A Random Forests Based Performance Ratio for Regulatory Asset Portfolio Management and Optimization

Abstract

The following paper proposes a portfolio performance measure to optimize, mostly bond asset portfolios usually held for regulatory purposes from a risk focused perspective. The measure is based on variations of the proximity measure introduced by the Random Forests framework, leading to a proximity based performance ratio. The proximities are modeled using a recursive conditional partitioning type of Random Forests, which allows for a ranking as well as an analysis of the risk drivers of the portfolio performance. The proximity based performance ratio is shown to, on average, outperform nine different and commonly known risk and performance ratios as well as the 1/N-balanced portfolio in three different tests, in- and out of the sample. The proximity based performance ratio can consider a large amount of risk rivers and is suitable for big data analysis for big and small financial institutions.

Keywords: Random Forests; Recursive Conditional Partitioning; Portfolio Optimisation; Sharpe Ratio; Big Data

JEL Classification: C02, D81, G11, G32