Title: Improving index tracking and portfolio performance with regularization Authors:  Dietmar Maringer - University of Basel (Switzerland) [presenting]
Sandra Paterlini - University of Trento (Italy)
Abstract: Historical data are commonly used for portfolio optimization when parametric models fail to capture essential features of asset returns. However, this approach poses a challenge: long time series may include outdated observations that no longer reflect current market conditions, while short series with recent data suffer from small-sample issues and risk overfitting, especially when the asset universe is large. Recent research shows that regularization techniques, such as Lasso (L1), Ridge (L2), or Elastic Net (combining L1 and L2), can enhance portfolio construction. Yet, practical guidance on calibration remains limited, and applications to passive management strategies are rarely explored. This paper addresses these gaps by investigating different regularization techniques in the context of index tracking and enhanced indexing. We conduct empirical studies to evaluate how different calibration choices affect portfolio performance, considering various objectives, training window lengths, and out-of-sample investment horizons. Our findings indicate that regularization can improve out-of-sample tracking error and related metrics, though the magnitude of improvement depends heavily on calibration choices and additional optimization objectives beyond tracking error.