Expert Systems with Applications 38 (2011) 83758383

Contents lists available at ScienceDirect

Expert Systems with Applications

journal homepage: www.elsevier.com/locate/eswa

Analysis of decision making factors for equity investment by DEMATEL and Analytic Network Process
Wen-Shiung Lee a, Alex YiHou Huang b,, Yong-Yang Chang c, Chiao-Ming Cheng b
a

Department of Business Administration, Tamkang University, 151 Ying-chuan Road, Tamsui, Taipei 25137, Taiwan College of Management, Yuan Ze University, 135 Yuan-Tung Road, Chung-Li, Taoyuan 32003, Taiwan c Department of Business and Entrepreneurial Management, Kainan University, 1 Kainan Road, Luzhu Shiang, Taoyuan 33857, Taiwan
b

a r t i c l e
Keywords: DEMATEL ANP Investment strategy

i n f o

a b s t r a c t
Existing methodologies of equity investment, such as fundamental analysis, technical analysis, and institutional investor analysis, explore important factors of stock price behaviors. However, the interdependent relationships of the key factors have not yet been fully studied. This paper provides the rst analysis on the interactive relationships among the factors in incorporating the methods of Decision Making Trial and Evaluation Laboratory (DEMATEL) and Analytic Network Process (ANP). The empirical results show that factors from the existing analytical methodologies have signicant interactive and self-feedback dynamics. Among the key factors, protability is the most important one affecting investment decision, followed by growth and trading volume. In addition, due to the complexity of the ANP, this study proposes a new methodology to simplify the process, and empirical evidences indicate that the approach is effective and efcient. 2011 Elsevier Ltd. All rights reserved.

1. Introduction Equity stock is the most popular nancial asset in capital markets. Academic researchers, industrial practitioners, and investors have employed various techniques to analyze the dynamics of stock prices. In particular, different methods have been developed to set certain selection criteria in choosing individual stocks with promising returns, and three methodologies have been broadly applied: the fundamental analysis, technical analysis, and institutional investor analysis. These methods are used to explore key factors that yield signicant impacts on stock prices and the effectiveness of the factors are often individually and independently analyzed. This study, on the other hand, employs the Decision Making Trial and Evaluation Laboratory (DEMATEL) and Analytic Network Process (ANP) to focus on the analyzes of interdependent relationships among the factors. Through the understanding of the interactions between key factors, this study seeks to provide more efcient and effective investment decision making process. Fundamental analysis studies key elements of nancial conditions such as earnings and dividend prospects along with overall market conditions such as interest rates to predict the movements of stock price. The method is based on the key assumption where stock performance of a rm could fully reect its operating condition. Recent researches have adopted the method include Tang and Shum (2003), Fama and French (2006), Wu and Xu (2006), Edirisin Corresponding author. Tel.: +886 3 4638800x2668; fax: +886 3 4354624.
E-mail address: huang@saturn.yzu.edu.tw (A.YiHou Huang). 0957-4174/$ - see front matter 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2011.01.027

ghe and Zhang (2007), Girard and Omran (2007), Chen and Zhang (2007), Cole, Moshirian, and Wu (2008), Samaras, Matsatsinis, and Zopounidis (2008), and Benjamas, Riza, and Ramesh (2008) and others. Technical analysis studies for recurrent and predictable patterns in stock prices such as return momentum and herding behaviors, and consequently, seeks to forecast future trends of stock movements. Recent examples with application of the approach are Lee and Rui (2002), Darrat, Rahman, and Zhong (2003), Wang and Chan (2006), Avramov, Chordia, Jostova, and Philipov (2007), Yamawaki and Tokuoka (2007), Zarandi, Rezaee, Turksen, and Neshat (2009), and Li and Kuo (2008). Institutional investor analysis investigates the relationships between stock price and trading activity of institutional investors such as the institutional buy/sell signals. Recent studies applying this methodology include Grifn, Harris, and Topaloglu (2003), Darrat et al. (2003), Wang and Cheng (2004), Cai and Zheng s(2004), Avramov, Chordia, and Goyal (2006), Bohl and Brzeszczyn ki, (2006), Chiyachantana, Jain, Christine, and Robert (2006, and Hirose, Kato, and Bremer (2009) and among others. These prior researches mainly investigate the unidirectional impacts of each individual factor on the dynamics of stock price rather than explore the interdependences and feedback effects among the factors and their relative marginal strengths. Consequently, due to lack of analyzes on these quantitative interdependences between factors, above approaches do not provide direct assessments and proper comparisons on the economic signicances of these factors.

8376

W.-S. Lee et al. / Expert Systems with Applications 38 (2011) 83758383

This research accordingly applies two quantitative methods, the DEMATEL and ANP, to evaluate the level of interdependences among those factors and to add on the understandings of dynamics of stock return. DEMATEL is one of tools on decision making of multiple criteria and is able to transform qualitative issues into quantitative tasks for analysis. ANP can be applied to analyze the interactive relations between factors with proper measurements of the weight for each factorial criterion. Consequently, each alternative in the decision making would be evaluated and prioritized to determine the length and area of improvements. By selecting the proper critical value in the process, the optimal combination of decision making can then be determined. Tzeng, Chiang, and Li (2007) applied the method of DEMATEL into multiple criteria decision making (MCDM) model, and Ou Yang, Shieh, Jun-Der Leu, and Tzeng (2008) proposed the application of both DEMATEL and ANP methods as the hybrid MCDM model. Cheng and Li (2007) used ANP to nd out key selecting criteria for a strategic partner and further to identify the relative importance among the criteria. Lee, Kim, Cho, and Park (2009a) uses ANP to analyze the problem of network technology, to consider the related technology network required, and to seek out the core technology and the characteristics of network technology. Additional recent researches adopted DEMATEL or ANP are Lin, Lee, and Chen (2009a), Hsieh, Lin, and Lin (2008), Lin, Chiu, and Tsai (2008), Lin, Tsai, Shiang, Kuo, and Tsai (2009b), Lee, Tzeng, Hsu, and Huang (2009b), Tzeng, Chen, Yu, and Shih (2009), and Li and Tzeng (2009) and others. This study contributes to literature in several ways. First, this paper provides the evidence in applications of ANP and DEMATEL on stock market dynamics, particularly the process of investment decision making. Second, the relations of interdependences among key factors in stock investment are quantitatively analyzed through investigations of experts perceptions. Third, theoretical analyzes are conrmed with empirical applications of recent data. Based on the empirical outcomes, the paper nds the methodologies are able to produce robust analyzes in generating valuable and practical investment strategy. The rest of this paper is organized as follows. In Section 2, key existing models of stock investment decision making are reviewed, and in Section 3, the application of ANP and DEMATEL is introduced. In Section 4, empirical outcomes are presented, and Section 5 concludes the research. 2. Factors in stock investment strategy Through literature review, this section discusses ndings of previous literatures regarding factors that affect investment strategy of equity stock. The factors were traditionally studied through three techniques: the fundamental, technical and institutional investor analyzes. The identication of the key factors provides theoretical supports to the empirical applications of this study. 2.1. Key factors under fundamental analysis Factors with impacts on stock return under fundamental analysis are basically the key variables of nancial statement of the underlying rm such as prot, operational management, leverage, account balances and so on. Macroeconomic variables such as GDP growth, unemployment rate, and interest rate are often controlled under the analysis. Recently, Fama and French (2006) examined the relationships between stock returns and equity ratio, expected protability, and expected investment and concluded that expected protability is most signicantly correlated with expected and average future stock returns. Wu and Xu (2006) studied stock markets of China with empirical data between 2000 and 2006 and constructed a stock price prediction model that consisted of protability, risk, growth, liquidity, and asset management. Their ndings showed

that these fundamental analysis factors can be applied successfully in prediction of stock price uctuations. Edirisinghe and Zhang (2007) studied the quarterly nancial statements of 230 US hightech companies and provided 18 analytical categorizations including protability, asset utilization, liquidity, leverage and growth, to construct representative indicators of nancial performance. Girard and Omran (2007) employed samples of ve emerging markets in Middle East to show that the political risk is a major factor inuencing movements of stock market. This result was similar to that of Tang and Shum (2003) which documented signicant relationship between political shocks and stock returns. Chen and Zhang (2007) concluded that factors related to protability have the most signicant impact on individual stock, and Cole et al. (2008) documented that stock market return leads economic growth where the two dimensions are also interactive to each other. In a study of Sweden stock market, Samaras et al. (2008) adopted the multicriteria decision support system and portfolio management to evaluate the strength and weakness of individual stocks based on method of fundamental analysis. Benjamas et al. (2008) surveyed Thailand stock market between 1975 and 2006 and concluded that key factors of fundamental analysis are level of dividend payout and earning changes. 2.2. Key factors under technical analysis Technical analysis studies the historical data surrounding price and volume movements of the stock by using charts as a primary tool to forecast future price movements. Brock, Lakonishok, and Lebaron (1992) researched Dow Jones index stocks and used variable-length moving average (VMA), xed-length moving average (FMA), and trading rage break (TRB) as indicators to evaluate the effectiveness of technical analysis. The study showed that investment returns under VMA and TRB are signicantly better than a strategy of buy-and-hold. Mizuno, Kosaka, and Yajima (1998) applied data of Japanese stocks and utilized VMA and relative strength indicators (RSI) of technical analysis to construct a stock prediction model with empirical ndings robust to the alternative methods. Wong, Meher, and Chew (2003) studied stock price changes in Singapore Stock Exchange with technique of technical analysis to distinguish the events of market entry and exit. The authors discovered that the performance of technical indicators positively related to investment returns. Wang and Chan (2006) surveyed Microsoft, Intel, and IBM and employed decision trees integrated with RSI to study timing points of stock purchase. The research showed this method can accurately predict optimal buying timing and enhance returns of investors. Yamawaki and Tokuoka (2007) used key stocks of companies in retail, PC, automobile, and petroleum industries from New York Stock Exchange and classied technical indicators into trend-type moving averages, volatile-type RSI, MACD, and momentum-type indicators, and showed that MA is the most effective performance indicator. Liu and Kwong (2007) studied factors that inuence stock market of Hong Kong. Using PXtract algorithm and technical analysis information such as daily highs and lows and daily opening and closing prices, the study found that a combination of PXtract and technical indicators can successfully predict price changes of stock. Li and Kuo (2008) used knowledge exploration as the basis and incorporated the K charts of technical analysis to successfully predict short-term trends of stock price changes. With data on stocks of Asian automobile companies, Zarandi et al. (2009) incorporated a type-2 fuzzy rule based-expert system with technical indicators of RSI, moving average convergencedivergence (MACD), and other moving averages. The authors developed the fuzzy multi-objective model for stock prediction/analysis and found that the model works successfully in anticipating stock price movements.

W.-S. Lee et al. / Expert Systems with Applications 38 (2011) 83758383

8377

2.3. Key factors under institutional investor analysis Lo and Mackinlay (1988) documented that equity markets cannot be perfectly efcient due to information asymmetry among investors. Grossman and Stiglitz (1980) classied investors as uninformed traders or informed traders depending on how comprehensively and quickly they react to market sensitive information. Easley and Hara (1987) showed that foreign institutional investors are usually informed traders and have superior abilities in stock market analyzes over domestic traders. Lee and Rui (2002) found a positive feedback relationship between stock market and institutional trading volume, and Darrat et al. (2003) discovered that trading volume would lead to stock price changes. Wang and Cheng (2004) concluded that higher trading volumes would lead to lower market return and vice versa for Shanghai index. Cai and Zheng (2004) found that stock market volatility is generally affected by institutional investment. In addition, the effect is more signicant when institutional investors increase their stock holdings. Chiyachantana et al. (2006) examined the impact of institutional investor trading on stock market in 43 countries and showed that stock markets are signicantly affected by institutional investor trading only during the period of transaction. ski (2006) employed Markov switching Bohl and Brzeszczyn GARCH to provide evidences that that institutional investor analysis can effectively predict short-term stock price movements. By sample of stocks in Japan and Korea, Ko, Kim, and Cho (2007) concluded that trading patterns of both domestic and foreign institutional investors would signicantly affect stock price movements. Zhu, Wang, Xu, and Li (2008) tested NASDAQ, NYSE and Singapore Strait Times Index with method of neural network system and showed that trading volume adequately affects mid- and longterm stock price movements. Hirose et al. (2009) provided evidences to support the signicant relationship between institutional trading and stock market return with data of Japanese stock market. In sum, prior studies show that all three methodologies of fundamental, technical and institutional investor analyzes may effectively predict stock price changes; however, no existing researches have provided evidences on the interdependent relationship between the factors across the three methods. Fundamental analysis contains key factors of risk, protability, growth, and asset management; technical analysis utilizes RSI, MA, and MACD factors; and factors of institutional investor analysis are trading volume and institutional investment. A summary of the major factors from the three analytical techniques is presented in Table 1. 3. Methodology This study employs DEMATEL to investigate interdependences between stock investment factors and then incorporates ANP to evaluate the magnitudes among factors and their dependences. This section, at rst, establishes network relationships through DEMATEL, and second, applies ANP to obtain the relative weights respective to each factor. Third, process of data collection according to the methodology is presented. 3.1. Application of DEMATEL for network relationship In order to construct an efcient model of investment strategy, levels of interaction and self-feedback relationships among key factors of stock return dynamics need to be analyzed. In multiple criteria decision making (MCDM), similar issues are commonly addressed by method of Decision Making Trial and Evaluation Laboratory (DEMATEL), which is mainly applied to resolve multidimensional issues by clarifying the relationships among essential elements. In recent years, DEMATEL has been popularly applied to

Table 1 Key factors in stock investment decision making. Dimensions Fundamental analysis Factors Risk Growth Protability Related studies Tang and Shum (2003), Wu and Xu (2006), and Girard and Omran (2007) Wu and Xu (2006), Edirisinghe and Zhang (2007), and Cole et al. (2008) Wu and Xu (2006), Fama and French (2006), Edirisinghe and Zhang (2007), and Chen and Zhang (2007) Wu and Xu (2006), and Edirisinghe and Zhang (2007) Mizuno et al. (1998), Wang and Chan (2006), Zarandi et al. (2009), and Yamawaki and Tokuoka (2007) Brock et al. (1992), Mizuno et al. (1998), Zarandi et al. (2009), and Yamawaki and Tokuoka (2007) Lee and Rui (2002), Darrat et al. (2003), Wang and Cheng (2004), and Zhu et al. (2008) Cai and Zheng (2004), Chiyachantana ski et al. (2006), Bohl and Brzeszczyn (2006), and Hirose et al. (2009) Easley and OHara (1987), Cai and Zheng ski (2006), (2004), Bohl and Brzeszczyn Chiyachantana et al. (2006), Ko et al. (2007), and Hirose et al. (2009)

Operation and asset management Technical analysis RSI (relative strength index) MA (moving average) Trading volume

Institutional investor analysis

Local institutional investors Foreign institutional investors

effectively pinpoint complicated structure of causal relationships. Through the observation of paired interaction within elements, the method utilizes matrix operations and mathematical techniques to quantify the causality and strength of dependences among the elements. Tzeng et al. (2007) showed that DEMATEL can enhance understandings of complex problems to identify feasible solutions within hierarchical structures. DEMATEL involves six steps, and in step 1, scales of evaluations need to be constructed. Using pair-wise comparisons of dimensions, the degrees of perceptions from the interviewees regarding the level of impacts of particular dimensions are examined. The measurement criteria of 0, 1, 2, 3, and 4 are used to illustrate no inuence, low inuence, medium inuence, high inuence, and extremely high inuence, respectively. In step 2, the direct-inuence matrix is constructed based on the degrees of relative impacts derived from the pair comparisons. An n n direct-inuence matrix Z with the directly observed relations is obtained where zij denotes the degree of impacts of the i factor on the j factor.

z11 6z 6 21 Z6 . 6 . 4 . zn1

z12 z22 . . . zn 2

3 z1n z2n 7 7 7 .. . 7 . . . 5 znn

Step 3 composes the normalized direct-relation matrix, which can be derived from Eqs. (2) and (3), where the matrix diagonal is coded 0 and the sum of each row and column does not exceed 1.

X sZ 2

1 1 5; P Pn ; s min4 max n j1 jzij j max i1 jzij j

i j

2 i; j 1; 2; . . . ; n 3

Step 4 derives the total-inuence matrix T from the equation, T = X + X2 + X3 + + Xm = X(I X)1, in which I represents the identity matrix, and step 5 obtains the prominence and relation by summing each row and column in T to yield D and R. Here, di is the sum of each row in T and the rows show the degrees of direct and indirect impacts over the other criteria, and rj is the sum of

8378

W.-S. Lee et al. / Expert Systems with Applications 38 (2011) 83758383

each column in T where columns indicates the degrees of inuences from other criteria. Numeric algorithm variable di, therefore, represents the factors that inuence others, rj represents factors that are inuenced by others, di + rj represents the strength of relationships between factors, di rj represents the strength of inuences among factors. In other words, di + rj and di rj represent the so called prominences and relations, respectively. Step 6 draws the network relation map (NRM) to simplify the interdependences in an easy-to-understand structure and to clearly express relationship between factors, levels of inuences, and the degree of impacts (please see Tamura, Nagata, & Akazawa, 2002 and Tzeng et al., 2007 for detail discussions). 3.2. Weight measurements by integrating DEMATEL and ANP In addition to DEMATEL in constructing the interdependences of factors, this study obtain accurate numerically weights of interactions by the method of Analytic Network Process (ANP), proposed by Saaty (1996). The ANP is designed to resolve the interdependence and feedback relations of criteria and liberalizes the limitations of Analytic Hierarchy Process (AHP). ANP can be seen as a generalized AHP, and the key difference between the two methods is that ANP is applied to decision-making criteria concerning their interdependences while AHP is applied with the assumption of independences among the criteria given only hierarchical relations are explored. As discussed in previous section, the factors affecting stock investment decision are certainly correlated with each other, and the analyzes of these factors by ANP should be more consistent with practical application. The new hybrid method adopted by this study is an extension of Ou Yang et al. (2008). The approach not only applies DEMATEL to conrm different degrees of impacts of the clusters but also normalizes DEMATELs total-inuence matrix T and incorporates it into unweighted supermatrix of ANP. Although key interdependences of clusters can be obtained via DEMATEL, those between clusters have to be determined by ANP algorithm. In addition, the dynamic inuence relationship implied in the total-inuence matrix acquired by DEMATEL is similar to the concept of ANP, which conrms the importance and inuence of criteria through questionnaires. The method of ANP is conducted by four steps. Step 1 constructs the structure of issues in conrm the decision process and establish the subject in interest. After clearly describing the objective, it is decomposed into network hierarchical structure. Step 2 involves building the unweighted supermatrix by normalizing the sum of inuence for each criterion in each hierarchy under the criteria total-inuence matrix of DEMATEL.

During the process of normalizing, the criteria total-inuence matrix TC yields Ta C shown as following equation.

11 where Ta is obtained as follows. C

di

n X j 1

t ij
2 a11 t C 11 7 6 7 6 . . 7 6 . 7 6 . . . 7 6 7 6 a11 11 7 11 t tC in =d2 7 6 6 C i1 7 6 7 6 . 7 6 . . 7 6 . . . 5 4 11 a11 t t11 = d 3 C nn C n1 t 11 =d1 C 1n
11

6
11 11

a11 TC

t11 11 =d1 6 C 6 . 6 . 6 . 6 6 11 11 6 = t i 1 6 C d2 6 6 . 6 . 6 . 4 11 = d3 t 11 C n1

t11 =d1 C 1j . . . t 11 =d2 C ij . . . t 11 =d3 C nj

a11 tC 1j

a11 tC 1n

. . .
a11 tC ij

11

. . .
a11 tC nj

11

7 7 . 7 . . 7 7 7 a11 7 tC in 7 7 7 . 7 . . 7 5 a11 tC nn 7

ann and TC is obtained same process. In the normalized criteria totalinuence matrix, the interdependence relationship among clustering is incorporated into the un-weighted supermatrix:

The derivation of W 11 and W 12 is based on Eq. (9) where vectors with blank or 0 show independent relationship between clustering and criteria, and W nn is obtained by the same process.

W.-S. Lee et al. / Expert Systems with Applications 38 (2011) 83758383

8379

9
Step 3 obtains the weighted supermatrix by normalizing the sum of impact for each hierarchy and each dimension in the dimensions total-inuence matrix as illustrated in (10).

weight of the factors, and identications of priorities among the factors in achieving the optimal investment strategy. Robust test is also provided with empirical results of leading stocks of seven major industrial sectors in Taiwan: MediaTek (semiconductor), Motech (optical electronics), HTC (communications), Cathay FHC (nance), Huaku (construction), China Steel (steel), and Taiwan Cement (cement). 4.1. Analytical technique relationship by DEMATEL This study applies DEMATEL to construct the structure in investment decision making and analyze the interdependent relationships of three analysis techniques and nine factors. The dimensions total-inuence matrix T derived is provided in Table 2 to illustrate the degree of interdependent relationships. The table shows signicant interactive relationships between the three techniques where all measures are non-zero. In addition, the prominences and relations between techniques are reected by the sums of inuences and provided in Table 3. The table shows greater d coefcient of fundamental analysis as 15.17 than 12.44 of technical analysis and 10.46 of institutional investor analysis, and the outcome indicates that comparing to other techniques, fundamental analysis is the most important stock evaluation technique. However, fundamental analysis also has the greatest r coefcient and d + r, and the results show the technique is easier to be affected by other techniques but at the same time, generates most total impacts to others in average. Institutional investor analysis has highest d r of 1.92 and is the technique with most direct effeteness to others. Table 4 shows that all factors possess interactive relationship with others and from the total-inuence results of Table 5, relatively to other factors, institutional investment (I2) is considered the most important one in terms of both direct and indirect relations. In contrast, factor of operation and asset management (F4) is considered the least important. Growth (F2) is rated the highest

6 . 6 . 6 . 6 i1 TD 6 6 tD 6 6 . 4 . .

t 11 D

j t1 D

n t1 D

. . . t ij D . . .

7 . 7 . . 7 7 7 t in D 7 7 . 7 . . 5

10

1 tn D

t nj D

t nn D

Normalizing the total inuence matrix TD yields T a D as following.

3 2 a11 3 1j 1j 1n 1n tD t 11 ta ta D =d1 t D =d1 t D =d1 D D 7 7 6 . 6 . . . . 7 6 . 7 6 . . . . . 6 . . . . . 7 6 . . 7 7 7 6 6 in 7 6 t ai1 t aij tain 7 6 i1 Ta t ij D 6 t D =d2 6 D D 7 D =d2 t D =d2 7 D 7 6 7 6 . . . . 7 6 . 7 6 . . . . . 4 . . . . . . 5 4 . . 5 2
nj 1 nn tn D =d3 t D =d3 t D =d3

anj an1 ann tD tD tD

11
The weighted supermatrix is obtained by incorporating the unweighted supermatrix into the normalized dimensions total-inuence matrix shown as below.

6 6 a12 21 6 tD W 6 6 . W6 . . 6 6 . 6 . 4 .
a1n W n1 tD

a11 tD W 11

12 21 ta D W 22 22 ta D W

. . .
aij W ij tD . . .

a2n tD W n2

3 an1 tD W 1n 7 . 7 . . 7 7 7 ni 7 ani tD W 7 7 . 7 . 5 . 12

Table 2 The outcomes of total-inuence matrix T. Analytical techniques Fundamental analysis Technical analysis Institutional investor analysis Fundamental analysis 7.04 5.66 5.01 Technical analysis 4.60 3.97 3.25 Institutional investor analysis 3.53 2.81 2.20

ann tD W nn

Step 4 obtains the limited supermatrix, or the weight of each criterion, by multiple productions of the weighted supermatrix as limz!1 W z , where W is the limited supermatrix and z is an arbitrary number. 3.3. Data collection This study surveys experts with extensive knowledge and experiences of nancial economics and investment strategy in capital markets. All experts are either industrial practitioners with over 3 years of administrative experiences in security houses or mutual fund managements, or academic scholars with research concentration in nancial investment and over 3 years of teaching experience in related subjects. Through in-depth interviews with the experts, this study obtains their perceptions of various investment criteria and the levels of portfolio performances evaluated from these criteria. Total of fteen questionnaires are conducted between October and December of year 2008. 4. Empirical outcomes This section presents outcomes on the stock investment factors according to the preferences of experts, the determinations of the

Table 3 The sum of inuences of analytical techniques. Analytical techniques Fundamental analysis Technical analysis Institutional investor analysis d 15.17 12.44 10.46 r 17.71 11.82 8.54 d+r 32.88 24.26 19.00 dr 2.54 0.62 1.92

Table 4 The factor total-inuence matrix T. Factors F1 F2 F3 F4 T1 T2 T3 I1 I2 F1 0.39 0.41 0.54 0.33 0.48 0.48 0.48 0.60 0.68 F2 0.57 0.42 0.64 0.45 0.50 0.53 0.53 0.67 0.72 F3 0.51 0.52 0.46 0.40 0.51 0.52 0.52 0.65 0.73 F4 0.39 0.37 0.41 0.23 0.39 0.36 0.36 0.46 0.50 T1 0.43 0.36 0.41 0.29 0.35 0.48 0.48 0.51 0.57 T2 0.47 0.43 0.49 0.32 0.51 0.39 0.52 0.59 0.59 T3 0.41 0.34 0.38 0.27 0.45 0.46 0.33 0.48 0.51 I1 0.49 0.50 0.56 0.38 0.49 0.50 0.50 0.63 0.55 I2 0.43 0.41 0.46 0.30 0.44 0.44 0.44 0.43 0.59

8380 Table 5 The sum of inuences of factors. Factors F1 F2 F3 F4 T1 T2 T3 I1 I2 d 4.09 3.76 4.35 2.97 4.12 4.16 4.16 5.02 5.44 r 4.39 5.03 4.82 3.47 3.88 4.31 3.63 4.60 3.94

W.-S. Lee et al. / Expert Systems with Applications 38 (2011) 83758383

4.2. Weight calculation by incorporating DEMATEL with ANP

d+r 8.48 8.79 9.17 6.44 8.00 8.47 7.79 9.62 9.38 dr 0.30 1.27 0.47 0.50 0.24 0.15 0.53 0.42 1.50

dr
I2

2
I1

I (19.00, 1.92), Institutional Investors Analysis (I1, I2)

T3 T1 T2 T (24.26, 0.62), Technical Analysis (T1, T2, T3)

1 15 0 20 25

30

d+r
F1 F4 F2 F3

-1

The inuence relationship map obtained through DEMATEL can provide us with an understanding of the entire structure. The ANP network relation map can thus be established, as shown in Fig. 2. The new analytical model proposed by this study is an integration of DEMATEL with ANP. According to the dynamic inuence relationship obtained through DEMATEL, an unweighted supermatrix is constructed. Then, a weighted supermatrix is established based on the degree of inuence of each dimension. Subsequently, a limited supermatrix is composed by taking into account the weight of each factor (overall weight), as illustrated in Table 6. Empirical results show that among the nine factors, growth, profitability and trading volume are the most important ones. Specically, growth is the most important factor with a weight of 0.128, followed by protability at 0.127 and local institutional investors at 0.122. Relative to other factors, experts perceive operation and asset management to be least important (weight = 0.090). With respect to each analysis technique, experts perceive protability and growth to be most important in the fundamental analysis; MA to be most important in the technical analysis; and local institutional investors to be most important under institutional investor analysis. In sum, there are two factors of fundamental analysis ranked in the top two among all, and this result shows that experts perceive fundamental analysis as the most important technique in the process of stock investment decision making. Moreover, although institutional investor analysis is perceived as less important, this dimension includes an important element, trading volume, which is ranked as the third important factor.
Table 6 Ranking and weight of factors. Analytical techniques/factors Fundamental analysis Risk Growth Protability Operation and asset management Technical analysis RSI MA Trading volume Institutional investor analysis Local institutional investors Foreign institutional investors Weight 0.464 0.119 0.127 0.128 0.090 0.310 0.101 0.115 0.094 0.226 0.122 0.104 Ranking 4 2 1 9 7 5 8 3 6

-2

F (32.88, -2.54),

Fundamental Analysis (F1,F2,F3,F4)

Fig. 1. Inuence relationship map.

in the inuence received category and local institutional investors (I1) is rated the highest in the prominence category (d + r) and perceived as the most important factor among all by experts. The highest score in the relation category (d r) is the foreign institutional investors (I2) and the result implies the factor has the greatest direct inuence on others. The factor with lowest d r is growth (F2), and the outcome indicates that the factor is most likely to be inuenced by others. Based on the results of Tables 4 and 5, an inuence relationship map is compiled as Fig. 1.

F1 Risk F2 Growth F3 Profitability F4 Operation and Asset Management

T1 RSI T2 MA T3 Trading Volume

(F) Fundamental Analysis I1 Local Institutional Investors I2 Foreign Institutional Investors

(T) Technical Analysis

(I) Institutional investor

Fig. 2. Network model of ANP.

W.-S. Lee et al. / Expert Systems with Applications 38 (2011) 83758383 Table 7 Outcomes of factors for sample stocks. Weight MediaTek Local 0.119 0.127 0.128 0.090 0.101 0.115 0.094 0.122 0.104 (F1) (F2) (F3) (F4) (T1) (T2) (T3) (I1) (I2) 2.800 4.333 4.333 3.600 3.600 3.533 3.400 3.600 4.000 ANP 0.333 0.550 0.555 0.324 0.364 0.406 0.320 0.439 0.416 3.707 1 Motech Local 3.133 3.600 3.467 3.000 3.467 3.067 3.200 3.667 3.467 ANP 0.373 0.457 0.444 0.270 0.350 0.353 0.301 0.447 0.361 3.355 4 HTC Local 3.333 4.200 4.133 3.267 3.533 3.400 3.200 3.667 3.733 ANP 0.397 0.533 0.529 0.294 0.357 0.391 0.301 0.447 0.388 3.637 2 Cathay Financial Holdings Local 2.267 3.133 3.133 3.867 3.133 3.133 3.067 3.600 3.600 ANP 0.270 0.398 0.401 0.348 0.316 0.360 0.288 0.439 0.374 3.195 5 Huaku Development Local 3.133 4.200 4.133 3.267 3.200 3.333 3.133 3.467 3.800 ANP 0.373 0.533 0.529 0.294 0.323 0.383 0.295 0.423 0.395 3.548 3 China Steel Local 2.467 3.333 3.267 3.467 2.933 2.867 2.933 3.400 3.467 ANP 0.294 0.423 0.418 0.312 0.296 0.330 0.276 0.415 0.361 3.124 6

8381

Taiwan Cement Local 2.000 3.000 3.133 3.133 3.067 2.667 2.933 3.467 3.667 ANP 0.238 0.381 0.401 0.282 0.310 0.307 0.276 0.423 0.381 2.999 7

Total Ranking

Table 8 Stock returns prior and after surveying of sample stocks. MediaTek (%) Panel A. Prior to the survey 2004 2005 2006 2007 2008 Panel B. After the survey 1 month (2009/01/052009/01/21) 2 month (2009/01/052009/02/27) 3 month (2009/01/052009/03/31) 4 month (2009/01/052009/04/30) 5 month (2009/01/052009/05/27) 6 month (2009/01/052009/06/30) 7 month (2009/01/052009/07/31) 8 month (2009/01/052009/08/31) 9 month (2009/01/052009/09/30) 10 month (2009/01/052009/10/30) ANP Ranking 1.90 25.53 30.35 38.04 52.98 50.96 72.80 74.17 85.62 71.41 1 (1) (1) (1) (2) (2) (2) (2) (1) (2) (2) 7.59 11.84 19.83 27.00 42.67 30.51 26.89 7.83 14.27 7.53 2 (2) (3) (3) (4) (4) (6) (7) (7) (7) 7.48 5.44 24.50 44.59 87.89 81.87 84.40 72.46 86.25 83.01 3 (3) (4) (2) (1) (1) (1) (1) (2) (1) (1) 16.09 5.96 12.68 11.62 44.34 43.95 40.39 20.03 26.50 24.55 4 (6) (5) (4) (5) (3) (3) (3) (6) (6) (5) 15.39 29.04 25.19 1.21 34.03 25.83 29.57 20.54 35.35 42.22 5 (5) (7) (7) (7) (5) (5) (5) (5) (3) (3) 6.80 7.91 8.60 5.65 13.54 15.33 35.58 29.73 29.73 27.70 6 (2) (6) (6) (6) (7) (6) (4) (3) (5) (4) 21.25 5.39 1.08 13.38 20.71 12.72 22.87 29.09 30.20 24.21 7 (7) (3) (5) (4) (6) (7) (7) (4) (4) (6) 16.63 72.60 1.29 31.53 54.87 HTC (%) 44.33 161.50 26.22 25.53 27.86 Huaku Development (%) 103.06 8.05 80.15 6.80 34.26 Motech (%) 111.51 134.82 39.66 7.80 112.74 Cathay Financial Holdings 27.61 4.80 28.79 7.13 52.76 China Steel 37.45 17.58 49.61 32.53 52.12 Taiwan Cement 28.18 20.10 30.66 47.95 45.42

Data source: Taiwan Economic Journal (TEJ) database.

4.3. Scores and evaluations of individual stock performances The data sample of this study contains the leading stocks of seven representative industrial sectors in Taiwanese stock market, including MediaTek, Motech, HTC, Cathay FHC, Huaku, China Steel, and Taiwan Cement. Fifteen expert scholars scored the performance of individual stocks in a range of 15, with 1 being the lowest score and 5 the highest. The averages are then calculated from the scores, and along with the relative weights acquired from ANP, results are shown in Table 7. Among these seven individual stocks, MediaTek received the highest overall performance score of 3.707, followed by HTC at 3.637, Huaku at 3.548, Motech at 3.355, Cathay FHC at 3.195, and China Steel at 3.124, and Taiwan Cement is the relatively poorest performer with a score of 2.999. Results of the ANP investment model show that the most important factor is protability (weight = 0.128). When a company possesses outstanding protability, it will attract more investments and increase company value, therefore, a companys protability naturally becomes a key factor in determining its future stock movement. Growth has a weight of 0.127 and is ranked second among all factors, and the result indicates that investors are willing to invest funds into a company with more stable growth outlook. Empirically, the results of overall performance show that MediaTek is the best choice among the seven stocks, followed by HTC, Huaku, Motech, Cathay FHC, and China Steel where the worst choice would be Taiwan Cement. In order to conrm the outcomes

Table 9 Outcomes of ordinary hybrid method and proposed method. Criteria F1 F2 F3 F4 Risk Growth Protability Operation and asset management T1 RSI T2 MA T3 Trading Volume I1 Local institutional investors I2 Foreign institutional investors Ordinary hybrid 0.112 0.138 0.146 0.069 0.116 0.112 0.082 0.103 0.122 The proposed method 0.119 0.127 0.128 0.090 0.101 0.115 0.094 0.122 0.104 Difference (0.007) 0.009 0.018 (0.011) 0.015 (0.003) (0.008) (0.019) 0.018

Fig. 3. Comparisons of weights of each criteria between the methods.

8382

W.-S. Lee et al. / Expert Systems with Applications 38 (2011) 83758383

Table 10 Outcomes of factors for sample stocks by ordinary hybrid method. Weight MediaTek Local 0.112 0.138 0.146 0.069 0.116 0.112 0.082 0.103 0.122 (F1) (F2) (F3) (F4) (T1) (T2) (T3) (I1) (I2) 2.800 4.333 4.333 3.600 3.600 3.533 3.400 3.600 4.000 ANP 0.314 0.598 0.633 0.248 0.418 0.396 0.279 0.371 0.488 3.743 1 Motech Local 3.133 3.600 3.467 3.000 3.467 3.067 3.200 3.667 3.467 ANP 0.351 0.497 0.506 0.207 0.402 0.344 0.262 0.378 0.423 3.370 4 HTC Local 3.333 4.200 4.133 3.267 3.533 3.400 3.200 3.667 3.733 ANP 0.373 0.580 0.603 0.225 0.410 0.381 0.262 0.378 0.455 3.668 2 Cathay Financial Holdings Local 2.267 3.133 3.133 3.867 3.133 3.133 3.067 3.600 3.600 ANP 0.254 0.432 0.457 0.267 0.363 0.351 0.251 0.371 0.439 3.186 5 Huaku Development Local 3.133 4.200 4.133 3.267 3.200 3.333 3.133 3.467 3.800 ANP 0.351 0.580 0.603 0.225 0.371 0.373 0.257 0.357 0.464 3.581 3 China Steel Local 2.467 3.333 3.267 3.467 2.933 2.867 2.933 3.400 3.467 ANP 0.276 0.460 0.477 0.239 0.340 0.321 0.241 0.350 0.423 3.127 6 Taiwan Cement Local 2.000 3.000 3.133 3.133 3.067 2.667 2.933 3.467 3.667 ANP 0.224 0.414 0.457 0.216 0.356 0.299 0.241 0.357 0.447 3.011 7

Total Ranking

of the model, Table 8 presents the actual short-term returns of the stocks after the conduction of surveys for comparisons. According to Table 8, stock returns prior to 2008 were generally positive for all seven stocks, but due to the recent nancial crisis, returns of 2008 are mostly negative across industries. The returns of 2009 are composed by different lengths of time periods from 1 to 10 months. The best stock selected from the model is MediaTek, and the returns of the stock were all ranked as top two for various time frames with 10 months return as 71.41%. The second and third choices of stocks, HTC and Huaku, both had their returns of various time frames within top four before June, especially for Huaku with best overall return of 83.01%. The Taiwan Cement is the worst choice according to the model and often had actual 2009 returns by various time frames as the bottom six or seven. Although the model does not provide perfectly accurate predictions of stock performance, the overall ranking of stocks is generally consistent with levels of actual returns.

5. Conclusions Equity stock is a key nancial asset of capital market, and researchers have adopted various techniques to analyze factors of its investment decision making. Majority of prior researches discussed the factors independently for their direct effeteness on stock performance, and the interactive impacts between factors were rarely studied. This paper adopts the methods of DEMATEL and ANP to analyze the interdependences between key factors of stock investment decision making. By integrating the dynamic inuence relationship obtained by DEMATEL with ANP, levels of direct and interactive impacts for factors are quantied and ranked, and the outcomes are robust to actual performances of sample stocks. The proposed method also simplies the existing models and raise efciency without affecting the key outcomes. The empirical results provide evidences of signicant interdependent and self-feedback relationships among factors of the three major analytical techniques. The weights of factors are numerically obtained and demonstrate that the most important factor is protability, followed by growth, local institutional investors, risk, moving average, foreign institutional investors, RSI, trading volume, and operation and asset management. The outcome is identical to results produced by existing process, which is more complex than proposed approach. Individual stocks ltered by the ranked factors show consistent predictions of actual performances. The nding consolidated from mathematical theories and judgments of experts is essentially useful to investors where investment strategy can be specically constructed according to the importance of factors and lead to better portfolio management. Acknowledgment Huang likes to thank support from the National Science Council of Taiwan (NSC95-2415-H-155-004). References
Avramov, D., Chordia, T., & Goyal, A. (2006). The impact of trades on daily volatility. Review of Financial Studies, 19(4), 12411277. Avramov, D., Chordia, T., Jostova, G., & Philipov, A. (2007). Momentum and credit rating. Journal of Finance, 62, 25032520. Benjamas, J., Riza, E., & Ramesh, P. R. (2008). Do Thai stock prices deviate from fundamental values? Pacic-Basin Finance Journal, 16(3), 298315. ski, J. (2006). Do institutional investors destabilize stock Bohl, M. T., & Brzeszczyn prices? Evidence from an emerging market. International Financial Markets, Institutions and Money, 16(4), 370383. Brock, W., Lakonishok, J., & Lebaron, B. (1992). Simple technical trading rules and the stochastic properties of stock returns. Journal of Finance, 47(5), 17311764. Cai, F., & Zheng, L. (2004). Institutional trading and stock returns. Finance Research Letters, 1(3), 178189. Chen, P., & Zhang, G. (2007). How do accounting variables explain stock price movements? Theory and evidence. Journal of Accounting and Economics, 43(2-3), 219244.

4.4. Robust test To conrm whether the new method proposed by this study is feasible, the hybrid method suggested by Ou Yang et al. (2008) is used to test the data and compare with above empirical results. Table 9 shows that the results obtained from the hybrid method and the comparisons generally suggest that similar ndings are obtained from the two methods. The smallest score difference between the two methods is the MA (T2) at 0.003 and the largest difference is local institutional investors (I1) at 0.019. The difference in protability (F3), operation and asset management (F4), RSI (T1), local institutional investors (I1), and foreign institutional investors (I2) are all lower than 0.02 and that for risk (F1), growth (F2), moving average (T2) and trading volume (T3) are less than 0.01. Fig. 3 provides a clearer description of the difference between the hybrid method and proposed method, and a clear correlation between the trends is observed. Growth (F2) and protability (F3) are seen as the most important factor for both methods, and trading volume (T3) and operation and asset management (F4) are the least important ones. The relative weights obtained from the ordinary hybrid method are also obtained and compared with the proposed model for the performance of individual stocks. Table 10 provides the outcome of hybrid model and shows that the best asset is MediaTek, with total score of 3.743, followed by HTC at 3.668, Huaka at 3.581, and then by Motech, Cathay FHC, China Steel, and Taiwan Cement. The result of ranking is exactly identical with the outcome by proposed model shown in Table 7. Therefore, we could conclude that the proposed method of the study effectively simplies the research process and raise efciency without affecting the validations of outcomes.

W.-S. Lee et al. / Expert Systems with Applications 38 (2011) 83758383 Cheng, E. W. L., & Li, H. (2007). Application of ANP in process models: An example of strategic partnering. Building and Environment, 42(1), 278287. Chiyachantana, N. C., Jain, K. P., Christine, J., & Robert, A. W. (2006). Volatility effects of institutional trading in foreign stocks. Journal of Banking and Finance, 30(8), 21992214. Cole, R. A., Moshirian, F., & Wu, Q. (2008). Bank stock returns and economic growth. Journal of Banking and Finance, 32(6), 9951007. Darrat, A. F., Rahman, S., & Zhong, M. (2003). Intraday trading volume and return volatility of the DJIA stocks: A note. Journal of Banking and Finance, 27(10), 20352043. Easley, D., & Hara, M. O. (1987). Price, trade size, and information in securities markets. Journal of Financial Economics, 19(1), 6990. Edirisinghe, N. C. P., & Zhang, X. (2007). Generalized DEA model of fundamental analysis and its application to portfolio optimization. Journal of Banking and Finance, 31(11), 33113335. Fama, E. F., & French, K. R. (2006). Protability, investment and average returns. Journal of Financial Economics, 82(3), 491518. Girard, E., & Omran, M. (2007). What are the risks when investing in thin emerging equity markets: Evidence from the Arab world. Journal of International Financial Markets, Institutions and Money, 17(1), 102123. Grifn, J. M., Harris, J. H., & Topaloglu, S. (2003). The dynamics of institutional and individual trading. Journal of Finance, 58(6), 22852320. Grossman, S. J., & Stiglitz, J. (1980). On the impossibility of informationally efcient markets. American Economic Review, 70(3), 393408. Hirose, T., Kato, H. K., & Bremer, M. (2009). Can margin traders predict future stock returns in Japan? Pacic-Basin Finance Journal, 17(1), 4157. Hsieh, L. F., Lin, L. H., & Lin, Y. Y. (2008). A service quality measurement architecture for hot spring hotels in Taiwan. Tourism Management, 29(3), 429438. Ko, K., Kim, K., & Cho, S. H. (2007). Characteristics and performance of institutional and foreign investors in Japanese and Korean stock markets. Journal of the Japanese and International Economies, 21(2), 195213. Lee, B. S., & Rui, O. M. (2002). The dynamic relationship between stock returns and trading volume: Domestic and cross-country evidence. Journal of Banking and Finance, 26(1), 5178. Lee, H., Kim, C., Cho, H., & Park, Y. (2009a). An ANP-based technology network for identication of core technologies: A case of telecommunication technologies. Expert Systems with Applications, 36(1), 894908. Lee, W. S., Tzeng, G. H., Hsu, C. Y., & Huang, J. M. (2009b). Combined MCDM techniques for exploring stock selection based on Gordon model. Expert Systems with Applications, 36(3), 64216430. Li, C. W., & Tzeng, G. H. (2009). Identication of a threshold value for the DEMATEL method using the maximum mean De-Entropy algorithm to nd critical services provided by a semiconductor Intellectual Property Mall. Expert Systems with Applications, 36(6), 98919898. Li, S. T., & Kuo, S. C. (2008). Knowledge discovery in nancial investment for forecasting and trading strategy through wavelet-based SOM networks. Expert Systems with Applications, 34(2), 935951. Lin, C. T., Lee, C., & Chen, W. Y. (2009a). An expert system approach to assess service performance of travel intermediary. Expert Systems with Applications, 36(2), 29872996. Lin, Y. H., Chiu, C. C., & Tsai, C. H. (2008). The study of applying ANP model to assess dispatching rules for wafer fabrication. Expert Systems with Applications, 34, 21482163.

8383

Lin, Y. H., Tsai, K. M., Shiang, W. J., Kuo, T. C., & Tsai, C. H. (2009b). Research on using ANP to establish a performance assessment model for business intelligence systems. Expert Systems with Applications, 36(2), 41354146. Liu, J. N. K., & Kwong, R. W. M. (2007). Automatic extraction and identication of chart patterns towards nancial forecast. Applied Soft Computing, 7(4), 11971208. Lo, A. W., & Mackinlay, A. C. (1988). Stock market prices do not follow random walks: Evidence from a simple specication test. The Review of Financial Studies, 1(1), 4166. Mizuno, H., Kosaka, M., & Yajima, H. (1998). Application of neural network to technical analysis of stock market prediction. Studies in Informatics and Control, 7(2), 111120. Ou Yang, Y. P., Shieh, H. M., Jun-Der Leu, J. D., & Tzeng, G. H. (2008). A novel hybrid MCDM model combined with DEMATEL and ANP with applications. International Journal of Operations Research, 5(3), 19. Saaty, T. L. (1996). Decision making with dependence and feedback: Analytic network process. Pittsburgh: RWS Publications. Samaras, G. D., Matsatsinis, N. F., & Zopounidis, C. (2008). A multicriteria DSS for stock evaluation using fundamental analysis. European Journal of Operational Research, 187(3), 13801401. Tamura, M., Nagata, H., & Akazawa, K. (2002). Extraction and systems analysis of factors that prevent safety and security by structural models. SICE Annual Conference 2002, Osaka, 3(3), 17521759. Tang, G. Y. N., & Shum, W. C. (2003). The conditional relationship between beta and returns: recent evidence from international stock markets. International Business Review, 12(1), 109126. Tzeng, G. H., Chen, W. H., Yu, R., & Shih, M. L. (2009). Fuzzy decision maps: A generalization of the DEMATEL methods. Soft Computing, 14(11), 11411150. Tzeng, G. H., Chiang, C. H., & Li, C. W. (2007). Evaluating intertwined effects in elearning programs: A novel hybrid MCDM model based on factor analysis and DEMATEL. Expert Systems with Applications, 32(4), 10281044. Wang, C. Y., & Cheng, N. S. (2004). Extreme volumes and expected stock returns: Evidence from Chinas stock market. Pacic-Basin Finance Journal, 12(5), 577597. Wang, J. L., & Chan, S. H. (2006). Stock market trading rule discovery using two-layer bias decision tree. Expert Systems with Applications, 30(4), 605611. Wong, W. K., Meher, M., & Chew, B. K. (2003). How rewarding is technical analysis? Evidence from Singapore stock market. Applied Financial Economics, 13(7), 543551. Wu, W., & Xu, J. P. (2006). Fundamental analysis of stock price by articial neural networks model based on rough set theory. World Journal of Modelling and Simulation, 1(2), 3644. Yamawaki, M. T., & Tokuoka, S. (2007). Adaptive use of technical indicators for the prediction of intra-day stock prices. Physica A, 383(1), 125133. Zarandi, M. H. F., Rezaee, B., Turksen, I. B., & Neshat, E. (2009). A type-2 fuzzy rulebased expert system model for stock price analysis. Expert Systems with Applications, 36(1), 139154. Zhu, X., Wang, H., Xu, L., & Li, H. (2008). Predicting stock index increments by neural networks: The role of trading volume under different horizons. Expert Systems with Applications, 34(4), 30433054.