Distinguishing skills from luck in trading and investment

Ma, Tren (2023) Distinguishing skills from luck in trading and investment. PhD thesis, University of Glasgow.

Abstract

This thesis develops a series of statistical frameworks to control type I errors in picking out-performers. It consists of three independent essays which assess performance of U.S. equity mutual funds, currency trading strategies and hedge funds.

The first essay introduces a novel multiple hypothesis testing method named the functional False Discovery Rate“plus” (fFDR+). The method incorporates informative covariates in estimating the False Discovery Rate (FDR)of predictive models’ “conditional” performance. In simulations, the fFDR + control swell the FDR and gains considerable power over prior methods that do not account for extra information. In empirical analyses, we construct portfolios based on several covariates and show that they enhance the performance of mutual fund portfolios, highlighting the value of extra information in the multiple hypothesis testing framework.

The second essay develops the multivariate functional false discovery rate (mfFDR) method that accounts for multiple informative covariates to examine the conditional performance of predictive models and gain a considerably higher power than prior methods including the one in the first essay. The proposed method is then applied to control luck in detecting profitable technical trading rules using 30 developed and emerging market currencies. It selects more profitable rules than prior methods; more importantly, these rules offer better out-of-sample performance.

The third essay introduces a new procedure to control for family error rate (FWER) in picking out-performers. The method utilizes multiple side information to more precisely estimate the FWER and gains much higher power in detecting out-performers compared to existing ones. In empirical analyses, the method allows investors picking out-performing hedge funds with very low FWER. The portfolios of hedge funds selected by the method beat passive benchmarks in various settings. Further analyses show that the new method detects truly out-performing hedge fund managers who can repeat their past performance over a long horizon.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Subjects: H Social Sciences > HG Finance
• Colleges/Schools: College of Social Sciences > Adam Smith Business School
• Supervisor's Name: Ewald, Professor Christian and Sermpinis, Professor Georgios
• Date of Award: 2023
• Depositing User: Theses Team
• Unique ID: glathesis:2023-83873
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 18 Oct 2023 09:51
• Last Modified: 20 Oct 2023 10:22
• Thesis DOI: 10.5525/gla.thesis.83873
• URI: https://theses.gla.ac.uk/id/eprint/83873