Learning from past out-of-sample errors

: an application of the Galton method in the United Kingdom stock market

Abstract

This Thesis is focused on an out-of-sample application of the Galton strategy in the United Kingdom stock market from January 1996 to December 2022. This allocation, exploiting the lack of perfect randomness of past out-of-sample errors, is able to provide a useful alternative to the classical plug-in approach in portfolio optimization. A crucial advantage of this investment strategy is the better risk-adjusted performance with respect to the benchmark and a classic version of Mean-Variance portfolio in terms of a higher Sharpe and Sortino ratio. Galton risk estimates are not too optimistic in predicting future volatility as opposed to competitors, recording a ratio of realized volatility over its ex-ante expectation close to unity. Five hundred random horse races confirm these results. In this context, the Ledoit and Wolf portfolio is the sole competitor beating in 26% and 61% of cases the Galton GMV and MV versions in terms of annualized Sharpe ratio. From a Risk Management perspective, Galton allocations have the best VaR hit rates at 95% and 99% confidence levels.

Date of Award	28 Jun 2023
Original language	English
Awarding Institution	Universidade Católica Portuguesa
Supervisor	Pedro Barroso (Supervisor)