A Statistical and Machine Learning Approach for Assessing the Impact of Financial Research Reports on Clients’ Trading Behaviour
Year: 2024
Type: dissertation
Abstract: BNP Paribas is a market maker that buys and sells securities on its account. Market makers play a crucial role in the market by facilitating securities trading and providing liquidity. BNP has a research team that analyses different securities and makes investment suggestions, which are grouped in a document called a research report. Those are sent to BNP’s clients, aiming to lead them to perform a transaction with the bank. This master dissertation seeks to determine the impact that BNP Paribas research reports containing bond trade suggestions have on clients' trade behaviour, particularly in requesting a quote (RFQ) and, subsequently, in performing a transaction. Based on a review of the literature, a problem that requires an explanation or a prediction answer leads to some differences in the model building process and the models that are advisable to be used. This research problem fits into the explanatory paradigm. The study starts with the search for variables associated with a higher likelihood of requesting an RFQ. The logistic regression model is used as an explanatory tool and assessed with standard statistical measurements like goodness-of-fit tests and residual analysis. The random forest model is then compared with the logistic regression using probability metrics rather than threshold ones aligned with the explanatory nature of the research question. Results show similar performance between the two models, with the logistic regression model ultimately chosen for its probabilistic framework. The logistic regression model answers the research question, showing that BNP clients who download a research report are 12% more likely to request an RFQ. However, the reports seem to not influence whether they choose to trade with BNP or another market maker. Another noteworthy finding is that clients who previously purchased a specific bond security before the first time it was suggested by the bank are 36% more likely to request an RFQ. This work also addresses the challenge of establishing causation, acknowledging the complexity of attributing the observed increase in RFQ requests directly to report downloads. Recommendations for future studies involve exploring causality through randomized control trials or other advanced methods. In conclusion, this research underscores the importance of both having good business knowledge for assessing the right variables and technical expertise in modelling them, with an emphasis on knowing whether the research question falls within an explanatory or predictive paradigm.
Author: Peixoto, Diogo Tomás dos Santos
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7005891031