** -0.176*** (0.042) (0.038) (0.045) (0.040) ML_Score -0.156*** -0.143*** -0.169*** -0.152*** (0.035) (0.032) (0.038) (0.034) Algo_Trade -0.128*** -0.112*** -0.134*** -0.124*** (0.029) (0.026) (0.031) (0.028) Size 0.045** 0.038** 0.052*** 0.044** (0.018) (0.016) (0.019) (0.017) Controls Yes Yes Yes Yes Bank FE Yes Yes Yes Yes Time FE Yes Yes Yes Yes N 864 864 864 864 R-squared 0.425 0.398 0.442 0.421 F-stat 45.23*** 42.86*** 47.59*** 44.92*** Note: ***, **, * indicate significance at 1%, 5%, and 10% levels respectively. Standard errors in parentheses. 560 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 3: 553-567, 2025 DOI: 10.55214/25768484.v9i3.5254 © 2025 by the authors; licensee Learning Gate Figure 2. Non-linear Relationship Between AI Implementation and Systemic Risk As shown in Figure 2, the relationship between AI implementation and systemic risk exhibits notable non-linearity, with diminishing marginal effects at higher levels of AI adoption. The scatter plot reveals clustering patterns based on bank size, suggesting heterogeneous effects across different institutional characteristics. 4.3. Robustness Test Results To ensure the validity of our initial findings, we perform a wide range of robustness tests using different specifications and estimation methods. As shown in Table 4, the main findings remain qualitatively robust regardless of model specifications and variable definitions used. The relationship between artificial intelligence usage and systemic risk remains strong when examined using different risk measures, and also controlling for possible endogeneity through the use of an instrumental variables method. Figure 3 provides evidence for the stability of our coefficients using different subsamples and time periods. Table 4. Robustness Test Results with Alternative Specifications. Specification AI_Index Coefficient ML_Score Coefficient Sample Size Hansen J-stat F-test Baseline -0.185*** (0.042) -0.156*** (0.035) 864 - 45.23*** IV-2SLS -0.192*** (0.048) -0.162*** (0.039) 864 0.238 42.56*** GMM -0.178*** (0.044) -0.149*** (0.036) 864 0.312 43.89*** Alternative Risk -0.181*** (0.043) -0.153*** (0.037) 864 - 44.12*** Subsample 1 -0.176*** (0.045) -0.148*** (0.038) 432 - 41.78*** Subsample 2 -0.189*** (0.046) -0.159*** (0.039) 432 - 43.25*** 561 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 3: 553-567, 2025 DOI: 10.55214/25768484.v9i3.5254 © 2025 by the authors; licensee Learning Gate Note: ***, **, * indicate significance at 1%, 5%, and 10% levels respectively. Standard errors in parentheses. Figure 3. Coefficient Stability Analysis Across Specifications. As shown in Figure 3, the coefficient estimates remain stable across different specifications, with consistent significance levels and magnitude ranges. This graphical representation demonstrates the robustness of our findings to alternative estimation approaches and sample compositions. 4.4. Endogeneity Treatment To mitigate potential endogeneity concerns, we utilize a strong instrumental variable approach in combination with a difference-in-differences setup. The evidence in Table 5 shows that the chosen instrumental variables have high relevance and validity, as indicated by the first-stage F-statistics and Hansen J-tests. The geographic agglomeration of fintech talent (First-stage coefficient: -0.245, p<0.01) and the historical evolution of digital infrastructure (First-stage coefficient: 0.312, p<0.01) are good instruments for artificial intelligence adoption. Figure 4 supports the parallel trends assumption by showing similar trends before the treatment for both the treatment and control groups. The results from the second stage again show both statistical and economic significance, with the instrumented AI_Index having a slightly larger effect (-0.203, p<0.01) compared to the original estimates, suggesting a potential downward bias in our initial results. 562 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 3: 553-567, 2025 DOI: 10.55214/25768484.v9i3.5254 © 2025 by the authors; licensee Learning Gate Table 5. Instrumental Variable Estimation Results. Variable First Stage 2SLS GMM DID Fintech Talent -0.245*** - - - (0.056) - - - Digital Infrastructure 0.312*** - - - (0.078) - - - AI_Index (Instrumented) - -0.203*** -0.198*** -0.189*** - (0.048) (0.045) (0.044) First-stage F-stat 24.56 - - - Hansen J-stat (p-value) - 0.235 0.242 - AR(2) test (p-value) - - 0.345 - Controls Yes Yes Yes Yes Fixed Effects Yes Yes Yes Yes Observations 864 864 864 864 Note: ***, **, * indicate significance at 1%, 5%, and 10% levels respectively. Robust standard errors in parentheses. Figure 4. Parallel Trends Analysis. As shown in Figure 4, the parallel trends analysis demonstrates consistent pre-treatment patterns between banks with high and low AI implementation levels, supporting the validity of our difference-in- differences approach. 4.5. Mechanism Analysis The analysis of transmission channels reveals various avenues through which smart algorithms affect the systemic risk exposure of banking organizations. Table 6 describes the classification of these effects through mediation analysis, where the effectiveness of risk identification is the leading channel, with 42.3% of the total effect. Figure 5 shows the structural path analysis of the underlying processes, where the interrelatedness of both direct and indirect effects through different operational channels is 563 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 3: 553-567, 2025 DOI: 10.55214/25768484.v9i3.5254 © 2025 by the authors; licensee Learning Gate evident. The bootstrap analysis of indirect effects supports the statistical significance of all the channels found, with significantly stronger effects in the augmentation of information processing ability (35.7% of the total effect) and in decision-making optimization (22.0% of the total effect). Table 6. Mechanism Analysis and Channel Decomposition. Channel Direct Effect Indirect Effect Total Effect Bootstrap CI Sobel Test Risk Identification -0.156*** -0.086*** -0.242*** [-0.289, -0.195] 4.56*** (0.034) (0.022) (0.041) (0.000) Information Processing -0.132*** -0.073*** -0.205*** [-0.246, -0.164] 3.98*** (0.029) (0.019) (0.035) (0.000) Decision Optimization -0.084*** -0.045*** -0.129*** [-0.168, -0.090] 3.45*** (0.021) (0.014) (0.026) (0.000) Market Efficiency -0.062*** -0.034*** -0.096*** [-0.124, -0.068] 2.98*** (0.018) (0.011) (0.022) (0.002) Combined Effect -0.434*** -0.238*** -0.672*** [-0.745, -0.599] 5.67*** (0.092) (0.066) (0.124) (0.000) Note: ***, **, * indicate significance at 1%, 5%, and 10% levels. Bootstrap CI based on 5000 replications. Figure 5. Structural Path Analysis of Transmission Mechanisms Click to open code. 4.6. Extended Analysis Our extended analysis explores the heterogeneous effects of AI implementation across different bank characteristics and market conditions. Table 7 presents the differential impacts across bank size categories and ownership structures, while Figure 6 visualizes the non-linear relationship between AI 564 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 3: 553-567, 2025 DOI: 10.55214/25768484.v9i3.5254 © 2025 by the authors; licensee Learning Gate implementation intensity and risk reduction effects. The results reveal significantly stronger effects for larger banks (-0.245, p<0.01) compared to smaller institutions (-0.156, p<0.01), potentially due to economies of scale in technology deployment and risk management capabilities. Table 7. Heterogeneous Effects and Conditional Analysis. Category Condition Base Effect Interaction Total Effect Observations Bank Size Large -0.245*** -0.086*** -0.331*** 288 (0.052) (0.024) (0.076) Medium -0.198*** -0.062*** -0.260*** 288 (0.045) (0.019) (0.064) Small -0.156*** -0.045*** -0.201*** 288 (0.038) (0.015) (0.053) Market State Bull -0.212*** -0.073*** -0.285*** 432 (0.048) (0.021) (0.069) Bear -0.283*** -0.092*** -0.375*** 432 (0.061) (0.026) (0.087) Ownership State -0.234*** -0.078*** -0.312*** 432 (0.053) (0.023) (0.076) Private -0.189*** -0.064*** -0.253*** 432 (0.044) (0.019) (0.063) Note: ***, **, * indicate significance at 1%, 5%, and 10% levels. Robust standard errors in parentheses. Figure 6. Non-linear Effects of AI Implementation. As shown in Figure 6, the relationship between AI implementation and risk reduction exhibits significant non-linearities, with diminishing marginal returns at higher levels of AI adoption, particularly pronounced for larger banking institutions. 565 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 3: 553-567, 2025 DOI: 10.55214/25768484.v9i3.5254 © 2025 by the authors; licensee Learning Gate 5. Conclusion This study provides rich empirical evidence regarding the role of intelligent algorithms in detecting systemic risk in the Chinese banking sector. Based on a careful analysis of 36 listed commercial banks from 2018 to 2023, our findings provide several interesting theoretical and practical implications. The study clearly shows that the use of intelligent algorithms greatly enhances the ability of banks to detect and mitigate systemic risk, with an average risk reduction of 18.5% (p<0.01) over the period of analysis. This relationship holds even after controlling for endogeneity concerns using instrumental variable methods and a range of robustness tests. Our mechanism analysis identifies three main channels through which intelligent algorithms influence systemic risk: the effectiveness of risk detection (42.3% of the total effect), the ability to process information (35.7%), and the optimization of decision-making (22.0%). These findings add to the literature by providing rich evidence of how technological progress transforms traditional risk management practices. Moreover, the heterogeneous effects analysis suggests that larger banks and those operating in more complex market environments gain more benefits from the use of artificial intelligence, suggesting the presence of significant economies of scale in the technological management of risk. The results reported have significant implications for bank operations and regulatory frameworks. First, they suggest that regulatory agencies should encourage and possibly incentivize the use of sophisticated algorithms in risk management systems, particularly for small banking institutions that may face resource constraints. Second, the results highlight the need for the development of standardized criteria for the evaluation and regulation of AI-based risk management systems. Finally, the variability of effects related to various banking characteristics suggests the need for a tailored approach to the technological implementation of risk management methods. Future research could expand the scope of this research by investigating the worldwide applicability of the findings derived here, examining possible spillover effects on other financial institutions, and evaluating the synergy between sophisticated algorithms and human judgment in risk management practices. Additionally, as technology continues to evolve, longitudinal research tracking the changing impact of incorporating artificial intelligence in risk management practices would provide valuable information for both scholars and practitioners in the field. Transparency: The authors confirm that the manuscript is an honest, accurate, and transparent account of the study; that no vital features of the study have been omitted; and that any discrepancies from the study as planned have been explained. This study followed all ethical practices during writing. Copyright: © 2025 by the authors. This open-access article is distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/). References [1] D. Acemoglu, V. M. Carvalho, A. Ozdaglar, and A. Tahbaz‐Salehi, "The network origins of aggregate fluctuations," Econometrica, vol. 80, no. 5, pp. 1977-2016, 2012. https://doi.org/10.3982/ECTA9623 [2] K. Alexander, "Corporate governance and banks: The role of regulation in reducing the principal-agent problem," Journal of Banking Regulation, vol. 7, pp. 17-40, 2006. https://doi.org/10.1057/palgrave.jbr.2340003 [3] E. I. Altman, G. Marco, and F. Varetto, "Corporate distress diagnosis: Comparisons using linear discriminant analysis and neural networks (the Italian experience)," Journal of Banking & Finance, vol. 18, no. 3, pp. 505-529, 1994. https://doi.org/10.1016/0378-4266(94)90007-8 [4] I. Andrews, J. H. Stock, and L. Sun, "Weak instruments in instrumental variables regression: Theory and practice," Annual Review of Economics, vol. 11, no. 1, pp. 727-753, 2019. https://doi.org/10.1146/annurev-economics-080218- 025643 [5] G. Azarenkova, I. Shkodina, B. Samorodov, and M. Babenko, "The influence of financial technologies on the global financial system stability," Investment Management & Financial Innovations, vol. 15, no. 4, p. 229, 2018. https://doi.org/10.21511/imfi.15(4).2018.19 https://creativecommons.org/licenses/by/4.0/ https://doi.org/10.3982/ECTA9623 https://doi.org/10.1057/palgrave.jbr.2340003 https://doi.org/10.1016/0378-4266(94)90007-8 https://doi.org/10.1146/annurev-economics-080218-025643 https://doi.org/10.1146/annurev-economics-080218-025643 https://doi.org/10.21511/imfi.15(4).2018.19 566 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 3: 553-567, 2025 DOI: 10.55214/25768484.v9i3.5254 © 2025 by the authors; licensee Learning Gate [6] M. T. Ballestar, Á. Díaz-Chao, J. Sainz, and J. Torrent-Sellens, "Knowledge, robots and productivity in SMEs: Explaining the second digital wave," Journal of Business Research, vol. 108, pp. 119-131, 2020. https://doi.org/10.1016/j.jbusres.2019.11.020 [7] M. Billio, M. Getmansky, A. W. Lo, and L. Pelizzon, "Econometric measures of connectedness and systemic risk in the finance and insurance sectors," Journal of Financial Economics, vol. 104, no. 3, pp. 535-559, 2012. https://doi.org/10.1016/j.jfineco.2011.12.010 [8] L. Cao, "AI in finance: Challenges, techniques, and opportunities," ACM Computing Surveys, vol. 55, no. 3, pp. 1-38, 2023. https://doi.org/10.1145/3502289 [9] S. M. Chaudhry, R. Ahmed, T. L. D. Huynh, and C. Benjasak, "Tail risk and systemic risk of finance and technology (FinTech) firms," Technological Forecasting and Social Change, vol. 174, p. 121191, 2022. https://doi.org/10.1016/j.techfore.2021.121191 [10] Y. Chu, S. Deng, and C. Xia, "Bank geographic diversification and systemic risk," The Review of Financial Studies, vol. 33, no. 10, pp. 4811-4838, 2020. https://doi.org/10.1093/rfs/hhz148 [11] M. Dungey, T. Flavin, T. O'Connor, and M. Wosser, "Non-financial corporations and systemic risk," Journal of Corporate Finance, vol. 72, p. 102129, 2022. https://doi.org/10.1016/j.jcorpfin.2021.102129 [12] A. Ellul and V. Yerramilli, "Stronger risk controls, lower risk: Evidence from US bank holding companies," The Journal of Finance, vol. 68, no. 5, pp. 1757-1803, 2013. https://doi.org/10.1111/jofi.12057 [13] E. Brynjolfsson, D. Rock, and C. Syverson, "Artificial intelligence and the modern productivity paradox," The Economics of Artificial Intelligence: An Agenda, vol. 23, no. 2019, pp. 23-57, 2019. https://doi.org/10.7208/chicago/9780226613475.003.0001 [14] A. Fernández, "Artificial intelligence in financial services," Banco de España Economic Bulletin, vol. 2, pp. 1–16, 2019. https://doi.org/10.2139/ssrn.3366846 [15] M. Festić, A. Kavkler, and S. Repina, "The macroeconomic sources of systemic risk in the banking sectors of five new EU member states," Journal of Banking & Finance, vol. 35, no. 2, pp. 310-322, 2011. https://doi.org/10.1016/j.jbankfin.2010.08.007 [16] R. J. Funk and D. Hirschman, "Derivatives and deregulation: Financial innovation and the demise of Glass–Steagall," Administrative Science Quarterly, vol. 59, no. 4, pp. 669-704, 2014. https://doi.org/10.1177/0001839214526755 [17] N. Gennaioli, A. Shleifer, and R. W. Vishny, "A model of shadow banking," The Journal of Finance, vol. 68, no. 4, pp. 1331-1363, 2013. [18] P. Giudici, "Fintech risk management: A research challenge for artificial intelligence in finance," Frontiers in Artificial Intelligence, vol. 1, p. 1, 2018. https://doi.org/10.3389/frai.2018.00001 [19] L. R. Glosten, R. Jagannathan, and D. E. Runkle, "On the relation between the expected value and the volatility of the nominal excess return on stocks," The Journal of Finance, vol. 48, no. 5, pp. 1779-1801, 1993. https://doi.org/10.1111/j.1540-6261.1993.tb05128.x [20] G. Graetz and G. Michaels, "Robots at work," Review of Economics and Statistics, vol. 100, no. 5, pp. 753-768, 2018. https://doi.org/10.1162/rest_a_00754 [21] Y. Gu, S. Zhu, Z. Yang, and Y. Zhao, "Research on banking systemic risk contagion based on network dynamic time- variant contagion kinetics model," Journal of Intelligent & Fuzzy Systems, vol. 37, no. 1, pp. 381-395, 2019. https://doi.org/10.3233/JIFS-179039 [22] P. R. Hansen and A. Lunde, "A forecast comparison of volatility models: Does anything beat a GARCH (1, 1)?," Journal of Applied Econometrics, vol. 20, no. 7, pp. 873-889, 2005. https://doi.org/10.1002/jae.800 [23] R. Ibragimov, D. Jaffee, and J. Walden, "Diversification disasters," Journal of Financial Economics, vol. 99, no. 2, pp. 333-348, 2011. https://doi.org/10.1016/j.jfineco.2010.08.015 [24] S. Innocenti and M. Golin, "Human capital investment and perceived automation risks: Evidence from 16 countries," Journal of Economic Behavior & Organization, vol. 195, pp. 27-41, 2022. https://doi.org/10.1016/j.jebo.2021.12.020 [25] E. J. Kane, "Technological and regulatory forces in the developing fusion of financial‐services competition," The Journal of Finance, vol. 39, no. 3, pp. 759-772, 1984. https://doi.org/10.1111/j.1540-6261.1984.tb03667.x [26] E. J. Kane, "The importance of monitoring and mitigating the safety-net consequences of regulation-induced innovation," Review of Social Economy, vol. 68, no. 2, pp. 145-161, 2010. https://doi.org/10.1080/00346760902968412 [27] E. N. Karimalis and N. K. Nomikos, "Measuring systemic risk in the European banking sector: A copula CoVaR approach," The European Journal of Finance, vol. 24, no. 11, pp. 944-975, 2018. https://doi.org/10.1080/1351847X.2017.1366350 [28] G. Kou, X. Chao, Y. Peng, F. E. Alsaadi, and E. Herrera Viedma, "Machine learning methods for systemic risk analysis in financial sectors," Technological and Economic Development of Economy, vol. 25, no. 5, pp. 716-742, 2019. https://doi.org/10.3846/tede.2019.8740 [29] C. Lan, Z. Huang, and W. Huang, "Systemic risk in China’s financial industry due to the COVID-19 pandemic," Asian Economics Letters, vol. 1, no. 3, p. 18070, 2020. https://doi.org/10.46557/001c.18070 [30] J.-P. Lee, E. M. Lin, J. J. Lin, and Y. Zhao, "Bank systemic risk and CEO overconfidence," The North American Journal of Economics and Finance, vol. 54, p. 100946, 2020. https://doi.org/10.1016/j.najef.2019.100946 https://doi.org/10.1016/j.jbusres.2019.11.020 https://doi.org/10.1016/j.jfineco.2011.12.010 https://doi.org/10.1145/3502289 https://doi.org/10.1016/j.techfore.2021.121191 https://doi.org/10.1093/rfs/hhz148 https://doi.org/10.1016/j.jcorpfin.2021.102129 https://doi.org/10.1111/jofi.12057 https://doi.org/10.7208/chicago/9780226613475.003.0001 https://doi.org/10.2139/ssrn.3366846 https://doi.org/10.1016/j.jbankfin.2010.08.007 https://doi.org/10.1177/0001839214526755 https://doi.org/10.3389/frai.2018.00001 https://doi.org/10.1111/j.1540-6261.1993.tb05128.x https://doi.org/10.1162/rest_a_00754 https://doi.org/10.3233/JIFS-179039 https://doi.org/10.1002/jae.800 https://doi.org/10.1016/j.jfineco.2010.08.015 https://doi.org/10.1016/j.jebo.2021.12.020 https://doi.org/10.1111/j.1540-6261.1984.tb03667.x https://doi.org/10.1080/00346760902968412 https://doi.org/10.1080/1351847X.2017.1366350 https://doi.org/10.3846/tede.2019.8740 https://doi.org/10.46557/001c.18070 https://doi.org/10.1016/j.najef.2019.100946 567 Edelweiss Applied Science and Technology ISSN: 2576-8484 Vol. 9, No. 3: 553-567, 2025 DOI: 10.55214/25768484.v9i3.5254 © 2025 by the authors; licensee Learning Gate [31] X. Li, D. Tripe, C. Malone, and D. Smith, "Measuring systemic risk contribution: The leave-one-out z-score method," Finance Research Letters, vol. 36, p. 101316, 2020. https://doi.org/10.1016/j.frl.2019.101316 [32] S. Liang, F. Moreira, and J. Lee, "Diversification and bank stability," Economics Letters, vol. 193, p. 109312, 2020. https://doi.org/10.1016/j.econlet.2020.109312 [33] D. Marek, R. Patrik, G. Veronika, and F. Marina, "Economic impacts of Covid-19 on the labor market and human capital," Terra Economicus, vol. 18, no. 4, pp. 78-96, 2020. https://doi.org/10.23683/2073-6606-2020-18-4-78-96 [34] T. Mizuta, "Artificial intelligence (ai) for financial markets: a good ai for designing better financial markets and a bad ai for manipulating markets," in Digital Designs for Money, Markets, and Social Dilemmas: Springer. https://doi.org/10.1007/978-3-030-60416-3_15, 2022, pp. 305-329. [35] W. Noonpakdee, "The adoption of artificial intelligence for financial investment service," in 2020 22nd International Conference on Advanced Communication Technology (ICACT), 2020: IEEE, pp. 396-400. [36] D. Petrone, N. Rodosthenous, and V. Latora, "An AI approach for managing financial systemic risk via bank bailouts by taxpayers," Nature Communications, vol. 13, no. 1, p. 6815, 2022. https://doi.org/10.1038/s41467-022-34199-0 [37] P. Aghion, B. F. Jones, and C. I. Jones, Artificial intelligence and economic growth. Cambridge, MA: National Bureau of Economic Research, 2017. [38] G. Pin and S. Yue, "The impact of internet finance on commercial banks’ risktaking: Theoretical interpretation and empirical test," China Finance and Economic Review, vol. 5, no. 3, pp. 89-109, 2016. [39] Y. Qi and J. Xiao, "Fintech: AI powers financial services to improve people's lives," Communications of the ACM, vol. 61, no. 11, pp. 65-69, 2018. https://doi.org/10.1145/3233137 [40] Y. Qian, J. Liu, L. Shi, J. Y. Forrest, and Z. Yang, "Can artificial intelligence improve green economic growth? Evidence from China," Environmental Science and Pollution Research, vol. 1, pp. 1-20, 2022. https://doi.org/10.1007/s11356-022-22599-4 [41] S. Shah, "The principal-agent problem in finance. CFA Institute Research Foundation L2014-1," 2014. [42] T. Schneider, P. Strahan, and J. Yang, "Bank stress testing, human capital investment and risk management," National Bureau of Economic Research. https://doi.org/10.3386/w30867, 2023. [43] A. Shleifer and R. W. Vishny, "Unstable banking," Journal of Financial Economics, vol. 97, no. 3, pp. 306-318, 2010. https://doi.org/10.1016/j.jfineco.2010.03.001 [44] Y. Sun and X. Tang, "The impact of digital inclusive finance on sustainable economic growth in China," Finance Research Letters, vol. 50, p. 103234, 2022. https://doi.org/10.1016/j.frl.2022.103234 [45] Z. Temelkov, "Fintech firms opportunity or threat for banks?," International Journal of Information, Business and Management, vol. 10, no. 1, pp. 137-143, 2018. [46] A. Tobias and M. K. Brunnermeier, "CoVaR," The American Economic Review, vol. 106, no. 7, pp. 1705-1721, 2016. https://doi.org/10.1257/aer.20150575 [47] W. Wagner, "Diversification at financial institutions and systemic crises," Journal of Financial Intermediation, vol. 19, no. 3, pp. 373-386, 2010. https://doi.org/10.1016/j.jfi.2009.10.001 [48] W. Wagner, "Systemic liquidation risk and the diversity–diversification trade-off " The Journal of Finance, vol. 66, no. 4, pp. 1141-1175, 2011. https://doi.org/10.1111/j.1540-6261.2011.01665.x [49] J. Wang, Y. Hu, and Z. Zhang, "Skill-biased technological change and labor market polarization in China," Economic Modelling, vol. 100, p. 105507, 2021. https://doi.org/10.1016/j.econmod.2021.105507 [50] Y. Wang, S. Chen, and X. Zhang, "Measuring systemic financial risk and analyzing influential factors: an extreme value approach," China Finance Review International, vol. 4, no. 4, pp. 385-398, 2014. https://doi.org/10.1108/CFRI- 05-2013-0083 [51] S. Wu, M. Tong, Z. Yang, and T. Zhang, "Interconnectedness, systemic risk, and the influencing factors: some evidence from China’s financial institutions," Physica A: Statistical Mechanics and Its Applications, vol. 569, p. 125765, 2021. https://doi.org/10.1016/j.physa.2020.125765 [52] Q. Xu, L. Chen, C. Jiang, and J. Yuan, "Measuring systemic risk of the banking industry in China: A DCC-MIDAS-t approach," Pacific-Basin Finance Journal, vol. 51, pp. 13-31, 2018. https://doi.org/10.1016/j.pacfin.2017.11.008 [53] C.-H. Yang, "How artificial intelligence technology affects productivity and employment: firm-level evidence from Taiwan," Research Policy, vol. 51, no. 6, p. 104536, 2022. https://doi.org/10.1016/j.respol.2022.104536 [54] D. Zhang, S. Mishra, and E. Brynjolfsson, "Artificial intelligence index report," Stanford Institute for Human-Centered Artificial Intelligence, vol. 1, pp. 1-230, 2022. [55] Z. Zhang, D. Zhang, F. Wu, and Q. Ji, "Systemic risk in the Chinese financial system: A copula-based network approach," International Journal of Finance & Economics, vol. 26, no. 2, pp. 2044-2063, 2021. https://doi.org/10.1002/ijfe.1820 https://doi.org/10.1016/j.frl.2019.101316 https://doi.org/10.1016/j.econlet.2020.109312 https://doi.org/10.23683/2073-6606-2020-18-4-78-96 https://doi.org/10.1007/978-3-030-60416-3_15 https://doi.org/10.1038/s41467-022-34199-0 https://doi.org/10.1145/3233137 https://doi.org/10.1007/s11356-022-22599-4 https://doi.org/10.3386/w30867 https://doi.org/10.1016/j.jfineco.2010.03.001 https://doi.org/10.1016/j.frl.2022.103234 https://doi.org/10.1257/aer.20150575 https://doi.org/10.1016/j.jfi.2009.10.001 https://doi.org/10.1111/j.1540-6261.2011.01665.x https://doi.org/10.1016/j.econmod.2021.105507 https://doi.org/10.1108/CFRI-05-2013-0083 https://doi.org/10.1108/CFRI-05-2013-0083 https://doi.org/10.1016/j.physa.2020.125765 https://doi.org/10.1016/j.pacfin.2017.11.008 https://doi.org/10.1016/j.respol.2022.104536 https://doi.org/10.1002/ijfe.1820