The role of tail network topological characteristic in portfolio selection: A TNA‐PMC model

• Published date: 01 March 2023
• Author: Mengting Li,Qifa Xu,Cuixia Jiang,Qinna Zhao
• Date: 01 March 2023
• DOI: http://doi.org/10.1111/irfi.12379

ORIGINAL ARTICLE

The role of tail network topological characteristic

in portfolio selection: A TNA-PMC model

Mengting Li

1

| Qifa Xu

1,2

| Cuixia Jiang

1

| Qinna Zhao

1

1

School of Management, Hefei University of

Technology, Hefei, Anhui, People's

Republic of China

2

Key Laboratory of Process Optimization

and Intelligent Decision-making, Ministry

of Education, Hefei, Anhui, People's

Republic of China

Correspondence

Cuixia Jiang, School of Management, Hefei

University of Technology, Hefei 230009,

Anhui, People's Republic of China.

Email: jiangcuixia@hfut.edu.cn

Funding information

Key Research and Development Plan Projects

in Anhui Province, Grant/Award Number:

202004a05020020; National Natural Science

Foundation of China, Grant/Award Numbers:

71671056, 72171070, 91846201; National

Social Science Foundation of China, Grant/

Award Number: 21BJY255

Abstract

To improve the performance of a large portfolio selection,

we consider the effect of tail network and propose a novel

tail network-augmented parametric mean-conditional value-

at-risk (CVaR) portfolio selection model labeled as TNA-

PMC. First, we adopt the least absolute shrinkage and selec-

tion operator-quantile vector autoregression (LASSO-QVAR)

approach to construct a tail network. Second, we parameter-

ize the weights of the mean-CVaR model as a function of

asset characteristics. Third, we incorporate the effect of the

tail network topological characteristic, namely eigenvector

centrality (EC), on the weights to construct the TNA-PMC

model. After that, we apply the model to the empiricalanaly-

sis on the Shanghai Stock Exchange 50 (SSE50) Index of

China from January 2010 to September 2020. Our empirical

results illustrate the effectiveness of the TNA-PMC model in

two aspects. First, the TNA-PMC model clarifies the eco-

nomic interpretationof the characteristics, such as the nega-

tive effective of EC on the portfolio weights. Second, the

TNA-PMC model performswell in terms of achieving diversi-

fication and attractiverisk-adjusted return.

KEYWORDS

mean-CVaR, network topology, parametric strategy, portfolio

selection, tail network

JEL CLASSIFICATION

G11

Received: 27 July 2021 Revised: 3 November 2021 Accepted: 18 February 2022

DOI: 10.1111/irfi.12379

© 2022 International Review of Finance Ltd.

International Review of Finance. 2023;23:37–57. wileyonlinelibrary.com/journal/irfi 37

1|INTRODUCTION

Portfolio refers to any combination of financial assets, such as stocks, bonds and cash, held by individual investors or

managed by financial professionals, hedge funds, banks and other financial institutions. Undoubtedly, portfolio perfor-

mance depends heavily on risk measurement. Since the pioneer work of Markowitz (1952)onmean–variance analysis,

various risk measurement methods have continuously emerged. Among them, quantile-based tail risk measurement has

gradually become a popular tool in financial risk management. The commonly used ones are value-at-risk (VaR) and condi-

tional value-at-risk (CVaR). Compared with VaR, CVaR is subadditive and convex. Therefore, CVaR is a coherent risk mea-

sure. In addition, the solution of CVaR-based portfolio selection models can be transformed into a quantile regression.

With the above advantages, CVaR becomes popular andhasbeenappliedinportfolioselectionmodels.

In a portfolio selection, the connections between financial assets have received widespread attention from both

academics and practitioners. First of all, the connections are common in reality, because financial assets are often

affected simultaneously by the same factors such as policy and economic environment. Moreover, with the continu-

ous advancement of economic globalization, their connections are getting closer and more complicated. The connec-

tions reflect the relationship between assets and will play a vital role in portfolio selection, see Yang, Yu, and

Ma (2019), Arreola Hernandez et al. (2020), and Wang et al. (2021). Therefore, the nature of the connection struc-

ture is of vital importance for portfolios. Compared to the one-to-one connection between financial assets, we turn

to use complex network for measuring multiple-to-multiple connections among all financial assets. From the point

view of complex network, financial assets are regarded as network nodes, and the connections between assets rep-

resent the edges. Generally, financial networks can comprehensively and systematically capture the interrelation-

ships among assets, characterize and visualize the interconnectedness among assets, see for instance Billio

et al. (2012), Diebold and Yilmaz (2014), Hautsch et al. (2015), Augustiani et al. (2015), Härdle et al. (2016), Demirer

et al. (2018), Barigozzi and Brownlees (2019), and Borges et al. (2020), among many others.

The existing literature has already introduced network effects into portfolio selection, see for instance Chuluun (2017),

Heipertz et al. (2019), and Nguyen et al. (2020). Based on the financial market network, Peralta and Zareei (2016)theoret-

ically investigate the negative relationship between the centrality of assets and their optimal weights. Following Peralta

and Zareei (2016), Výrost et al. (2019) propose an additional constraint in portfolio optimization, which requires weights

to satisfy the ordering imposed by centrality measures. Eom and Park (2017) empirically investigate the effects of com-

mon factors on the connectivity of the stock network and the distribution of portfolio weights. Zareei (2019)usesthe

cross-dependency in returns to construct financial assets network, and analyze the portfolio risk of various stylized net-

work structures. Arreola Hernandez et al. (2020) point out that the risks arising from interdependence of banks threaten

the portfolio return. Lee and Woo (2019), Yang, Zhao, andWang (2019 )andLietal.(2019) document that the topological

characteristics of the stock network are related to stock returns and their dynamics can be used for identifyingthe optimal

assets and constructing a stock portfolio. The existing complex network in portfolio selection models is considered under

the mean framework. However, compared to mean dependence, tail dependence among multiple financial assets is more

important for risk management. Furthermore, previous literature primarily focuses on selecting assets to invest with the

help of the network topological characteristic or theoretically showing the relationship between the assets' network topo-

logical characteristic and their optimal portfolio weights. Unfortunately, the intuitiveeffect of network topological charac-

teristics on portfolio weights may be overlooked. Thus, how to measure this complex tail dependence among financial

assets and take it into account in portfolio selection models is still a challenging work.

Moreover, the above mentioned network effect in portfolio selection is a static one. In reality, a static model is

difficult to capture time-varyingmarket information. It has been well documented that the time-varyingmarket infor-

mation is important for asset allocation, see Gârleanu and Pedersen (2013), Zhou et al. (2019), Ma et al. (2019), and

Antonakakis et al. (2019). In Fama and French (1996) and Lynch (2001), the asset characteristics like price/earnings

(PE) and book-to-market (BTM), are related to financial asset returns. On this basis, Brandt et al. (2009) propose a

new approach to optimize the portfolio selection model, which parameterizes directly the weight of each asset as a

function of its time-varying characteristics including market value (ME). The parametric portfolio policy can well

38 LI ET AL.