Mathis, Class of 2022, presented research on predicting Federal Reserve rate decisions using causal modelling at the 19th Financial Risks International Forum in Paris. The work addresses the limits of traditional models in high-dimensional macroeconomic environments.
Monetary policy signals shape financial markets, yet modelling their dynamics remains complex as data volumes and interdependencies grow.
Monetary policy as a driver of financial markets
Decisions taken by the Federal Open Market Committee (FOMC) influence borrowing costs, asset pricing and investment strategies across global markets. Beyond the immediate adjustment of the federal funds rate, forward guidance also affects expectations and risk perception.
These mechanisms place monetary policy at the centre of macro-financial dynamics. Anticipating rate changes therefore remains a key objective for researchers and practitioners working in quantitative finance.
At the 19th Financial Risks International Forum, held in Paris on March 30 and 31, 2026, this question was addressed through a research contribution presented by Mathis Guenet, class of 2022, PhD candidate at De Vinci Research Center (De Vinci Higher Education). The work, co-authored with Matthieu Garcin and Martino Grasselli, introduces a causal framework to improve the prediction of policy rate adjustments.
Mathis Guenet during his presentation
A research forum focused on hidden financial risks
The International Risks Forum, organised by the Institut Louis Bachelier in cooperation with the Fondation du Risque and the Europlace Institute of Finance, brings together researchers and industry experts working on financial risk analysis.
The 2026 edition focused on “Hidden Financial Risks”. This theme reflects transformations in financial systems driven by technological progress, globalisation and the emergence of new data sources.
Quantitative finance, artificial intelligence and decentralised finance have contributed to the development of models capable of processing large volumes of information. At the same time, these evolutions have revealed limits in traditional approaches, particularly when risks are nonlinear, embedded in complex systems or poorly observable.
The forum provided a setting for discussing new methods to detect and model these risks more effectively.
Limits of traditional models in high-dimensional data
Macroeconomic datasets are characterised by numerous variables that interact with each other. These interdependencies make it difficult to distinguish causal relationships from simple correlations.
Standard regression models face several challenges in this context. Confounding bias may arise when relevant variables are omitted, while mediation bias can appear when intermediate variables are incorrectly controlled.
Another difficulty lies in the structure of the data itself. Macroeconomic datasets are often high-dimensional, meaning they include many variables, but the number of observations remains limited. This imbalance can lead to models that perform well on historical data but fail to generalise to new situations.
A causal modelling approach using directed graphs
To address these issues, the research presented at the forum employs a causal approach using Directed Acyclic Graphs (DAGs). Instead of including all available variables in a regression, the method focuses on identifying the direct causes of the target variable.
In a causal graph, each variable is conditionally independent of its indirect causes once its direct causes are considered. This property enables the isolation of a reduced set of predictors that contains the relevant information for prediction.
Applied to predicting policy rate adjustments, this approach yields a more parsimonious model. Variables that do not contribute additional information are excluded, reducing noise and improving interpretability.
Improved predictive performance and robustness
The results presented show that models based on direct causal relationships achieve in-sample performance comparable to that of standard regressions. However, the gap between in-sample and out-of-sample performance is reduced.
This translates into better predictive accuracy when the model is applied to new data. By mitigating overfitting, the causal approach yields more robust forecasts of Federal Reserve decisions across time horizons.
Beyond this specific application, the research highlights the relevance of causal modelling in environments where data is complex and highly interconnected.
Research at the intersection of finance and data science
This contribution reflects broader developments in quantitative finance, where statistical methods are increasingly combined with causal reasoning and machine learning techniques.
The Institut Louis Bachelier, which hosts more than 70 research programs across environmental, digital, demographic and financial transitions, provides a framework for collaborations between academic and industry partners. Each year, several hundred researchers contribute to these initiatives.
Presentations such as this one highlight the link between engineering training and advanced research topics, particularly at the intersection of data science and financial economics.
This dynamic is supported by the De Vinci Research Center, which brings together the research activities of De Vinci Higher Education through four areas: business, digital, finance and modelling, along with a dedicated partnership research unit.
Learn more about ESILV’s programmes