Using neural networks to improve cross-sectional momentum strategy performance

Loading...
Thumbnail Image

URL

Journal Title

Journal ISSN

Volume Title

School of Business | Master's thesis

Date

2019

Major/Subject

Mcode

Degree programme

Finance

Language

en

Pages

39+5

Series

Abstract

Machine learning and in particular, neural networks, are powerful tools for predictive analysis and used in a wide range of applications. I combine the neural networks and momentum investing with the aim to construct a neural network enhanced momentum strategy. I train the neural network to predict future relative performance of stocks in US between 1964 and 2012 using past returns and risk indicators as a feature set. I find that neural network momentum strategy generates monthly average returns of 3.18% (t-stat 11.81), outperforming the traditional momentum strategy by 1.39%-points. I also find that long-term momentum is the main driver of the relative performance predictions and that short and mid-term historical volatilities are the most important risk factors explaining variation in the predictions.

Description

Thesis advisor

Nyberg, Peter

Keywords

neural, networks, neural networks, momentum, momentum crashes

Other note

Citation