International Asset Allocation with Regime Switching: Evidence from the ETFs

• Published date: 01 October 2015
• Author: Pan Jiang,Qingfu Liu,Yiuman Tse
• Date: 01 October 2015
• DOI: http://doi.org/10.1111/ajfs.12109

International Asset Allocation with Regime

Switching: Evidence from the ETFs*

Pan Jiang

Institute for Financial Studies, Fudan University

Qingfu Liu**

Institute for Financial Studies, Fudan University

Yiuman Tse

College of Business Administration, University of Missouri – St. Louis

Received 5 October 2014; Accepted 6 July 2015

Abstract

We develop a dynamic investment strategy with Markov regime switching (MRS) in asset alloca-

tion with international iShares exchange-traded funds (ETFs). Using daily ETF data, we show

that a portfolio based on the dynamic MRS strategy outperforms one based on static mean-vari-

ance strategies after transaction costs. This dynamic investment strategy not only captures the

regime shifts in the highly frequent trading process but also can be practically used with tradable

ETFs. We investigate the reasons for predictive misjudgments and assess the contribution of

each regime’s investment strategy, providing insight into the characteristics of the MRS model

and modifying our views on why the MRS strategy outperforms traditional strategies.

Keywords Asset allocation; International ETF; Markov regime switching

JEL Classification: C12, F31, G15

1. Introduction

Various studies have shown that international equity markets exhibit higher

volatility and correlations during market downturns than during market upturns.

Several recent studies have considered these asymmetric effects while examining

international asset allocation with Markov regime-switching (MRS) models (e.g.,

Ang and Bekaert, 2004; Guidolin and Timmermann, 2008; Kritzman et al., 2012;

*We thank the editor and anonymous referees for their insightful comments and suggestions.

Liu acknowledges the supports from the National Nature Science Funds of China

(71473042). This research started when Tse was visiting Fudan University.

**Corresponding authors: Qingfu Liu, Institute for Financial Studies, Fudan University, Shang-

hai, 200-433 Handan Rd, China. Tel: 86-21-6564-3821, Fax: 86-21-6511-2913, email: liuqf@

fudan.edu.cn.

Asia-Pacific Journal of Financial Studies (2015) 44, 661–687 doi:10.1111/ajfs.12109

©2015 Korean Securities Association 661

Dou et al., 2014). The overall results show that the dynamic MRS models out-

perform static mean-variance allocation.

In this study, we develop a dynamic investment strategy with MRS for asset allo-

cation with international iShares exchange-traded funds (ETFs). iShares are ETFs

that represent diversified portfolios tracking a specific market index. They have

become popular investment vehicles in recent years, as they represent diversified

securities portfolios and have the best qualities of both closed and open-ended

mutual funds. We consider different transaction costs and risk preferences in the

model for both in-sample and out-of-sample analyses. We find a difference in

regime judgment between the out-of-sample prediction and the robustness test,

which leads us to identify the reason for the misjudgments. We then examine the

contributions of the strategies under different regimes.

Our research contributes to the literature in several ways. First, our data sample

is composed of 21 iShares series, covering market information about Asia, Europe,

and America. A cross-region portfolio made up of these ETFs greatly lowers the

nonsystematic risk and thus conveys more information about the world’s systematic

risk. ETFs are convenient trading vehicles that can be traded like stocks. But unlike

stocks, ETFs can be short sold on a downtick (Elton et al., 2002). Prior studies use

MSCI global equity indexes, which are not tradable. In addition, the ETFs listed on

the New York Stock Exchange (NYSE) can be traded during synchronous US trad-

ing hours in dollars.

Second, we select daily ETF data, which are different from the monthly data

used by Ammann and Verhofen (2006), Ang and Bekaert (2002), Dou et al. (2014),

and Nilsson and Graflund (2003). Although portfolio adjustments based on

monthly data may avoid the high costs of frequent regime shifts, they lose more

information than do adjustments based on daily data and thus may not be sensitive

to the changes occurring in financial markets. We use daily data to examine

whether transaction costs offset the benefits of the investment strategies.

Third, we construct a “world portfolio” for regime estimation in a simple way—

by constructing a volume-weighted mean portfolio. Note that this “world portfolio”

itself is static and functions only as an instrument for the final dynamic portfolio allo-

cation. Following Ang and Bekaert (2004) and Dou et al. (2014), we apply an inte-

grated portfolio’s returns as the only switching variable in the in-sample Markov

regime-switching estimation. This switching variable is better than that used in Kritz-

man et al. (2012) because it includes all the information for 21 iShares and diversifies

most of the nonsystematic risk. As our investment strategy is based on such an inte-

grated portfolio, we call the strategy “systematic-risk-adaptive strategy.”

1

1

Portfolio allocation can also be based on each ETF’s own regime switching and its correla-

tion with other ETFs. However, this method uses a multivariate expected return vector and

covariance matrix and, therefore, requires a much more complicated investment strategy.

Thus, we use a volume-weighted mean portfolio in this paper.

P. Jiang et al.

662 ©2015 Korean Securities Association