Close this search box.
Close this search box.

How machine learning could be used to predict future events?

How machine learning could be used to predict future events?

Artificial Intelligence & Machine Learning

A Review Of ‘Real-life experience: Using ML and distance-to-default to predict distress risk’ by Tobias Hoogteijling & Pim van Vliet

Assessing Tail Risk in Equity Markets: A Review of Machine Learning and Distance-to-Default Measures

Investors pay close attention to tail risk, as the major fluctuations in stock prices can significantly impact investment portfolios. The ability to pinpoint stocks that may undergo substantial price downturns is of paramount importance in financial risk management.

Understanding DtD and ML Signals

The DtD signal, drawing from the Merton model, gauges a firm’s proximity to default, factoring in asset volatility—a crucial element in the assessment of financial stability. This measure has been crucial in Robeco’s Quantitative Equity strategies since 2011, playing a significant role in stock selection, particularly within Conservative Equities strategies.

In 2021, Robeco incorporated an ML risk signal within its Quantitative Equity models, designed to identify potential severe stock price drops. ML’s adaptability to data and its proficiency in recognizing complex relationships make it particularly suited for detecting nonlinear patterns that traditional risk measures might miss.

Hoogteijling & van Vliet compare the performance of traditional beta and volatility measures against DtD and ML signals within Robeco’s investment strategies.

This includes a granular year-by-year examination, stock case studies, power curve assessments, and analyses of high-risk stock portfolios.

The year-by-year analysis since the incorporation of DtD in 2011 reveals that while high-risk stocks sometimes lead in bullish markets, they tend to underperform during market downturns, with DtD consistently outstripping traditional risk metrics.

Case studies from 2022 illustrate the practical applications of these signals.

Notably, firms like Silicon Valley Bank and Snap, which were flagged early on by poor DtD and ML risk scores, experienced significant market value drops. As a result, demonstrating the predictive strength of these advanced measures. See our piece: How Did Silicon Valley Bank Fail?

Power curves, a statistical tool for prediction accuracy, show that both DtD and ML signals demonstrate greater predictive ability than traditional risk indicators, with ML exhibiting the highest accuracy.

Portfolio analyses focused on the highest-risk stocks as determined by the three indicators further confirm the enhanced effectiveness of DtD and ML measures. These stocks typically lag behind the market, but the advanced measures identified them more accurately, suggesting their potential to improve returns through risk avoidance or short-selling strategies.

Hoogteijling & van Vliet’s detailed evaluation of DtD and ML risk indicators underscores their value in predicting tail risk in equity investment strategies. The advanced measures have proven their worth, with real-life applicability and superior performance compared to traditional metrics. This insight is invaluable for strategic investment planning and robust risk management, aiming to safeguard portfolios from severe market downturns.

Download the full paper: Real-life experience: Using ML and distance-to-default to predict distress risk (