AI, a tool for better predicting company bankruptcies

What will happen in the coming months, with slow growth and in an economic and political context that remains marked by uncertainty? For the most accurate understanding of economic and financial developments, traditional statistical models used by the financial industry could be complemented with new tools based on artificial intelligence (AI). Scientific research is questioning the advantages and disadvantages of these AI models.

Indeed, it is worth noting that among the various applications of AI, one area is particularly noticeable: the prediction of various phenomena, ranging from weather conditions to stock market trends to the prediction of business bankruptcies. Using machine learning algorithms, AI, fueled by a large amount of data, is trained to recognize early signs of financial difficulties and thus assess the risk of bankruptcy. Furthermore, the various models for predicting business bankruptcy incorporating AI are gradually improving as they integrate new information.

[An article from The Conversation written byVanessa Serret,Professor, IAE Metz School of Management – University of Lorraine and Sami Ben Jabeur,Associate Professor, ESDES Lyon Business School, Catholic University of Lyon (UCLy)]

Surpassed classical models

Concretely, the different AI models used in bankruptcy prediction are fueled by financial data (balance sheets, income statements, debts, leverage, liquidity ratios, etc.), operational data (cash flow, company size, industry sector, supply chain, etc.), and external data (market conditions, competition, technological changes, economic indicators, etc.).

One of our studies proposes a new approach to bankruptcy prediction based on hybrid machine learning, exploiting the complementarity of traditional methods and AI. It emerges that this approach exceeds the classical statistical methods used by bankers, allowing for improved prediction accuracy. For example, among all the models tested over a one-year horizon, hybrid machine learning was the most accurate with an exceptional prediction power reaching 95.11%.

Another study has highlighted the qualities of a sophisticated machine learning algorithm called CatBoost, designed to classify and predict business bankruptcies. Compared and tested against eight other traditional methods, this approach was applied to data related to business bankruptcies in France between 2014 and 2016. The average prediction over one, two, and three years from the eight classical models varies between 65.38% and 80.34%, while the average prediction of the AI model is 82.90%, highlighting its superiority. Furthermore, this AI model can take into account textual data such as the texts of financial and extra-financial activity reports, which have become essential in understanding corporate strategies.

The various models for predicting business bankruptcies incorporating AI are gradually improving as they integrate new information.
Pexels, CC BY-SA

Finally, a study confirms the superiority of AI models in terms of prediction accuracy. This study compares the performance of eight classical models in predicting the bankruptcy of French companies between 2013 and 2017 with another AI model called XGBoost, which correctly predicts the financial distress of a company on average at a rate of 84.7%. It has the advantage of detecting signs of failure up to five years before bankruptcy. Furthermore, this model is highly interpretable and reveals the relative importance of various financial characteristics (especially accounting variables), increasing its transparency and credibility with stakeholders.

The prospects for predicting business bankruptcy with AI are therefore promising. In future years, AI techniques could be capable of analyzing risks in real time, thus facilitating a more prompt and appropriate response of companies to changing environments. They could especially incorporate additional data sources such as social media trends, geopolitical events, or environmental, social, and governance factors to enrich their analyses.

Read also >> Does the brain choose money or morals when negotiating?

Important challenges to overcome

However, global economic upheavals, such as the Covid-19 pandemic, have revealed the ability of rapid and unforeseen changes to drastically affect the economic landscape, thus challenging all predictive models. AI is not the solution to everything. Firstly, the accuracy of these tools depends heavily on the availability of abundant, reliable, and high-quality data. Furthermore, the increasing complexity and interconnection of global supply chains make bankruptcy prediction increasingly difficult.

In general, during major crises, the reliability of AI models based on historical data may be compromised, as this data may not necessarily reflect the current or future state of markets and operational conditions.

The increasing sophistication of AI tools also raises questions about their credibility. To maintain the confidence of economic and financial sector stakeholders and ensure compliance with current regulations, it is essential for these advanced systems to remain understandable and transparent in their decision-making processes, at the risk of raising ethical issues that could lead to their banishment.

Finally, although AI represents assistance in predicting bankruptcy, its effectiveness is dependent on analysis by business experts. In-depth knowledge and understanding of business ecosystems are indeed indispensable. They are a prerequisite for building and interpreting these tools so that decisions are not only based on reliable data, but also on a strategic and contextual vision.

Based on current knowledge, AI models in predicting financial distress are imperfect but help to reduce the uncertainty surrounding business ecosystems. In this regard, they can improve the coordination of stakeholders such as banks, regulators, etc.

Read also >>How to lower the cost of electric car consumption?

The Conversation

Leave a Reply

Your email address will not be published. Required fields are marked *