How to Invest in Stocks of Family Business Groups? Case Study of WALSIN Group in Taiwan

The impact of family business groups on the industry has drawn different views. A major challenge for investors is to understand the business model of enterprises and enhance the trading performance in the financial market. Using the data of WALSIN family group, this article adopts generalized autoregressive conditional heteroscedasticity model (EGARCH) to investigate the volatility of stock prices of WALSIN family group. The overall evidence suggests that the relationship between the leverage effect after including the leading company variable and the absence of the leading company appears to be significant at the 1% level. Therefore, the leverage effect after including the leading company has a certain effect on the financial operation of the family business group. In other words, it is helpful to include the leverage effect of the leading company in the portfolio, which can stabilize the trading performance of the WALSIN family group. Furthermore, this study adopts Granger causality and program trading to test the strategy of following the leader of WALSIN family group. The net trading profit during this period is 71.03%. The results show that the technical analysis tested in this study can lead to trading profits and investors can increase their trading profits by following the leader in the family owned business.

Therefore, this study adopts the behavioral finance theory to simulate program trading strategies. This can help individual investors trade the stocks of family groups. The organization of the paper is as follows. Section 2 is the literature review. Section 3 discusses the EGARCH model, Granger causality and the estimation method for program trading. Section 4 describes that data source. Section 5 presents that empirical results and analyses. Finally, the conclusion is presented in the last section.

Literature Review
The literature on stock market predictions can trace back to the random walk theory proposed by Bachelier (1900) who argues that the stock market volatility is similar to the Brownian motion in Physics. Later, Samulson (1965) from the perspective of Modern Economics suggests that stock prices are unpredictable if all information and anticipations of the market participants are included. Fama (1965) proposes that stock market prices are random. Fama (1970) then summaries the past literature and formally proposes an Efficient Market Hypothesis (EMH); that is, a weak form, semi-strong form and strong form of market efficiency.
After 1980s, empirical evidence shows anomalies exist in the market. For example,  finds that the results from NYSE listed companies are inconsistent with the CAPM model of Sharp (1964); that is, the equity premium puzzle and over-reaction puzzle. Other anomalies include the size effect (Banz, 1981), season effect (Lakonishok et al., 1988), calendar effect (Cadsby, 1989) and IPO underpricing puzzle (Reilly et al., 1969). In addition, there are anomalies caused by investors' perception bias, such as anchoring effect, availability heuristics, intuitive heuristics, confirmation bias and framing dependence bias, and investors' psychological characteristics, such as overconfidence, ambiguity aversion, loss aversion, disposition effect and herding effect. Therefore, according to the behavior finance, investors have "bounded rationality".
Compared with the traditional financial theory, behavioral finance does not have well-structured pricing and investment portfolio theory. Some important breakthroughs in the field include the BAPM (behavior asset pricing model) of Shefrin and Statman (1994). Barberis, Shleifer and Visney (1998) argue that investors often over-or under-react. This changes the predictability of nonsystematic risks and therefore, gives rise to a predictable investment return model (that is, the BSV model). Daniel, Hirsheifer and Subrahmanyam (1998)

divide investors into informed and uninformed groups and
investigate the continuity in short-term stock returns and the long-term reversion (that is, the DHS model). Moreover, William and Huang (1995) examine two basic benchmarks in herding effect, stock market returns and dispersion in investment portfolio returns. The smaller the dispersion, the more prominent the herding effect. Investors often conduct noise trading or positive feedback trading. The latter is caused by herding trading behavior, extrapolative expectation and technical analyses. As institutional investors are more informed about their industry partners and have higher ability in prediction, they are more likely than individual investors to have herding trading behavior. In the situation of a lack of information, mimicking the trading behavior of others can reduce the costs of www.scholink.org/ojs/index.php/rem Research in Economics and Management Vol. 2, No. 3, 2017 108 Published by SCHOLINK INC.
gathering information. Also, when making losses, investors can shirk responsibility, blame others and have less regrets. As herding behavior involves many investors, it has great impacts on the market stability and efficiency (Kim & Wei, 1999). Previous studies on herding effect include Lakonishok et al. (1992) who study stock funds in the US between 1985, Froot et al. (1994 who study the herding behavior of institutional investors and Werners (1999) who studies the herding behavior of mutual funds using a larger sample (between 1975 and 1994).
To investigate the complex behavior of financial markets that emerge from decisions made by many traders, Alanyalil et al. (2013)  As for the mean reversion effect, some investors mistake that this effect exists in a short and mid-term period. De Bondt et al. (1985) find that good performing stocks in the previous year also perform better than the bad performing stocks in the following year. They also find a revision effect in stock market returns in the long-run. Fama and French (1988) analyze the NYSE listed stocks between 1926 and 1985 and find that in the long-term, stocks are negatively serial correlated, but positively serial correlated in the short-term. Poterba and Summers (1988) also find similar results using a different sample period and data from more countries. Moreover, Jegadeesh and Timan (1993) report momentum effects in the short and mid-term. Rouwenhourst (1998) adopts Jegadeesh et al.'s (1993 methods and find similar results in 12 European countries. Further, Conrad and Kaul (1988) show that adopting a reversion trading strategy in the long-term and a momentum strategy in the mid-term is a successful trading strategy.
All the above suggests that anomalies exist in the market. Hence, this study proposes a new testing method. Specifically, we use econometric models and optimized program trading and the trading strategy proposed by behavioral finance to conduct two stages of tests in order to find an investment strategy for investing family-owned business groups. The hypothesis tested is as follows: After selecting a leading company in the industry, we can use technical analyses to enhance the trading performance of family-owned business groups.

Theoretical Models and Estimation Methods of VAR and Granger Causality
Sim ( (i.e., k = 1) can be presented as: The error term t  is white noise; m is the constant, a is the coefficient and Ω is the positive definite variable and covariance matrix. Granger (1969,1988) and the application of Granger causality analysis proposed in Lan et al. (2017) can be used to clarify the relationship between the two.

Theoretical Model and Estimation Method of EGARCH
The Generalized Autoregressive Conditional Heteroscedasticity Model (GARCH) is developed by Bollerslev (1986) based on a modification of Engle's (1982) in the previous period t-1. The model is as follow: and GARCH (p, q) model overcome the condition of non-negative  . Therefore, the model can be presented as below: Research in Economics and Management Vol. 2, No. 3, 2017 110 where h t is the conditional variance of the GARCH model, p is the order of GARCH terms h 2 and q is the order of ARCH terms  2 . As in finance, negative news often has greater impact on stock prices than positive news. The conditional variance of EGARCH model becomes: If the coefficient of leverage effect r ≠ 0, it implies that there is an asymmetric response of conditional variance on positive error term and negative error term. From the volatility viewpoint, this study compares the impact on stock market when the data of WALSIN is included and excluded.
This study follows the method in Lan et al. (2014) and draws the news impact curve. The method is briefly described below: . From the EGARCH model, we estimate the conditional variance series  2 and take the square root, which is then divided by the error term to derive z.
2) Rank z from the lowest to the highest and structure a new series containing z.
3) Use the coefficients of the EGARCH model to derive s: 4) Plot z and s on a graph to draw the news impact curve and to observe the impact on the stock market.
If the curve tilts upwards to the left with a large angle, it suggests a high degree of panic.

Experimental Design and Estimation Method
In traditional behavioral finance, there are two research methods in experimental designs. One is the Structural Equation Modeling (SEM) and the other is experimentation method. This study adopts a different method using Multi Charts to simulate a model for following leading company. In this study, the leading company is WALSIN in model 1 and HannStar Board in model 2. Data 2 is the price of 0050; Data 3 is the price of leading company. In addition, the trading strategy using RSI technical index in this study is based on the closing price and the breakthrough by the 20-day moving average. Three conditions of a "system buy" are: (1) today's closing price of data 2 is greater than 20-day moving average price of data 2; (2) today's closing price of data1 is greater than the 20-day moving average price of data1; and (3) the RSI of today's stock prices is higher than the best buying point's RSI (the simulation interval is 50~80). Conversely, three conditions of a "system sell" are: (1) today's closing price of data2 is lower than 20-day moving average price of data 2; (2) today's closing price of data1 is lower than the 20-day moving average price of data1; and (3) the RSI of today's stock prices is lower than the best selling point's RSI (the simulation interval is 20~40). We use the optimized trading to find the optimal number of days in moving average. The position is closed out if the profit is greater than 500 points or the loss is greater than 100 points.

Data
Based on the information provided by Unique Business Weekly (Issue 1016, p. 14), WALSIN family business group includes WALSIN (1605), HannStar Board (5469), Walsin Technology (2492) In order to ensure that the model is completed, the experiment is divided into two stages. The first stage covers from 2010.9.27~2013.10.31 and the second stage covers from 2010.9.27~2017.1.19. The second stage adopts optimal parameters from the first stage. The trading cost in the simulated models is assumed to be about 1% of the stock price.  (0)), it is a I(0) stationary series. However, when intercept and trend are considered (-2.7740(0)) or when no intercept and trend are considered (-1.3614(0)), null hypotheses are not rejected in both cases. That is, the series are not stationary, have fat tail and have autocorrelation. Therefore, I(0) is not stationary. After taking a difference, I(1) becomes stationary and rejects the null hypothesis (Table 1). Therefore, we can proceed with VAR and Granger causality test.  Note: According to Mackinnon (1991), *, **, *** shows significance level at 1%, 5% and 10%. (0) shows that when the lag period is 0, it has the minimal AIC. Sample code is as provided in Section 4.

Lag Period Test of WALSIN's VAR Model
Before proceeding with VAR model estimation, we must first test the lag period. The results show that the AIC, SC, HQ and FPE of WALSIN are at their minimum when the data is lagged one period (Table   2). Therefore, the VAR is estimated using lag period of one. According to AR roots graph, the reciprocals of the variables' unit roots are all within the circle (Figure 1). Therefore, the model is proven to be stationary.         (Table 4). Therefore, this model is estimated using a lag period of two.  In other words, these two variables refuse to reject the information of each other. Therefore, we can proceed with the next step of investment simulations.

Estimation Results of EGARCH Model (Excluding WALSIN)
The model in this section does not include WALSIN and shows that all variables are significant at the 1% level where α = 1.458280, β = 0.865143 and γ = -0.234075. This suggests that the leverage effect has greater impacts on psychological panicking reaction when there is bad news. The leverage effect from good news can be presented as: 1.224205 = (1.458280 -0.234075). The leverage effect from bad news can be presented as: 1.692355 = (1.458280 + (-0.234075) * (-1) ( Table 6). The next step is to draw the news impact curve from the EGARCH results (excluding WALSIN) ( Figure 2). The Figure shows that when the news impact is less than 0, the bad news impact curve is steeper and the good news impact curve is more flat. The results suggest that bad news will cause stock prices to become more volatile. That is, the sample without WALSIN faces greater risks.   (Table 7). Similarly, we can draw the news impact curve of EGARCH model (including WALSIN) (Figure 3).
The Figure shows that when the news impact is less than 0, the bad news impact curve is more flat and the good news impact curve is steeper. The results suggest that good news will cause stock prices to become more volatile. That is, the sample with WALSIN faces lower risks.   The results show that only two companies in model 2 reduce investment returns in the second stage. If the investment portfolio includes ten WALSIN business group companies, the trading is profitable (Table 9). Between 27 September 2010 and 31 December 2013, the profit from the investment portfolio is $78. When the sample period is extended to 19 January 2017, the profit increases by $19.36 (21.97%), which is better than the overall market of 8.2% but worse than model 1 (50.35%). The results suggest that the model with WALSIN as the leading company is better than the model with HannStar Board as the leading company.

Conclusion and Discussion on Investment Strategy
Researchers have different views on the impact of family businesses on the industry. For investors, they must understand business models of enterprise groups in order to enhance their trading performance.
This study analyses the impact of WALSIN family business group on the industry and examines the influences of leading company on the related family business companies.
The news impact curve based on the EGARCH model (including the leading company WALSIN) shows that when the news impact is less than 0, the curve is more flat in the case of bad news. On the other hand, good news impact curve is steeper, suggesting that good news will cause greater volatilities in stock prices. That is, the risks are lower when WALSIN is included. The results suggest that including the leading company, WALSIN, in the investment portfolio is good for the stability of market trading.
Further, when including WALSIN and HannStar Board as the leading company in model 1 and 2, respectively and extending the examination period to 19 January 2017, the net profits increase by 50.35% and 21.97%, respectively. The increment is larger than the overall market of 8.2%. The results suggest that this program trading can lead to profits and model 1 is better than model 2. Therefore, investors can employ the property of leading company to enhance their trading performance in financial markets. Due to the space and time limit, future research could study the stocks of other family groups in Taiwan to conduct optimal back testing and run the simulations.