An ELECTRE III Based CBR Approach to Combinatorial Portfolio Selection

• Author: Veera Boonjing,Laor Boongasame,Pisit Chanvarasuth
• Date: 01 June 2019
• DOI: http://doi.org/10.1111/ajfs.12260
• Published date: 01 June 2019

An ELECTRE III Based CBR Approach to

Combinatorial Portfolio Selection*

Pisit Chanvarasuth

School of Management Technology (MT), Sirindhorn International Institute of Technology, Thammasat

University, Thailand

Laor Boongasame**

Department of Mathematics, Faculty of Science, King Mongkut’s Institute of Technology Ladkrabang,

Thailand

Veera Boonjing

International College, King Mongkut’s Institute of Technology Ladkrabang, Thailand

Received 1 July 2018; Accepted 5 November 2018

Abstract

Investors generally learn from historical data and use it to improve future investment deci-

sions. However, existing portfolio selection research rarely considers such a concept. This

paper proposes a novel approach to the combinatorial portfolio selection problem. Our solu-

tion employs the Case-Based Reasoning (CBR) approach with ELECTRE III, based on the

Industry Classification Benchmark standards. The target stocks are ranked based on their

similarities to a reference case, allowing specified selection criteria. Similarities are calculated

based on selected ELECTRE III metrics. Experimental results with the Stock Exchange of

Thailand data show that our CBR approach with ELECTRE III outperforms the traditional

method.

Keywords Case-Based Reasoning; ELECTRE III; Multiple criteria decision making; Stocks;

Financial analysis

JEL Classification: G12, G18, G21, G28, G34, G38

1. Introduction

Portfolio selection, as articulated by Markowitz in 1952, focuses on constructing

portfolios to optimize or maximize the expected return based on a given level of

market risk. According to the existing literature, various machine learning

*This research is supported by a research grant provided by Thammasat University.

**Corresponding author: Department of Mathematics, Faculty of Science, King Mongkut’s

Institute of Technology Ladkrabang, Bangkok, Thailand. Tel: +662-329-8400-11 ext 283, Fax:

+662-329-8412, email: laor.bo@kmitl.ac.th.

Asia-Pacific Journal of Financial Studies (2019) 48, 386–409 doi:10.1111/ajfs.12260

386 ©2019 Korean Securities Association

techniques for solving the portfolio selection problem with a single security have

been developed. Recent studies have presented encouraging results on portfo lio

selection using reinforcement learning (Moody et al., 1998; Moody and Saffell,

2001; Almahdi and Yang, 2017), neural networks (Kimoto et al., 1993), fuzzy algo-

rithms (Yunusoglu and Selim, 2013; Chourmouziadis and Chatzoglou, 2016; Fer-

reira et al., 2018; Liagkouras and Metaxiotis, 2018), genetic algorithms (Mahfoud

and Mani, 1996; Allen and Karjalainen, 1999; Dempster et al., 2001; Mandziuk and

Jaruszewicz, 2011), decision trees (Tsang et al., 2004), support vector machines

(Tay and Cao, 2002; Cao and Tay, 2003; Lu et al., 2009), boosting and expert

weighting (Creamer, 2007, 2012; Creamer and Freund, 2007), Dempster-Shafer evi-

dence theory (Thakur et al., 2018), and artificial bee colony algorithm (Kumar and

Mishra, 2017). In practice, investors apply different techniques using historical data

to realize insights and use them to guide their future investment decisions. A new

problem can be solved by recalling and reusing specific knowledge from past experi-

ence. However, using this problem-solving method with case-based reasoning

(CBR) for a portfolio has not been done. Due to its strengths, researchers have suc-

cessfully applied CBR to many areas: supply chain management and scheduling (Li

et al., 2012), bond rating (Shin and Han, 2001) business failure prediction (Li and

Sun, 2009), and stock market prediction (Chun and Park, 2005, 2006). This paper

presents a method to measure forward-looking covariance risk for any two assets

even when an explicit market for barter trading does not exist. We apply CBR with

ELECTRE III to solve the portfolio selection problem using historical stock market

data.

As mentioned above, case-based reasoning is the process of solving new

problems based on the solutions of similar past problems, while ELECTRE III is a

multi-criteria decision analysis method to help investors compare uncertain multi-

criteria businesses. In addition, we also highlight the use of the Industry

Classification Benchmark (ICB) standards (www.icbenchmark.com) instead of

utilizing typical financial ratios such as P/DPS, Price/Earnings ratio (PE), P/B, and

Enterprise value (BV/MV), which are used in existing financial research (Basu, 1977;

Chan and Lakonishok, 2004; Greenblatt, 2006). The work of (Basu, 1977) reveals that

the returns of stocks with a low Price/Earnings ratio (PE) are higher than those with

a higher PE. The studies conducted by Chan and Lakonishok (2004) identified Book

to Market, Earnings to Price, and Cash Flow to Price ratios as potential measures to

yield above-average returns in Japan. Greenblatt (Greenblatt, 2006) shows that simple

stock selection rules based on Return on Capital and the EBIT to Enterprise Value

(BV/MV) can also give above-average returns. To the best of our knowledge, there is

no research that emphasizes using the ICB to assist in investing.

The ICB is the pattern adopted to show the definition of industry classes. Classi-

fication of industries enables practitioners to systematically arrange firms in terms

of their similarity. Firms which belong to each class can be evaluated through the

same set of industry/commerce criteria (Courtis, 1978; Greig, 1992; Holthausen and

Larcker, 1992; Ou and Penman, 1992; Penman, 1992; Bernstein and Wild, 1999;

Combinatorial Portfolio Selection

©2019 Korean Securities Association 387