Market Regime Forecasting using Correlation Networks and Machine Learning Classifiers

Public Deposited

Analytics

Last Modified

Creator

Abstract

Market regime classification has been practically influential to financial practitioners. Through the top-down approach to identify market regimes, we partition historical data on U.S. equity market returns into bear and bull markets (Hanna, 2018). Then we create a correlation network for each market state respectively. The topological features of those networks will be fed into the machine learning classifiers, which will provide a model whose machine learning algorithm generates the highest accuracy among all the candidate algorithms. We found that, with 1-months duration and 1-week lagged industry return data, the Fine Tree model has the highest forecasting accuracy of 91.7%. For data with longer horizon, the Kernel Naive Bayes model with 3-month duration and 12-month lagged data performs the best with an accuracy of 66.7%. The applications of the acquired model can not only benefit financial practitioners from acting in advance of the market, but also benefit macro-economists from preparing better for the extreme market events.

Date of publication

Keyword

Subject

DOI

Resource type

Rights statement

Advisor

Aguilar, Michael
- College of Arts and Sciences, Department of Economics
- Other Affiliation: UNC Kenan-Flagler Business School
Fruehwirth, Jane
- College of Arts and Sciences, Department of Economics

Degree

Graduation year

Language

Relations

Parents:

This work has no parents.

In Collection:

Thumbnail	Title	Date Uploaded	Visibility	Actions
	Honor_Thesis_Fan_Zhu.pdf	2020-05-01	Public	Download