Machine learning methods have achieved great success in modeling complex patterns in finance such as asset pricing and credit risk that enable them to outperform statistical models. In addition to the predictive accuracy of machine learning methods, the ability to interpret what a model has learned is crucial in the finance industry. We address this challenge by adapting interpretable machine learning to the context of corporate bond recovery rate modeling. In addition to the best performance, we show the value of interpretable machine learning by finding drivers of recovery rates and their relationship that cannot be discovered by the use of traditional machine learning methods. Our findings are financially meaningful and consistent with the findings in the existing credit risk literature

This paper explores ways to improve the existing systemic risk measures by incorporating machine learning algorithms into the measurement. We aim to overcome the shortcomings of existing methods that rely on restricted modeling and are difficult to tap into various data resources. To this end, this paper unifies a dynamic quantification framework for systemic risk and links it to a two-step supervised learning problem, which allows for hierarchical structure of the systemic event and the return dependence. We leverage the generalization and predictive powers of machine learning to statistically model the tail events and the co-movements of the equity returns during the shocks to the macro-economy. Our results show that most machine learning algorithms enhance the systemic risk measure’s predictive power. Numerous comparative and sensitivity backtesting studies for United States and Hong Kong markets are conducted, from which we recommend the best machine learning algorithm for systemic risk measurement

This paper compares the out-of-sample predictive performance of different early warning models for systemic banking crises using a sample of advanced economies covering the past 45 years. We compare a benchmark logit approach to several machine learning approaches recently proposed in the literature. We find that while machine learning methods often attain a very high in-sample fit, they are outperformed by the logit approach in recursive out-of-sample evaluations. This result is robust to the choice of performance metric, crisis definition, preference parameter, and sample length, as well as to using different sets of variables and data transformations. Thus, our paper suggests that further enhancements to machine learning early warning models are needed before they are able to offer a substantial value-added for predicting systemic banking crises. Conventional logit models appear to use the available information already fairly efficiently, and would for instance have been able to predict the 2007/2008 financial crisis out-of-sample for many countries. In line with economic intuition, these models identify credit expansions, asset price booms and external imbalances as key predictors of systemic banking crises

Yes, they can! Machine learning models predict leverage better than linear models and identify a broader set of leverage determinants. They boost the out-of-sample R2 from 36% to 56% over OLS and LASSO. The best performing model (random forests) selects market-to-book, industry median leverage, cash and equivalents, Z-Score, profitability, stock returns, and firm size as reliable predictors of market leverage. More precise target estimation yields a 10%–33% faster speed of adjustment and improves prediction of financing actions relative to linear models. Machine learning identifies uncertainty, cash flow, and macroeconomic considerations among primary drivers of leverage adjustments

This paper uses a comprehensive set of predictor variables from the five largest Eurozone countries to compare the performance of simple univariate and machine learning-based multivariate models in forecasting stock market crashes. In terms of statistical predictive performance, a support vector machine-based crash prediction model outperforms a random classifier and is superior to the average univariate benchmark as well as a multivariate logistic regression model. Incorporating nonlinear and interactive effects is both imperative and foundation for the outperformance of support vector machines. Their ability to forecast stock market crashes out-of-sample translates into substantial value-added to active investors. From a policy perspective, the use of machine learning-based crash prediction models can help activate macroprudential tools in time

We develop a state‐of‐the‐art fraud prediction model using a machine learning approach. We demonstrate the value of combining domain knowledge and machine learning methods in model building. We select our model input based on existing accounting theories, but we differ from prior accounting research by using raw accounting numbers rather than financial ratios. We employ one of the most powerful machine learning methods, ensemble learning, rather than the commonly used method of logistic regression. To assess the performance of fraud prediction models, we introduce a new performance evaluation metric commonly used in ranking problems that is more appropriate for the fraud prediction task. Starting with an identical set of theory‐motivated raw accounting numbers, we show that our new fraud prediction model outperforms two benchmark models by a large margin: the Dechow et al. logistic regression model based on financial ratios, and the Cecchini et al. support‐vector‐machine model with a financial kernel that maps raw accounting numbers into a broader set of ratios

Practitioners allocate substantial resources to technical analysis whereas academic theories of market efficiency rule out technical trading profitability. We study this long-standing puzzle by applying a diverse set of machine learning algorithms. The results show that an investor can find profitable technical trading rules using past prices, and that this out-of-sample profitability decreases through time, showing that markets have become more efficient over time. In addition, we find that the evolutionary genetic algorithm’s attitude in not shying away from erroneous predictions gives it an edge in building profitable strategies compared to the strict loss-minimization-focused machine learning algorithms

We construct real-time machine learning strategies based on a “universe” of fundamental signals. The out-of-sample performance of these strategies is economically meaningful and statistically significant, but considerably weaker than those documented by prior studies that use curated sets of signals as predictors. Strategies based on a simple recursive ranking of each signal’s past performance also yield substantially better out-of-sample performance. We find qualitatively similar results when examining past-return-based signals. Our results underscore the key role of feature engineering and, more broadly, inductive biases in enhancing the economic benefits of machine learning investment strategies

We test the efficient market hypothesis by using machine learning to forecast stock returns from historical performance. These forecasts strongly predict the cross-section of future stock returns. The predictive power holds in most subperiods and is strong among the largest 500 stocks. The forecasting function has important nonlinearities and interactions, is remarkably stable through time, and captures effects distinct from momentum, reversal, and extant technical signals. These findings question the efficient market hypothesis and indicate that technical analysis and charting have merit. We also demonstrate that machine learning models that perform well in optimization continue to perform well out-of-sample

We use machine learning methods and high‐dimensional detailed financial data to predict the direction of one‐year‐ahead earnings changes. Our models show significant out‐of‐sample predictive power: the area under the receiver operating characteristics curve ranges from 67.52% to 68.66%, significantly higher than the 50% of a random guess. The annual size‐adjusted returns to hedge portfolios formed based on the prediction of our models range from 5.02% to 9.74%. Our models outperform two conventional models that use logistic regressions and small sets of accounting variables, and professional analysts’ forecasts. Analyses suggest that the outperformance relative to the conventional models stems from both nonlinear predictor interactions missed by regressions and the use of more detailed financial data by machine learning