Footnote:
In: Machine Learning and Data Sciences for Financial Markets: A Guide to Contemporary Practices (2023). Edited by A. Capponi and C.A. Lehalle. Cambridge University Press
Nach Informationen von SSRN wurde die ursprüngliche Fassung des Dokuments July 25, 2021 erstellt
Description:
AI/ML models are used for many financial applications ranging from portfolio selection to efficient credit allocation. However, the drawback to applying these models in practice is that performance (i.e., predictive power) is generally inversely related to model complexity. In this chapter, we formalize the problem that a risk manager faces when trading off the benefits of model performance with the drawbacks of complexity. Because it is difficult to define a single metric that quantifies complexity across all settings, we use several metrics to capture model complexity. We use the case of a volatility managed portfolio to show the impact of model complexity on performance by constructing a mean-variance-complexity efficient frontier. We find that for any level of risk, there is a positive relation between model complexity and portfolio return. Moreover, we show that model complexity affects the classic mean-variance efficient frontier. For low levels of complexity, the risk-return tradeoff is an inverted u-shape, while at high levels of complexity the risk-return tradeoff is positive