Selected work

Representative Papers

It Takes Two to Tango, but More to Assess Systemic Risk: Credit Networks Through the Lens of Hypergraphs

Jul 12, 2026

Traditional binary networks struggle to capture the multilateral credit linkages arising when multiple financial institutions are simultaneously exposed to the same firm, thereby limiting the accurate assessment of systemic risk. This study addresses this gap by introducing a hypergraph framework into financial credit networks, constructing a bank–firm multilateral credit relationship model using loan registry data from the Central Bank of Argentina. The authors propose a nonlinearly weighted H-eigenvector centrality measure that integrates both neighboring institution centrality and loan exposure amounts, and combine it with an enhanced DebtRank algorithm to evaluate shock propagation effects. Empirical results demonstrate that institutions identified as systemically important under this approach exhibit significantly stronger risk amplification during adverse shocks, offering financial regulators a more precise and complementary analytical tool for systemic risk monitoring.

0 citationsRead paper

Geopolitics, Geoeconomics and Risk:A Machine Learning Approach

Oct 14, 2025

This study examines the predictive power of news-derived geopolitical risk (GPR) and economic policy uncertainty (EPU) indices for sovereign credit risk—measured by CDS spreads—and assesses their incremental value over conventional indicators such as VIX and Fed rate expectations. Using daily panel data from 42 countries, we construct a high-frequency, multidimensional GPR–EPU framework integrating news text analytics with nonlinear machine learning models—particularly random forests—to detect non-linear interaction effects and region-specific heterogeneity in risk transmission. Results demonstrate that news-based indicators significantly improve out-of-sample CDS spread forecasting accuracy; random forests consistently outperform traditional econometric models; and the impact of GPR and EPU exhibits state dependence, intensifying markedly during market turmoil. The study provides novel empirical evidence and a methodological framework for modeling unstructured information–driven sovereign risk, advancing both financial economics and macro-financial early-warning systems.

0 citationsRead paper

Causal Interventions in Bond Multi-Dealer-to-Client Platforms

Jun 22, 2025

This paper addresses the core challenge in multi-to-multi bond trading platforms where dealers cannot observe counterparties’ quotes and struggle to ensure profitability. We propose the first general analytical framework integrating causal inference with probabilistic graphical models. Methodologically, we distinguish generative versus discriminative modeling approaches, explicitly capture the dynamic impact of RFQ (Request-for-Quote) negotiation mechanisms, identify key pricing drivers via causal interventions, and design prediction evaluation metrics tailored for optimal pricing. Our contribution lies in the first incorporation of structured causal modeling into electronic RFQ decision-making—overcoming the limitations of traditional black-box predictive models. Empirical results demonstrate significant improvements in price prediction accuracy and revenue estimation reliability, thereby enhancing dealers’ pricing capability and profitability under information asymmetry.

0 citationsRead paper

What Can 240,000 New Credit Transactions Tell Us About the Impact of NGEU Funds?

Mar 16, 2025

This paper investigates the causal impact of the NextGenerationEU (NGEU) fund—delivered via public procurement—on credit markets, specifically examining whether NGEU-related procurement stimulates new credit more effectively than conventional procurement and identifying heterogeneous responses across firm size, industry, loan maturity, and value-chain position. Leveraging precise matching between 240,000 high-frequency credit transactions and NGEU procurement contracts, we estimate dynamic policy effects using a panel local projections model with multidimensional fixed effects. This study pioneers the integration of high-frequency fiscal and financial data, revealing that NGEU procurement significantly increases new credit, with an effect magnitude approximately 30% larger than traditional procurement. The differential impact stems primarily from higher actual fund utilization rates rather than inherent policy design features. Our findings provide novel empirical evidence and a methodological framework for “precision irrigation” of public funds, combining rigorous causal identification with direct policy relevance.

0 citationsRead paper

REX: Causal Discovery based on Machine Learning and Explainability techniques

Jan 22, 2025

Existing causal discovery methods struggle to simultaneously achieve high accuracy and interpretability, particularly under complex, noisy data conditions. To address this, we propose REX—the first method to deeply integrate Shapley values into a machine learning–driven causal graph learning framework, unifying causal structure identification with quantitative, attribution-based interpretation of individual causal edges. REX synergizes neural network modeling capacity with the local interpretability of Shapley values, supporting both nonlinear functional relationships and additive noise models; it further enhances robustness via edge significance assessment. Experiments demonstrate that REX consistently outperforms state-of-the-art methods on synthetic benchmarks. On the Sachs single-cell protein signaling dataset, REX achieves an accuracy of 0.952 with zero false-positive edges—marking substantial improvements in both reliability and interpretability of causal discovery.

0 citationsRead paper

Recent publications

Latest Papers

It Takes Two to Tango, but More to Assess Systemic Risk: Credit Networks Through the Lens of Hypergraphs

Jul 12, 2026

Traditional binary networks struggle to capture the multilateral credit linkages arising when multiple financial institutions are simultaneously exposed to the same firm, thereby limiting the accurate assessment of systemic risk. This study addresses this gap by introducing a hypergraph framework into financial credit networks, constructing a bank–firm multilateral credit relationship model using loan registry data from the Central Bank of Argentina. The authors propose a nonlinearly weighted H-eigenvector centrality measure that integrates both neighboring institution centrality and loan exposure amounts, and combine it with an enhanced DebtRank algorithm to evaluate shock propagation effects. Empirical results demonstrate that institutions identified as systemically important under this approach exhibit significantly stronger risk amplification during adverse shocks, offering financial regulators a more precise and complementary analytical tool for systemic risk monitoring.

0 citationsRead paper

Geopolitics, Geoeconomics and Risk:A Machine Learning Approach

Oct 14, 2025

This study examines the predictive power of news-derived geopolitical risk (GPR) and economic policy uncertainty (EPU) indices for sovereign credit risk—measured by CDS spreads—and assesses their incremental value over conventional indicators such as VIX and Fed rate expectations. Using daily panel data from 42 countries, we construct a high-frequency, multidimensional GPR–EPU framework integrating news text analytics with nonlinear machine learning models—particularly random forests—to detect non-linear interaction effects and region-specific heterogeneity in risk transmission. Results demonstrate that news-based indicators significantly improve out-of-sample CDS spread forecasting accuracy; random forests consistently outperform traditional econometric models; and the impact of GPR and EPU exhibits state dependence, intensifying markedly during market turmoil. The study provides novel empirical evidence and a methodological framework for modeling unstructured information–driven sovereign risk, advancing both financial economics and macro-financial early-warning systems.

0 citationsRead paper

Causal Interventions in Bond Multi-Dealer-to-Client Platforms

Jun 22, 2025

This paper addresses the core challenge in multi-to-multi bond trading platforms where dealers cannot observe counterparties’ quotes and struggle to ensure profitability. We propose the first general analytical framework integrating causal inference with probabilistic graphical models. Methodologically, we distinguish generative versus discriminative modeling approaches, explicitly capture the dynamic impact of RFQ (Request-for-Quote) negotiation mechanisms, identify key pricing drivers via causal interventions, and design prediction evaluation metrics tailored for optimal pricing. Our contribution lies in the first incorporation of structured causal modeling into electronic RFQ decision-making—overcoming the limitations of traditional black-box predictive models. Empirical results demonstrate significant improvements in price prediction accuracy and revenue estimation reliability, thereby enhancing dealers’ pricing capability and profitability under information asymmetry.

0 citationsRead paper

What Can 240,000 New Credit Transactions Tell Us About the Impact of NGEU Funds?

Mar 16, 2025

This paper investigates the causal impact of the NextGenerationEU (NGEU) fund—delivered via public procurement—on credit markets, specifically examining whether NGEU-related procurement stimulates new credit more effectively than conventional procurement and identifying heterogeneous responses across firm size, industry, loan maturity, and value-chain position. Leveraging precise matching between 240,000 high-frequency credit transactions and NGEU procurement contracts, we estimate dynamic policy effects using a panel local projections model with multidimensional fixed effects. This study pioneers the integration of high-frequency fiscal and financial data, revealing that NGEU procurement significantly increases new credit, with an effect magnitude approximately 30% larger than traditional procurement. The differential impact stems primarily from higher actual fund utilization rates rather than inherent policy design features. Our findings provide novel empirical evidence and a methodological framework for “precision irrigation” of public funds, combining rigorous causal identification with direct policy relevance.

0 citationsRead paper

REX: Causal Discovery based on Machine Learning and Explainability techniques

Jan 22, 2025

Existing causal discovery methods struggle to simultaneously achieve high accuracy and interpretability, particularly under complex, noisy data conditions. To address this, we propose REX—the first method to deeply integrate Shapley values into a machine learning–driven causal graph learning framework, unifying causal structure identification with quantitative, attribution-based interpretation of individual causal edges. REX synergizes neural network modeling capacity with the local interpretability of Shapley values, supporting both nonlinear functional relationships and additive noise models; it further enhances robustness via edge significance assessment. Experiments demonstrate that REX consistently outperforms state-of-the-art methods on synthetic benchmarks. On the Sachs single-cell protein signaling dataset, REX achieves an accuracy of 0.952 with zero false-positive edges—marking substantial improvements in both reliability and interpretability of causal discovery.

0 citationsRead paper