Board of Directors' Profile: A Case for Deep Learning as a Valid Methodology to Finance Research

Abstract

This paper presents a Deep Learning (DL) model for natural language processing of unstructured CVs to generate a six-dimensional profile of the professional experience of the Spanish companies' board of directors. We show the complete process starting with open data extraction and cleaning, the generation of a labeled dataset for supervised learning, the development, training and validation of a DL model capable of accurately analyzing the dataset, and, finally, a data analysis work based on the automated generation of the professional profiles of more than 6,000 directors of Spanish listed companies between 2003 and 2020. An RNN-LSTM neural network has been trained in three phases starting from a random initial state, (1) learning of basic structures of the Spanish language, (2) fine tuning for scientific texts in the field of economics and finance, and (3) regression modeling to generate a six-dimensional profile based on a generalization of sentiment classification systems. The complete training has been carried out with very low computational requirements, having a total duration of 120 hours of processing in a low-end GPU. The results obtained in the validation of the DL model show great accuracy, obtaining a value for the standard deviation of the mean error between 0.015 and 0.033. As a result, we have been able to outline with a high degree of reliability the profile of the listed Spanish companies' board of directors. We found that the predominant profile is that of directors with experience in executive or consultancy positions, followed by the financial profile. The results achieved show the potential of DL in social science research, particularly in Finance.

Keywords

• artificial intelligence
• board
• deep learning
• finance
• long short term memory
• recurrent network