The effect of financial deregulation on stock markets interdependence and efficiency : evidence from Shanghai-Hong Kong Stock Connect

(1)

MSc Business Economics, Finance track

Master Thesis

The effect of financial deregulation on stock

markets interdependence and efficiency: evidence

from Shanghai-Hong Kong Stock Connect

Date: 06-07-2015

Student Name: Liwei Luo

Student Number: 10824162

(2)

1

Statement of Originality

This document is written by Student Liwei Luo who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

2

Acknowledgement

I would like to express my sincere gratitude to my supervisor, Dr. Vladimir Vladimirov, for checking the correctness of the results in this paper and giving me a lot of help throughout the thesis.

I would also like to thank my second reader for giving me some useful advice for this paper, even though I do not know who he or she is.

(4)

3

Abstract

The purpose of this paper is to study how financial deregulation affects interdependence between different markets and their respective market efficiency through implications of the Shanghai-Hong Kong Stock Connect. From an index-level perspective, this paper studies the interdependence between Shanghai and Hong Kong market by focusing on bilateral causality relation between A-share, B-share markets in shanghai and stock market in Hong Kong. Conducting error correction model Granger Causality tests, I found that this program facilitates return causality from Shanghai A-share to Hong Kong stock market. All the eligible shares under the program represent most of the total market capitalization in each market, which makes the performance of all these shares very close to market performance. Hence, for company-level tests, I conduct event studies on companies dual-listed in Shanghai and Hong Kong instead of all the eligible shares and find that this program has a greater impact on these companies’ A-share in Shanghai than on H-share in Hong Kong, which is indicative of a higher efficiency in Hong Kong Stock market. However, the reduced trading restrictions due to this financial deregulation did not reduce or eliminate the valuation gap between A-share in Shanghai and H-share in Hong Kong as expected.

(5)

4

Introduction

There are generally two types of capital controls in a country. One is to impose restrictions on the foreign ownership of its domestic equity markets. The other is to control domestic capital outflow to foreign capital markets. Financial deregulation to some extent reduces trading restrictions generated by these two forms of capital controls. The objective of this paper is to study the impact of the new financial deregulation in China ---Shanghai-Hong Kong Stock Connect--- on the relationship between these two stock exchanges and their market efficiency. Shanghai Stock Exchange (SSE) consists of two segmented stock markets, share and B-share markets. A-share securities are quoted in RMB (Chinese currency) and traded mostly by residents in Mainland China. B-share securities in SSE are traded in US dollar and mainly by foreign investors. Hong Kong, as a special administrative region of People’s Republic of China, has a high degree of autonomy and therefore exemption from many financial controls of the central government. Hence, its capital market has a higher level of financial liberalization than that of its Mainland counterparts. Even though Shanghai and Hong Kong are both part of China, there exist many trading restrictions between their stock exchanges. Prior to Shanghai-Hong Kong Stock Connect, equity markets in Mainland China were highly regulated and open only to part of foreign investors. Only licensed foreign institutional investors1 can trade A shares listed in Mainland China. Foreign retail investors were not allowed to trade in A-share markets. Likewise, only licensed domestic investors2_in Mainland China were able to buy securities listed in Hong Kong. Goetzmann, Li, and Rouwenhorst (2005) show that emerging markets play an increasingly important role in international diversification. Aiming at improving interaction and reducing trading barriers between Shanghai and Hong Kong stock exchanges, this Stock Connect program provides a new opportunity for investors from both sides to diversify their portfolio. So the study on this program would draw much attention from investors seeking diversification benefits. Another purpose of this program is

1_{Prior to Shanghai-Hong Kong Stock Connect, foreign investors can only trade shares in Mainland China through two programs:}

Qualified Foreign Institution Investors (QFII starting in 2007) and Renminbi Qualified Foreign Institution Investors (RQFII starting in 2011) programs. Only institutional investors can apply for the two program. And for them to be qualified, they must meet a strict criteria set forth by the China Securities Regulatory Commission (CSRC), including minimum thresholds of capital, years of business experience, and assets under management. Prior to QFII and RQFII, foreign investors were not allowed to invest in capital markets in Mainland China.

2_{Similarly, domestic investors can only invest in foreign capital markets through Qualified Domestic Institutional Investors (QDII}

starting in 2001) and Qualified Domestic Individual Investors (QDII2 starting in 2013) programs before the Stock Connect program. There are also a series of strict requirements for applicant to meet. Prior to QDII and QDII2, domestic investors were not able to trade foreign financial products.

(7)

6

to attract capital inflows, its effectiveness would therefore be of much concern to policy makers. In this paper, I will analyze the implication of this program from two angles.

First, I study the impact on market interdependence between the two stock exchanges by conducting index-level tests before and after this program. The tests in this part are mainly in the context of the return causality and co-integrating relationship between Shanghai A-share index, B-share index and Hong Kong Hang Seng market index. Shanghai-Hong Kong Stock Connect creates a link between Shanghai A-share market and Hong Kong stock market. So the economic mechanism behind interaction between A-share market index and Hong Kong Hang Seng market index is quite straight-forward. Financial openness reduces trading restrictions, which in turn would theoretically strengthen relation between these two markets. Whereas the situation for B-share market is a bit more complicated. The performance of B-B-share market could be influenced by both positive and negative transmission effect from this program. The positive is from the effect of this good news on the overall market. The negative one can be explained by demand and supply. Sun and Tong (2000) argue that H-share and red-chip listed in Hong Kong Stock Exchange are good substitutes for B-share. Under this program, A-share has been largely though not fully open to foreign investors, which could make it a possible substitute for B-share. This being the case, the demand for B shares should be decreased due to the increasing demand for A-share and Hong Kong stock, which in turn might lead to underperformance. Hence, the final situation hinges on which side of force to dominate. Taking all above into consideration, the examination of the relationship between these three pairs of indices sheds light on how these indices interact with each other, which may be of interest to both policy makers and investors.

Many previous studies have documented that stock prices tend to follow a pattern of random walk, which probably leads to spurious regression. In order to avoid this problem, I calculate the first difference for the three indices: Shanghai A-share (SHA), Shanghai B-share (SHB), and Hong Kong Hang Seng (HKHS) indices. These three indices make up three index pairs (SHA-SHB, SHA-HKHS, and SHB-HKHS). I then conduct bivariate error correction model (ECM) Granger test among these pairs. In order to construct the error correction term, I test for co-integrating relationship within each pair before testing the causality. Intuitively, the launch date of this program may serve as a breaking point among the linkage between these markets. I use Chow test to examine whether the launch of this program would lead to a structural break in stock markets’

(8)

7

causality relationships. In robustness check, I construct another econometric model based on dynamic correlation within the three index pairs to check the index-level results.

Second, I examine the market efficiencies of these two markets by conducting event studies on companies dually listed in Shanghai and Hong Kong. These dual-listed companies issue both A-share in Shanghai and H-share in Hong Kong. Their A-share and H-share usually trade at different price, which causes the existence of “A-and-H price difference” phenomenon. Much literature on disparities in the pricing of cross-listed stocks points out that the differential required returns between foreign and domestic investors may be the reason behind this phenomenon. Investors can be able to diversify their investment portfolio by buying foreign stocks. If so, foreign investors may be willing to pay a higher price than domestic investors. Nevertheless, studying the price premium between Chinese A-share and B-share, Bailey (1994) found an opposite phenomenon in China---domestic investors pay a higher price than foreign investors. Applying the “investor sentiment” notion introduced by Lee et al. (1991), Ma (1996) argues that relatively limited investment alternatives in Mainland China can induce a high speculative sentiment in its stock markets. He further suggests that information asymmetry due to market segmentation can also be a contributing factor to the price difference in segmented markets. Foreign investors may not have access to as much information about local firms as domestic investors. Hence, foreign investors may require a compensation for this, which would to some extent lower share price. There are also some arguments regarding different level of market efficiency in the two markets. In my paper, I will examine how A-and-H price difference reacts to Shanghai-Hong Kong Stock Connect and compare the market efficiency of these two markets by investigating both A-share and H-share market response to the regulatory change. In addition, I explore the relationship between A-share’s market reaction to this program and its corresponding firm characteristics, which may be useful to investor when making investment decision.

The main contribution of this paper is to provide empirical studies on this recent program and add new evidence from this program to the existing literature. Policy makers may care about the index-level analyses, while investors may focus on the company-level results. China is gradually opening up its capital market. Currently, the scheme is limited to part of shares listed in Shanghai, but the extension plans for Shenzhen Stock Exchange---another exchange in Mainland China---are already under the way. The expansions of capital markets China---are likely to experience a similar

(9)

8

pattern. So this paper may also contribute to later studies on further deregulations in Chinese capital markets.

The rest of this paper is organized as follows. Section I provides background information about Chinese stock markets and the Shanghai-Hong Kong Stock Connect. Section II presents a review of pertinent literature. Section III describes methodology. Section IV describes data and descriptive statistics. In Section V, I describe my empirical tests and discuss the results. Section VI presents robustness checks and additional results. Finally, Section VII concludes.

Background

Chinese Stock Markets

There are two stock exchanges in Mainland China: Shanghai Stock Exchange and Shenzhen Stock Exchange. Companies in China can issue A-share, B-share and H-share. Both A-share and B-share are listed in Mainland China, i.e. Shanghai or Shenzhen bourse. H-share is listed in Hong Kong Stock Exchange. Initially, A shares were traded only by citizens in Mainland China and B and H shares only by Hong Kong and Foreign investors. Exchanges in Mainland China were established much later than that of Hong Kong, so its stock markets are less developed than Hong Kong. In order to keep up with Hong Kong or global market, China has been trying different types of financial reforms. B-share became available to domestic investors on Feb 20, 2001because the trading restrictions has been removed by Chinese government. Foreign institutional investors were allowed to invest in A-share in the following year, while they first need to be approved by China Securities Regulatory Commission. Financial institutions in Mainland China had not been able to invest in overseas capital markets until 2006. Because the process of being approved for both domestic and foreign institutions is relatively complicated and rigorous, the trading restrictions problem had not largely been resolved. Moreover, trading between these markets has historically been limited by strict quotas. A further financial reform was still needed to reduce trading restrictions, which is how Shanghai-Hong Kong Stock Connect emerged.

Shanghai-Hong Kong Stock Connect

Shanghai-Hong Kong Stock Connect, as manifested by its name, creates a link between Shanghai Stock Exchange (SSE) and Stock Exchange of Hong Kong (SEHK). It was announced on 10 Apr 2014 and launched on 17 Nov 2014. This program essentially provides two new trading

(10)

9

channels consisting of Northbound and Southbound Trading. Geographically, Shanghai is situated at north of Hong Kong. So northbound trading is one new trading channel from Hong Kong to Shanghai Stock Markets, allowing global investors to invest in eligible shares listed in Shanghai. In contrast, southbound trading is another new trading channel from Shanghai to Hong Kong, allowing residents in Chinese Mainland to trade eligible shares on Stock Exchange of Hong Kong. Eligible shares in northbound trading consist of 568 stocks in Shanghai, including constituent stocks of SSE 180 index3_{and SSE 380 index}4_{and SSE-SEHK A+H shares}5_{, which}_{account for}

90% of Shanghai stock market’s capitalization6. Shares eligible to be traded through the Southbound Trading Link comprise all the constituent companies of the Hang Seng Composite Large-Cap Index and the Hang Seng Composite Mid-Cap Index, and shares of all companies listed on both SSE and SEHK. Under this program, investor base has been increased in both Shanghai and Hong Kong stock markets and trading quotas has been increased by 50%. This is a huge step for China to open up its capital market to the world.

Literature Review

This paper mainly contributes to three strands of the literature. First, I provide empirical evidence on stock market interdependence. Many studies use return or volatility transmission mechanism between different stock indices to investigate the interdependence of their respective stock markets. For instance, employing a multivariate GARCH model of international transmission of stock returns and volatility, Karolyi (1995) finds that the bi-directional causality between the United States and Canadian stock markets has been overstated. Using the same method as Karolyi (1995) to test the interaction among different US sector indexes, Hassan and Malik (2007) show that information affecting a certain sector would have spillover effects on all sectors because of their interdependence. Another method often used to test causality is Granger Causality test. For example, Aaltonen and Ostermark (1997) conduct a rolling test of Granger Causality between two

3_{SSE 180 Index selects 180 of the most representative stocks listed on SSE based on sector representation, size and liquidity to}

reflect the overall situation and operation of Shanghai securities market and serve as performance benchmark as well as underlying instrument for financial derivatives

4_{SSE 380 index consists of the 380 stocks with Midcap, high growth and good earning records, which aims to comprehensively}

reflect the performance of the Shanghai new blue chip stocks.

5_{SSE-SEHK A+H shares refer to shares issued by companies dual-listed in Shanghai and Hong Kong.}

6 _{This makes the}_{performance of all the eligible shares very close to market performance, which is why I conduct event studies only}

(11)

10

stock markets (free and restricted share markets) in Japan and Finnish stock market and find that the relationship with Finnish stock markets seems to be stronger for free share markets than the restricted one. Using a vector auto regression (VAR) model to study the causality between Indian ADR and its respective domestic shares, Hansda and Ray (2003) show that although there is a dominant-satellite relation between NASDAQ and Bombay Stock Exchange, Indian ADR and its domestic counterparts Granger-cause each other. Using the impulse response analysis of a nine-country VAR model, Eun and Shim (1989) demonstrate that the U.S, market can Granger causes the other countries, while the opposing relation is not statistically significant. For Chinese stock markets, using a four factor VAR model, Lin and Swanson (2008) examine the causality relationship among four internal Chinese markets and find that there is no causality between Shanghai A-share market and Shanghai B-share market, whereas Shenzhen A-share market statistically Granger causes Shenzhen B-share market. They further conduct causality tests between stock markets in Mainland China and regional markets and find that Stock markets in Mainland China Granger causes Hong Kong without reverse effects. Similarly, Yeh and Lee (2000) find the same relation between stock markets in Mainland China and Hong Kong. However, they show that Hong Kong stock market Granger causes B-share markets in both Shanghai and Shenzhen. In this paper, I add an error correction model to the Granger Causality test and use it to examine the relation between Shanghai A-share, B-share and Hong Kong stock markets before and after the event and further conduct a chow test to examine whether this program causes a structural change in the causality relationship among these indices.

Secondly, this paper is related to the price difference phenomenon in cross-listed firms. The most common reason used to explain this economic phenomenon is market segmentation theory. Domowitz et al. (1998) point out that cross-market differences are primarily caused by segmented markets. Consistent with this view, Sun and Tong (1999) further study the effect of market segmentation on stock performance in china and find that different required returns and information asymmetry, and different market sentiments arose from market segmentation in Mainland China and Hong Kong have explanatory power to the price gap between A-share and its corresponding H-share. The values of different classes of share for one company should theoretically be the same because these shares are entitled to the same rights (Cai et al. (2010)). Nevertheless, this situation exists only in a perfect world. According to the argument of different required returns, the share targeting at foreign investors should be priced higher than its counterpart

(12)

11

traded by domestic investors in that foreign investors can gain diversification benefits through international investment. Asymmetric information argument assumes that foreign investors cannot easily glean information about local companies and therefore need a compensation for this. Different market sentiment leads to different degree of market reaction to a certain event. These three factors may drive market price towards different directions and to different extent in different times. From a different angle to test market segmentation, Merton (1987) documents that an increase in the size of a company’s investor base will lower investors’ expected return and therefore increase the market value of the company’s securities. Many analysts anticipated that the Stock Connect program would reduce trading restrictions in segmented markets and therefore the valuation gap. My study provides visual inspection on how the A-and-H price difference behaves before and after the Stock Connect.

Finally, this paper is linked to the efficient market hypothesis. The concept of market efficiency was formally introduced by Fama (1970). He proposes that a market in which security prices at any time fully reflect all available information is efficient. In such a market, the cost of acquiring certain information and transaction fees should equal the potential profits made from the information. Hence, market participants cannot earn abnormal return based on the available information. By examining the weak-form efficient market hypothesis (EMH) and the role of the banks, Groenewold et al. (2003) find that the efficiency in Chinese stock markets tended to decrease when the banks are excluded from stock markets and increase when they are re-admitted. Applying parametric and non-parametric variance ratio test, Fifield and Jetty (2008) show that Chinese A-share markets are more efficient than B-share markets. By using rolling multiple variance ratio tests to reexamine the weak-form EMH in Chinese stock markets, Hung (2009) finds that only Shanghai A-share market is the weak-form efficient market. Comparing the market efficiency before and after the removal of trading restrictions in B-share markets, he argues that market efficiency tends to be improved following financial deregulation and liberalization. In this paper, I conduct event study to test for the significance of abnormal returns of dual-listed shares (A-share in Shanghai and H-share in Hong Kong) around the Stock Connect and further compare the efficiency of the two markets. I also study the relation between the abnormal return and firm characteristics.

(13)

12

Methodology

The empirical test in this paper is divided into two main parts. The first part is performed at the index level and its main objective is to examine the bilateral relationship between A-share and B-share market in Shanghai and stock market of Hong Kong and whether there is a structural change in these relationships prior and subsequent to the Shanghai-Hong Kong Stock Connect. The second one is at the firm level and it aims to address how companies dually listed in Shanghai and Hong Kong respond to the Stock Connect program in their respective markets (A-share in Shanghai and H-share in Hong Kong).

Index-level Tests

In order to test the first one, I conduct a Granger Causality test under the context of a bivariate vector error correction model (VECM) and use Chow test to further test for the significance of the change in the causality relationships. Before conducting the Granger test, I examine the stationary and co-integration relation of time-series data. To be more specific, the first part proceeds as follows:

1. Lag length selection using Akaike information criterion (AIC)

Many economic time series have been shown to be correlated with its lag values. In order to construct more accurate models, I use Akaike information criterion (AIC) to select the lag length to be included in estimation models. But how many lags should be included in an estimated auto-regression model? In determining the order of an auto-auto-regression, Stock and Watson (2011) explain that if the number of lags is too high, additional error would be introduced into regression model, while if the order is too low, useful information contained in the more distant lagged values could possibly be ignored. So a more accurate choice requires balancing these two aspects. The most widely used method in practice is Akaike information criterion (AIC), which is

𝐴𝐼𝐶(𝑝) = ln [𝑆𝑆𝑅(𝑝)

𝑇 ] + (𝑝 + 1) 2

𝑇 (1) The value of p represents the number of lags and is the estimator of interest. SSR (p) is the sum of squared residuals of the estimated auto-regression model AR (T is the total number of observations). The first term in equation (1) decreases with p, whilst the second one increases with p. This equation trades off these two forces. The value of p that minimizes AIC (p) is optimal. In this paper, I choose three p values for A-share and B-share in Shanghai and Hang Seng stock

(14)

13

market index in Hong Kong respectively. The lag lengths selected in this step would be used in models in later tests.

2. Unit root test using Augmented Dickey Fuller test

A large number of studies have shown that stock price in many countries has a stochastic trend, which is also called unit root. For instance, Wang et al. (2014) demonstrate that stock prices of seven Asian countries (Mainland China, Hong Kong, Japan, South Korea, Singapore and Malaysia) follow a random walk over a period from Dec 1990 to Mar 2013. Using a variable having a stochastic trend could lead to biased estimations. Stock and Watson (2011) list three problems caused by stochastic trends: zero autoregressive coefficients, non-normal distributions of t-statistics and spurious regression (two time series appear related when they are not). In order to avoid these problems, I use the first difference of the logarithm of daily index prices, which is approximately equal to daily return7. But at first I would conduct an Augmented Dickey Fuller test with a deterministic trend term to see if these time series have a trend-stationary (without the trend term, the alternative hypothesis would be the series is stationary, not stationary around a deterministic linear time trend). The Dickey fuller regression model is as below:

ΔY t = β0 + αt +δ Y t-1 + γ1 ΔY t-1 + γ2 ΔY t-2 +…+γ p ΔY t-p + u t (2) where ΔY t represents the first difference of the logarithm of daily prices for each of the three

indices at time t. P is the lag length selected according to AIC. δ is the coefficient of interest under the null hypothesis: δ=0 meaning the time series has a unit root and the alternative hypothesis is the series is stationary around a deterministic linear time trend. The model is used to all the unit root tests for Shanghai A-share index and B-share index and Hong Kong Hang Seng index.

3. Co-integration Test

After running unit root test for Shanghai A-share index (SHA) and B-share index (SHB) and Hong Kong Hang Seng index (HKHS), I test their co-integration relationships in pair (SHA-SHB, SHA-HKHS, SHB-HKHS). Two series that are said to be co-integrated have a common stochastic trend. If they do have the same stochastic trend, then computing their difference can eliminates this trend. Hence, co-integration test is essentially unit root test. If the co-integrating coefficient is

7_{The change of the logarithm of a variable is close to the proportional change of that variable: ln (Y}

t) - ln (Yt-1) = ln (Yt-1+ΔY) -

(15)

14

unknown, Engle Granger Augmented Dickey Fuller (EG-ADF) test is appropriate for co-integration test (Stock and Watson (2011)). It is a two-step procedure. The first step is an OLS regression, which is

Y t = α + θ X t + z t (3) θ is the co-integrating coefficient. Y t and X t represent market prices of indices in each pair at

time t. The second step is basically a Dickey Fuller t-test as the one in unit root test except for no time trend. To be more precise, the second step is to test whether the residual, z t from regression (3) has a unit root using an ADF model with an intercept but no time trend.

ΔZ t = β0 +δ Z t-1 + γ1 ΔZ t-1 + γ2 ΔZ t-2 +…+γ p ΔZ t-p + u t (4) The null hypothesis is the same as regression (3), that is, δ=0, which means Z hasa unit root. The alternative hypothesis, however, is that Zis stationary instead of trend-stationary in the case of regression (3). If Z is stationary, then Y t - θ X t is stationary and the common stochastic trend in Y and X could be eliminated. So if the null hypothesis is rejected, then Y and X are co-integrated.

4. Granger CausalityTest

The aim of Granger Causality test is to detect the direction of transmission effect in each pair. If the results of last step show there does exist co-integration relation between a certain pair, I would construct vector error correction model (VECM) based on the co-integrating coefficients to Granger-test for this pair. The Granger test regression based on VECM is shown as follows:

ΔY t = β 10 + β11 ΔY t-1 + β12 ΔY t-2 +…+ β1 p ΔY t-p + γ11 ΔX t-1 + γ12 ΔX t-2 +…+γ 1 q ΔX t-q + α1 (Y t-1 - θ X t-1) + u 1t (5)

Δ X t = β 20 + β 21 ΔY t-1 + β 22 ΔY t-2 +…+ β 2 p ΔY t-p + γ 21 ΔX t-1 + γ 22 ΔX t-2 +…+γ 2 q ΔX t-q + α 2 (Y t-1 - θ X t-1) + u 2t (6)

For equation (5), the null hypothesis H0: γ11 = γ12 =…= γ1q = 0 For equation (6), the null hypothesis H0: β11 = β12 =…= β1p = 0

If the null hypothesis of equation (5) is rejected, then ΔX Granger causes ΔY. If the null hypothesis of equation (6) is rejected, then ΔY Granger causes ΔX. ΔY t and ΔX t represent daily return of the two indices in each pair at time t. Y t-1 - θ X t-1 is the error correction term in which θ is a

(16)

long-15

term multiplier representing the long-run relationship between market prices of the paired indices. γ11, γ12…γ1p and β11, β12… β1p model short-run responses. P and q are the number of lags for ΔY and ΔX respectively. α1 and α 2 are both adjustment parameter. When the past values are above the equilibrium value, they willtheoretically drag the values down in the next period. Similarly, if the previous values are below the equilibrium value, they will move upwards under the influence of the adjustment parameter.

If there is no co-integrating relationship between the paired indices, the regression model used in Granger test excludes the error correction term in VECM. Then it is simply a two-factor vector auto regressive (VAR) model, which is

ΔY t = β10 + β11 ΔYt-1 + β12 ΔYt-2 +…+ β1 p ΔY t-p + γ11 ΔXt-1 + γ12 ΔX t-2 +…+γ 1 P ΔX t-p + u 1t (7) ΔX t = β 20 + β21 ΔY t-1 + β22 ΔY t-2 +…+ β2 p ΔY t-p + γ21 ΔX t-1 + γ22 ΔX t-2 +…+γ2 P ΔX t-p +u 2t (8) Except for the error correction term, other specifications and null hypothesis remain similar to

regression (5) and (6).

The Granger Causality test is performed under three sub-periods for each pair to compare the degree of significance in each period. The first period is 6-month from the announcement of the Stock Connect program backwards, the second is from the announcement to the launch date and the final period is 6-month from the launch date onwards. The comparisons shed light on how the announcement and launch of this program affect the relations between these stock markets.

5. Chow Test

Shanghai - Hong Kong Stock Connect is an important financial change in China. It reduces trading restrictions between stock markets in Shanghai and Hong Kong and opens capital market in Mainland China to retail investors around the world. Chow test in this paper can detect whether there exists a structural change in the causality relationship between the two markets due to this unprecedented program. The regression model used is as follows:

ΔY t = β0 + α 1 ΔXt-1 +…+ α p ΔX t-p + γ0 D + γ1 [ΔXt-1× D] +…+γ p [ΔX t-p ×D] + u t (9) This model is used to test whether there is a break caused by the launch of the Stock Connect program. D t is a binary variable that equals 0 before the launch date and 1 after. P is the lag length selected in earlier tests. If there is not a structural change, then the regression function is the same

(17)

16

before and after a certain day, so the binary variable do not affect the regression. Hence, the null hypothesis for equation (9) is γ0 = γ1 =0, which means there is no break. If the null hypothesis is rejected, then structural change occurs.

6. Dynamic Correlation test

Using moving correlation within the three index pairs as a measure of market interdependence, I conduct an OLS regression to check the robustness of the index-level results. In this part, I incorporate the announcement and launch date of the Stock Connect program into one model to analyze the change in paired indices’ correlation over time. The regression function is as below:

ρ i j ,t = α 0 + α 1 D1 + α 2 D2 +u t (10) ρ i j ,t denotes the dynamic correlation coefficient of index i and j at time t. D1 and D2 are binary variable. D1 equals 0 before the announcement date and 1 between the announcement date and the launch date. D2 equals 0 before the launch of this program and 1after. The coefficients of interest are α 1 andα 2, which reflect how the two dates affect the co-movement of each index pair.

Company-level Tests

1. Event Study

In order to analyze the effect of the Stock Connect program at the company level, I conduct an event study on companies cross-listed in Shanghai and Hong Kong which is mainly the second part of empirical test in this paper. The details are as follows:

Event window: ±20 Days (41-day window)

Normal Return Model ∶ 𝑅𝑖𝑡 = 𝛼 + 𝛽𝑅𝑚𝑡+ 𝜀𝑖𝑡

Aggregate Abnormal Return: 𝐶𝐴𝑅̂ (−20, +20) = ∑_𝑖𝑡 20_𝑡=−20𝜀̂_𝑖𝑡 The Average across firms: 𝐶𝐴𝑅̅̅̅̅̅̅̅̅(−20, +20) =_𝑖𝑡 1

𝑁∑ 𝜀̂(−20, +20)𝑖𝑡 𝑁

𝑖=1

The final step is to test for significance using the following statistic: 𝐽 =𝐶𝐴𝑅̅̅̅̅̅̅̅̅(−20,+20)𝑖𝑡

𝜎

The null hypothesis in the event study is that average cumulative abnormal return equals zero, which means the event does not affect the stock performance of these companies.

(18)

17 2. OLS regression

In order to further examine whether the level of abnormal return has to do with firm characteristics, I run an OLS regression as follows:

CARs i = β0 + β1 MV i + β2 STATERATIO i + β3 AGE i + β3 PE i + β4 I i +u i (11) CARs i is cumulative abnormal return of company i. MV i, AGE i and PE i represent the market value, the age and PE ratio of company i respectively. STATERATIO i denotes the ratio of state-owned share to the total number of outstanding share. Ii represents the industry the company i belongs to. β1,2,3,4 captures different companies’ reaction to this program. I further test whether CAR is different for financial and non-financial industry. In this case, Ii equals 1 for financial industry and 0 for non-financial.

Data and Descriptive Statistics

Data

The data used in this paper is composed of three datasets. The first one is time series data used for empirical test at the index level, consisting of Shanghai A-share, Shanghai B-share and Hang Seng market index. I begin with the daily close prices of these three indices from RESSET and then construct variables needed for the later empirical tests. First, I take the logarithm of the daily prices and calculate the daily returns of these three indices based on the logarithmic prices:

R i, t = (ln pi, t – ln pi, t-1) × 100

For each index pair, I use these daily returns to compute the time series of stock price growth correlation. I calculate the dynamic correlation over a 126-trading day (6-month) rolling window. This correlation and the integration relationship are two main measures of stock price co-movement in this paper. The main study period is from 10 Nov 2013 to 1 Apr 2015. I further divide this period into three sub-sample periods, from 10 Oct 2013 to 10 Apr 2014, 10 Apr 2014 to 17 Oct 2014, 17 Oct 2014 to 1 Apr 2015, respectively. The first one is a 6-month period before the announcement of the Stock Connect program. The second covers the time period from the announcement date to a month before the launch date. The third ranges from a month before the

(19)

18

launch of the program to the most recent data available on RESSET. I take a month before because the launch of this program had been actually delayed for a month.

The second dataset is a panel data used for event study. The original data contains information about 69 dual-listed firms from 1 Jan 2014 to 1 Apr 2015. It includes daily close prices for companies’ A-share in Shanghai and H-share in Hong Kong. Because I need the market indices in the two market to compute normal return, I merge this dataset with the first dataset by date to include shanghai A-share and Hang Seng market index. The event window in the empirical is 41 days, 20 days before and after the launch date. The estimation window used for computing normal return is 90 days prior to the event window. I drop observations with event window less than 41 and estimation window less than 90, which makes the resulting dataset a sample of 56 dual-listed companies. By subtracting normal return from the actual return, I receive abnormal return and use these data to calculate the cumulative abnormal return (CAR) across the event window for each company in the sample.

The third dataset consists of CARs gained from the last dataset and variables reflecting firm characteristics. The primary data source for the second and third dataset is also the RESSET website. These variables include PE ratio, the date of establishment, state-owned shares, the total shares outstanding, industrial code and market value of these dual-listed companies. By dividing the total shares by state-owned shares, I get the state-share ratio. I calculate the age of a company by deducting the date of establishment from the launch date. By merging with the CARs from the second dataset, the final sample in this dataset consists of 56 observations.

Descriptive Statistics

Table 1 reports a summary of basic statistics of daily return for A-share and B-share market in Shanghai and Hong Kong stock market in three different periods. Period 1 is a 6-month period (from 10 Oct 2013 to 10 Apr 2014) before the announcement of the Shanghai-Hong Kong Stock Connect. Period 2 is from the announcement to one month before the launch of the program (from 10 Apr 2014 to 17 Oct 2014). Period 3 is from one month prior to the launch to the most recent trading day whose data are available on RESSET (from 17 Oct 2014 to 01 Apr 2015). Hence, period 1 represents pre-announcement period and period 2 for post-announcement but pre-launch period and period 3 for post-launch period. As can be clearly shown in Table1, the mean of daily returns for these three indices moved from negative in period 1 to positive in period 2. The growth

(20)

19

in the daily return of Hong Kong Hang Seng index is much higher than that of Shanghai A-share and B-share index. There is also a big jump from period 2 to period 3. The increase in the return of Shanghai A-share in this interim, however, is double than that of Shanghai B-share and Hong Kong Hang Seng index. More specifically, the average daily return of Shanghai A-share has quadrupled from 0.00111 in period 2 to 0.00484, while that of Shanghai B-share and Hang Seng index just doubled in that interim. I use a conventional decile sorting process to allow for a more detailed examination for the growth of return. Consistent with what I found in the mean of the returns, the growth in return of Shanghai A-share index in the top five deciles is higher than that of Shanghai B-share and Hong Kong Hang Seng index.

As with return, the volatilities of these three indices have all experienced an increase from period 2 to period 3, as can be shown from the standard deviation in Row 4. It is worth noting that the standard deviations of the three indices are roughly the same in the first two periods, whereas things have changed in period 3. The standard deviations of Shanghai B-share and Hong Kong Hang Seng market index remained close to each other and experience just a modest increase. In contrast, the figure for Shanghai A-share in period 3 is more than double that in period 2 and also that of contemporaneous Shanghai B-share and Hang Seng index. The distribution of the returns of Shanghai A-share and B-share has become more peaked than their respective distribution in the previous two periods, as can be shown in kurtosis, which approximate 7 in period 3 for both A-share and B-A-share in Shanghai. The figure for Hong Kong Hang Seng market index remain almost the same (somewhere around 3.2) throughout the three periods. The lower kurtosis indicates that its distribution is more flatter than that of Shanghai A and B share, which is consistent with the finding by Huang, Yang and Hu (2000) that Shanghai markets tend to have the greater kurtosis than Hong Kong stock markets .

Undeniably, the Stock Connect program does cause some changes in stock markets in Shanghai and Hong Kong. Such changes are also detectable from the AIC values in Table2. It can be clearly seen that the significant AIC values for the three indices are the same and remain unchanged in the first two periods, which is 2. However, appropriate orders of lag length based on their AIC values have all changed from 2 in period 2 to 1 in period 3. The change in AIC values is indicative of the change in each market itself due to this program.

(21)

20

Table 1. Descriptive Statistics for Daily Index Return

Table1 presents summary statistics for daily market returns of A-share and B-share index in Shanghai and Hong Kong Hang Seng market index in three different periods. Period 1 is a 6-month period (from 10 Oct 2013 to 10 Apr 2014) before the announcement of the Shanghai-Hong Kong Stock Connect. Period 2 is from the announcement to one month before the launch of the program (from 10 Apr 2014 to 17 Oct 2014). Period 3 is from one month prior to the launch until the most recent trading day whose data are available on RESSET (from 17 Oct 2014 to 01 Apr 2015). The index market return here is calculated according to the first difference of the logarithms of the index prices, that is, for each index of the three indices in the sample,R i, t = (ln pi, t – ln

pi, t-1) × 100. These returns are then sorted into ten equal parts. q1-10 represent the average return in each decile. Skewness reflects

the asymmetry of the probability distribution of the daily returns. Positive skew means the right tail of the distribution is longer and therefore right-tailed. In contrast, the negative suggests that the corresponding distribution is left-tailed. Kurtosis reflects the peakedness of the probability distribution of the daily index returns. The higher the value, the more peaked the distribution and the lower, the flatter.

period 1 period 2 period 3

from 10 Oct 2013 to 10 Apr 2014 from 10 Apr 2014 to 17 Oct 2014 from 17 Oct 2014 to 01 Apr 2015

Shanghai Hong Kong Hang Seng market index

Shanghai Hong Kong Hang Seng market

index A-share B-share A-share B-share A-share B-share

Mean -0.00023 -0.00081 -0.00002 0.00111 0.00092 0.00028 0.00484 0.00200 0.00046 Max 0.02839 0.01766 0.02690 0.02389 0.01696 0.02273 0.06388 0.02800 0.01927 Min -0.02901 -0.02992 -0.02265 -0.01838 -0.01938 -0.01954 -0.08032 -0.03986 -0.02619 Standard deviation 0.00976 0.00932 0.00925 0.00791 0.00688 0.00790 0.01980 0.00990 0.00892 Skewness 0.13401 -0.57431 -0.01657 0.19541 -0.17537 -0.05196 -0.81790 -0.98932 -0.28640 Kurtosis 3.66278 3.89406 3.23442 3.35104 3.01119 3.34042 6.93907 6.61296 3.20235 q 1 -0.01743 -0.01937 -0.01728 -0.01269 -0.01127 -0.01419 -0.03294 -0.01857 -0.01529 q 2 -0.00932 -0.00908 -0.00843 -0.00625 -0.00639 -0.00702 -0.01138 -0.00561 -0.00930 q 3 -0.00638 -0.00565 -0.00553 -0.00361 -0.00353 -0.00362 -0.00395 -0.00169 -0.00499 q 4 -0.00302 -0.00336 -0.00327 -0.00167 -0.00181 -0.00182 0.00042 0.00071 -0.00258 q 5 -0.00157 -0.00089 -0.00067 -0.00014 0.00025 -0.00039 0.00432 0.00202 -0.00025 q 6 0.00027 0.00054 0.00142 0.00154 0.00222 0.00097 0.00657 0.00359 0.00225 q 7 0.00281 0.00240 0.00320 0.00347 0.00393 0.00274 0.01159 0.00530 0.00411 q 8 0.00585 0.00527 0.00502 0.00665 0.00568 0.00488 0.01674 0.00728 0.00626 q 9 0.00871 0.00763 0.00874 0.00913 0.00766 0.00808 0.02203 0.01056 0.01029 q10 0.01802 0.01504 0.01671 0.01582 0.01262 0.01501 0.03878 0.01816 0.01576

(22)

21

Table 2. Akaike information criterion (AIC) of Daily Logarithmic Prices

Table 2 exhibits the appropriate lag lengths of Shanghai A-share and B-share market index and Hong Kong Hang Seng market index in three periods by providing AIC values for 10 lags.Period 1 is a 6-month period before the announcement of the Shanghai-Hong Kong Stock Connect. Period 2 is from the announcement to one month before the launch of the program. Period 3 is from one month prior to the launch until the most recent trading day whose data are available on RESSET.

Lag

period 1 period 2 period 3

from 10 Oct 2013 to 10 Apr 2014 from 10 Apr 2014 to 17 Oct 2014 from 17 Oct 2014 to 01 Apr 2015

Shanghai _{Hong Kong} Hang Seng market index

Shanghai _{Hong Kong} Hang Seng market index A-share B-share A-share B-share A-share B-share

0 -3.9442 -3.4825 -4.2939 -2.9068 -2.8209 -3.7502 -1.5833 -3.5701 -4.57706 1 -6.4322 -6.4990 -6.5466 -6.6196 -6.7901 -6.7062 -5.0122* -6.4128* -6.63232* 2 -6.4400* -6.5170* -6.5589* -6.6247* -6.8047* -6.7066* -4.9955 -6.3937 -6.59554 3 -6.4254 -6.5061 -6.5546 -6.6181 -6.7970 -6.7003 -4.9766 -6.3864 -6.57901 4 -6.4111 -6.5003 -6.5375 -6.6170 6.7954 -6.6937 -4.9547 -6.3766 -6.54388 5 -6.4121 -6.4831 -6.5460 -6.6202 -6.7924 -6.6947 -4.940 -6.3578 -6.51671 6 -6.4041 -6.4850 -6.5360 -6.6180 -6.7970 -6.6914 -4.9389 -6.3475 -6.50105 7 -6.3893 -6.4767 -6.5258 -6.6186 -6.7983 -6.6852 -4.917 -6.3259 -6.47025 8 -6.3896 -6.4713 -6.5104 -6.6152 -6.7977 -6.6777 -4.9028 -6.3039 -6.43629 9 -6.3772 -6.4659 -6.4943 -6.6084 -6.7967 -6.670 -4.8809 -6.3215 -6.39428 10 -6.3615 -6.4493 -6.4809 -6.6008 -6.7921 -6.663 -4.8636 -6.2995 -6.38205 *5% significance level

(23)

22

Preliminary analyses from the basic statistics seem to show that this program has a stronger impact on Shanghai A-share market than B-share and Hong Kong stock market. As such, more detailed empirical tests are necessary for the study on the different effects of this program on different markets.

An investigation at a more micro level can reveal how the average prices of A-share and its H- share counterpart behave under the influence of this program. From figure 1a, we can see the valuation gap between A-share and H-share begin to shrink after the announcement of the program. But when the two prices go closer and closer to the extent that they almost touch each other, they diverge from each other again. What drives them away is not the different moving direction (they actually move towards the same direction since the announcement) but the different growth rates of their prices. Another thing to note is that the A-share price has been higher than its H-share counterpart throughout the whole period, which is similar to the finding by Li, Yan and Greco (2005) who found that H-share had been traded at a discount relative to its corresponding A-share during the Asian financial crisis.

The valuation gap between A-share and H-share is most conspicuous after the launch date of the program (Figure 1a). Figure 1b and c provide supporting evidence for this phenomenon. Figure 1b displays the moving patterns of A-share and H-share for companies with positive A-and-H price difference and Figure 1c for those with negative A-and-H price difference. The moving paths of A-share and H-share, whether in Figure 1b or Figure 1c, are almost parallel during the period from the announcement to the launch of the program. But things have changed since the launch of Shanghai-Hong Kong Stock Connect. For companies whose A-share is traded at a premium compared to their H-share counterparts, their A-share moves upwards at a much higher growth rate than H-share and therefore widens their valuation gap. For those whose A-share is traded at a discount relative to their corresponding H-share, their A-share also increases at a higher rate and overtakes H-share short after the launch of the program, which causes their price difference to reduce first and then enlarge again.

Surprisingly, what has been shown in Figure 1 seems to be in contradiction with what we expected. The resulting reduction in trading restrictions due to this program is expected to help narrow the A-and-H price difference .The fact, however, is on the contrary. The A-and-H price difference tends to increase after the launch of the program, which can be explained by the different

(24)

23

Figure 1. Price Moving Patterns of A-share and H-share

Figure 1a,b,c show the moving pattterns of the average prices of A-share and H-share across 69 companies dual-listed in Shanghai and Hong Kong from 01 Jan 2014 to 01 Apr 2015. Prices here are not logarthimic prices but instead the actual prices that are quoted on their respective stock exchange. H-share price is denominated in Hong Kong dollar and A-share in RMB (currency in Mainland China). In order for them to be comparable, I first convert H-share into RMB denomination. Using the adjusted prices, I calculate the price difference in the A-share of a company and its H-share counterpart. I further group these dual-listed companies into two tpyes based on their price differences (postive and negative) at the announcement of the Stock Connect program. Of the 69 firms, 47 fall into the positive A-and-H price difference group and the rest into negative one. Figure 1a show the average price movement of all the 69 AH firms and Figure 1b,c for the positive and negative group, respectively. There are two vetical dash lines in each of the three figures. The left one represents the day of the annoucement of the Stock Connect program and the right one for the launch date of this program.

6 8 10 12 14 16 Ad ju st e d p ri ce

01jan2014 01apr2014 01jul2014 01oct2014 01jan2015 01apr2015 A-share H-share

Figure 1a

(25)

24 4 6 8 10 12 14 Ad ju st e d p ri ce

Figure 1b

Positive A-and-H Price Difference Firms

10 12 14 16 18 20 Aj ust ed p rice

Negative A-and-H Price Difference Firms

(26)

25

growth rates as manifested at Figure 1a, b, c. And the different growth rates suggest that the Stock Connect program may have different degrees of impact on Shanghai and Hong Kong stock markets. So a further study will be conducted to test whether the impacts of this program on the two markets are really different.

Empirical Results

Index-level Results

1. Unit root and Co-integration Tests

The unit root test in table 3 determines whether each time series of the market price of Shanghai A-share, B-share or Hong Kong Hang Seng index follows a stochastic trend. And co-integration test in table 4 further detects whether these time series have the same stochastic trend in common, that is, whether they are co-integrated with each other. The sample period is divided into three sub-periods, from 10 Oct 2013 to 10 Apr 2014, denoting the period before the announcement of the Shanghai-Hong Kong Stock Connect; from 10 Apr 2014 to 17 Oct 2014 denoting the period between the announcement and the launch of the program; from 17 Oct 2014 to 01 Apr 2015, denoting the period after the launch of the program. As mentioned earlier, the appropriate number of lag length for the three indices in the first two periods is 2, and 1 in the third period. So equation 12 is used in the Augmented Dickey Fuller (ADF) test period1 and 2 and equation 13 for test in period 3.

Δ Yit = β0 + αt +δ Yi,t-1 + γ1 Δ Yi,t-1 + γ2 Δ Yi,t-2+ Z i t (12) Δ Yit = β0 + αt +δ Yi,t-1 + γ1 Δ Yi,t-1 + Z i t (13) ΔY t represents the first difference of the logarithm of daily prices for each of the three indices at time t. The ADF test is essentially to test whether the null hypothesis H0: δ = 0, meaning the time series has a unit root and the alternative hypothesis is the series is stationary around a deterministic linear time trend. From table 3, the null hypotheses for all the three indices in all the three periods fail to be rejected. That is to say each of the three indices has a stochastic trend, which is consistent with large prior literature.

(27)

26

It can be seen from table 4 that except for the pair of Shanghai A-share and B-share in period 3, the three index pairs are all significantly co-integrated. To be more specific, the co-integration between Shanghai A-share and B-share becomes insignificant after the launch of the program, indicating the different level of impacts of this program on the two markets. Similarly, the co-integrating relation between Shanghai A-share and Hang Seng index is dampened from 1% significance level before the launch of the program to 5% after. The co-integrating relation between Shanghai B-share and Hong Kong Hang Seng index, however, becomes more significant after the program. The results can be explained by the finding from the summarized statistics that the program has a stronger impact on Shanghai A-share market than the other two markets, which weakens its co-integrating relation with either of the two markets. The reason for the stronger response of Shanghai A-share could be speculative fever of investors in Mainland China. Sun and Tong (2000) use the speculative argument to explain the price premium of A-share relative to its B-share counterpart and document that A-share prices can be pushed beyond their intrinsic value due to the speculative fever of unseasoned Chinese investors. Market participants in the stock markets of Mainland China are mainly retail investors (Fend and Seasholes (2004)). Retail investors typically do not have expertise to analyze companies’ performance and tend to trade based on market sentiment instead of fundamentals. The Shanghai-Hong Kong Stock Connect would undoubtedly be perceived as good news to stock markets and therefore leads to optimistic market sentiment which may in turn stimulate the speculative mania in Chinese investors. In contrast, market participants in Shanghai B-share are primarily seasoned institutional investors and established much earlier than stock markets in Mainland China, Hong Kong stock market is much more mature. As for the increased co-integrating relationship between Shanghai B-share and Hang Seng index, the further openness of Hong Kong stock market to investors in Mainland China increases similarities of these two markets and therefore the co-movement of their prices.

2. Granger Causality Test

Table 5 examines the return causality within each pair during the three periods. For index pairs exhibiting co-integrating relation in table 4, a vector error correction term (VECM) Granger Causality test is performed to test the bilateral return causality within themselves. Hence, except for Shanghai A-share and B-share in period 3, the model used is as follows:

(28)

27

ΔX t = β 20 + β 21 ΔY t-1 + β 22 ΔY t-2+ γ 21 ΔX t-1 + γ 22 ΔX t-2 +α 2 (Y t-1 - θ X t-1) + u 2t (15) ΔY t and ΔX t represent daily return of the two indices in each pair at time t. Y t-1 - θ X t-1 is the

error correction term in which θ is co-integrating coefficient gained from co-integrating test. α1 and α 2 are both short-run adjustment parameter. This paper focus mainly on the existence of return causality within the three pairs, which can be examined by testing the null hypothesis: γ11 = γ12= 0 for equation 14 and β11= β12 =0 for equation 15, each representing a one-way causal relation in the pair. When the null hypothesis γ11 = γ12= 0 is retained, index X does not Granger-cause index Y. If the null hypothesis β11 = β12 = 0 is also accepted, then there is not causality the other way around, that is, from index Y to index X. Because there is no co-integration between Shanghai A-share and B-share in period 3, the model used excludes the error correction term, that is,

ΔY t = β 10 + β11 ΔY t-1 + β12 ΔY t-2 + γ11 ΔX t-1 + γ12 ΔX + u 1t (16) ΔX t = β 20 + β 21 ΔY t-1 + β 22 ΔY t-2+ γ 21 ΔX t-1 + γ 22 ΔX t-2 + u 2t (17)

The null hypotheses and corresponding interpretations are similar to that in equation 14 and 15. It can be seen from table 5 there is no significant causal relationships, whether bilateral or unilateral, for all pairs in period 1, suggesting relatively weak interdependence among these stock markets before the announcement of the program. In period 2, there exists unilateral causality from Shanghai stock markets (both Shanghai A-share and B-share) to Hong Kong stock market. However, only causality from Shanghai A-share remains significant in period 3. In addition, there is significant bilateral causality within Shanghai stock exchange in period 3.

It is not groundless that the program creates unilateral causality from Shanghai stock markets to Hong Kong stock market. Studying the effect of four financial reforms in Mainland China on China’s stock market integration within itself and with world markets, Lin and Swanson (2008) found transmission from Mainland China to Hong Kong without reverse effects. They relate this to the political power of Mainland China over Hong Kong. Besides, the close economic relation between them can also be a contributing factor to this unilateral causality. Nearly half of stocks listed in Hong Kong are from Mainland China. So stock market in Hong Kong can be largely affected by Mainland China’s economy. Stock market performance can be viewed as a barometer for economic situation. Hence, unilateral return causality from Shanghai to Hong Kong after the program indicates that this program increases the economic influence from Mainland to Hong

(29)

28

Table 3. Unit Root Test

Table 3 presents results of unit root tests for Shanghai A-share, B-share and Hong Kong Hang Seng index in three different

periods. Period 1 is a 6-month period (from 10 Oct 2013 to 10 Apr 2014) before the announcement of the Shanghai-Hong Kong Stock Connect. Period 2 is from the announcement to one month before the launch of the program (from 10 Apr 2014 to 17 Oct 2014). Period 3 is from one month prior to the launch until the most recent trading day whose data are available on RESSET (from 17 Oct 2014 to 01 Apr 2015). I use the Augmented Dickey Fuller test with a deterministic trend term, which is Δ Yit = β0 + αt +δ

Yi,t-1 + γ1 Δ Yi,t-1 + γ2 Δ Yi,t-2 +…+γ p ΔY t-p + Z i t. ΔY t represents the first difference of the logarithm of daily prices for each of

the three indices at time t. According to the AIC value computed in Table 2, I set P equal to 2 in the first two periods and 1 for the third period. The null hypothesis is that index i (Shanghai A-share, B-share or Hang Seng index) has a unit root. The unit root test is essentially to test whether the residual term Z it has a unit root. If the t statistics of Z it falls into the acceptance interval, then the

null hypothesis is accepted, otherwise being rejected. Z (SHA), Z (SHB) and Z (HK) represent the residual term of Shanghai A-share, B-share and Hong Kong Hang Seng index, respectively. The critical values of each ADF Test at 1%, 5%, and 10% significance level are shown on the right-hand side of the corresponding t-statistic. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.

Residual Series t statistics

Critical Values of ADF Test for Unit Root

1% Critical Value 5% Critical Value 10% Critical Value

Period 1: from 10 Oct 2013 to 10 Apr 2014

Z(SHA) -2.106 -4.035 -3.448 -3.148

Z(SHB) -2.398 -4.035 -3.448 -3.148

Z(HK) -2.348 -4.035 -3.448 -3.148

Period2: from 10 Apr 2014 to 17 Oct 2014

Z(SHA) -2.912 -4.025 -3.444 -3.144

Z(SHB) -1.903 -4.025 -3.444 -3.144

Z(HK) -1.657 -4.025 -3.444 -3.144

Z(SHA) -1.779 -4.06 -3.459 -3.155

Z(SHB) -2.813 -4.06 -3.459 -3.155

Z(HK) -3.026 -4.06 -3.459 -3.155

(30)

29

Table 4. Co-integration test

Table 4 investigates the co-integration relations within three index pairs between Shanghai A-share, B-share and Hong Kong

Hang Seng index for three different periods. Period 1 is a 6-month period (from 10 Oct 2013 to 10 Apr 2014) before the announcement of the Shanghai-Hong Kong Stock Connect. Period 2 is from the announcement to one month before the launch of the program (from 10 Apr 2014 to 17 Oct 2014). Period 3 is from one month prior to the launch until the most recent trading day whose data are available on RESSET (from 17 Oct 2014 to 01 Apr 2015). The co-integration test is a two-step procedure. The regression model in the first step is Y t = α + θ X t + z t. The second step is a unit root test for residualZ t. The null hypothesis is

that the two indices in pair are not co-integrated, that is, their residual Z (i ,j ) has a unit root. If their residual Z (i ,j ) does not have a unit root, then the null hypothesis is rejected, i.e. they are co-integrated with each other.This tablepresents t statistics of the residual series and respective critical value. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.

Residual Series

Z ( i ,j ) t statistics

Critical Values of ADF Test for Unit Root

1% Critical Value 5% Critical Value 10% Critical Value

Z ( SHA, SHB ) -4.536 *** -2.58 -1.95 -1.62

Z ( SHA, HK ) -3.338 *** -2.58 -1.95 -1.62

Z ( SHB, HK ) -2.269 ** -2.58 -1.95 -1.62

Period2: from 10 Apr 2014 to 17 Oct 2014

Z ( SHA, SHB ) -3.475 *** -2.58 -1.95 -1.62

Z ( SHA, HK ) -3.27 *** -2.58 -1.95 -1.62

Z ( SHB, HK ) -2.18 ** -2.58 -1.95 -1.62

Z ( SHA, SHB ) -0.917 -2.58 -1.95 -1.62

Z ( SHA, HK ) -2.169 ** -2.58 -1.95 -1.62

Z ( SHB, HK ) -2.677 *** -2.58 -1.95 -1.62

(31)

30

Table 5. Granger Causality Test

Table 5 looks at the bilateral causality relationship within the paired indices for three different periods.Period 1 is a 6-month period (from 10 Oct 2013 to 10 Apr 2014) before the announcement of the Shanghai-Hong Kong Stock Connect. Period 2 is from the announcement to one month before the launch of the program (from 10 Apr 2014 to 17 Oct 2014). Period 3 is from one month prior to the launch until the most recent trading day whose data are available on RESSET (from 17 Oct 2014 to 01 Apr 2015). The null hypothesis is that the return of one index does not granger-cause that of the other in certain pair. Each row in the table displays one-way causality relation in each pair in the three periods. The first two rows show the bilateral causality relation between Shanghai A-share and Hong Kong Hang Seng index, the third and fourth rows for Shanghai A-share and B-share, and the last two rows for Shanghai B-share and Hong Kong Hang Seng index. This table provides F statistics and its corresponding p value for each causality test. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.

Direction of Causality Relation

Period 1 Period 2 Period 3

from 10 Oct 2013 to 10 Apr 2014 from 10 Apr 2014 to 17 Oct 2014 from 17 Oct 2014 to 01 Apr2015

F statistics p value F statistics p value t statistics p value

HKHS causes SHA 0.00 0.9964 0.01 0.9922 -0.43 0.6662 SHA causes HKHS 0.60 0.5502 3.69 ** 0.0278 -2.15 ** 0.0334 SHA causes SHB 1.55 0.2170 1.64 0.1982 -2.90 *** 0.0050 SHB causes SHA 0.29 0.7488 1.14 0.3218 2.01 ** 0.0470 SHB causes HKHS 0.08 0.9236 6.09 *** 0.0030 -0.77 0.4460 HKHS causes SHB 1.20 0.3043 0.04 0.9628 -1.18 0.2390

(32)

31

Kong. The bilateral causality between Shanghai A-share and B-share implies that market interdependence between Shanghai A-share and B-share has been increased following the program. It can be explained by the increased foreign investor base due to this program. Shanghai B-share are primarily traded by foreign investors. Prior to the program, Shanghai A-share can only be traded by a limit number of institutional investors. This program has removed trading restrictions on Shanghai A-share for foreign investors, which makes Shanghai A-share and B-share more similar to each other.

3. Chow Test

Chow test is conducted to test whether the Shanghai-Hong Kong Stock Connect has significantly changed the return causality in each index pair. The estimation specification used is ΔY t = α 0 + α 1 ΔXt-1 + α 2 ΔX t-2 + γ0 D + γ1 [ΔXt-1× D] +γ 2 [ΔX t-2 ×D] + u t (18) ΔY, ΔX represent the daily return of indices in each pair. The number of lags is selected according to Akaike information criterion (AIC) presented in table 2. D is a dummy variable that equals 1 for post-program period and 0 otherwise. If there is not a structural change caused by the program, then the regression function is the same before and after the program, so the dummy variable do not affect the regression. Hence, the null hypothesis of the Chow test is γ0 = γ1 = γ2 =0, which means there is no break. If the null hypothesis is rejected, then structural change occurs.

Table 6 reports the results of chow tests. From p value in the table, the causality that Hong Kong stock market is affected either by Shanghai A-share or B-share have not significantly changed after the program. On the other hand, the return causality relationships between Hong Kong stock market to Shanghai stock markets, whether A-share or B-share, have changed at 5% significance level. Similarly, this program gave rise to structural changes in causality relation within Shanghai stock exchange. It can be seen from table 6 that the unilateral causality relation between Shanghai A-share and B-share has changed significantly. It can be deduced from the above evidence that Hong Kong stock market is more efficient than Shanghai stock market based on market efficiency hypothesis.

(33)

32

Table 6. Chow test for structural change in causality relation within each index pair

Table 6 explores whether there exists a structural break in the bilateral causality relationship within the three index pairs. The estimation model used is ΔY t = α 0 + α 1 ΔXt-1 + α 2 ΔX t-2 + γ0 D + γ1 [ΔXt-1× D] +γ 2 [ΔX t-2 ×D] + u t (the number of lags is selected

according to Akaike information criterion (AIC)). . D is a dummy variable that equals 1 for post-program period and 0 otherwise. ΔY, ΔX represent the daily return of indices in each pair. For example, in column 1, ΔY, ΔX are assigned to the daily returns of SHA and HKHS to test for structural change in causality from HKHS to SHA. SHA, SHB, and HKHS represent Shanghai A-share, Shanghai A-share, and Hong Kong Hang Seng index, respectively. The null hypothesis is γ0 = γ1 =γ 2=0 for the non-existence of a

structural break. The p value for the null hypothesis is displayed in the lowest row. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.

Dependent variable: daily return of SHA (Shanghai A-share)

Dependent variable: daily return of HKHS (Hong Kong Hang Seng index)

Dependent variable: daily return of SHB (Shanghai B-share) (1) (2) (3) (4) (5) (6) test for structural change in causality from HKHS to SHA test for structural change in causality from SHB to SHA test for structural change in causality from SHA to HKHS test for structural change in causality from SHB to HKHS test for structural change in causality from SHA to SHB test for structural change in causality from HKHS to SHB constant 0.00030 0.00030 -0.00007 -0.00006 0.00006 0.00011 (0.00081) (0.00081) (0.0005549) (0.00055) (0.00054) (0.00055) Daily Return of SHA (t-1) 0.05353 0.16383 ***

(0.0634321) (0.06255)

Daily Return of SHA (t-2) -0.01321 -0.01650

(0.063481) (0.06260) Daily Return of SHB (t-1) 0.05168 -0.06934 (0.10021) (0.06872) Daily Return of SHB (t-2) 0.00804 0.02540 (0.10033) (0.06881) Daily Return of HKHS (t-1) 0.03945 0.11502 * (0.09547) ( 0.06486) Daily Return of HKHS (t-2) -0.00582 -0.02215 (0.09547) (0.06485) Dummy (Post-program) 0.00428 *** 0.00370** 0.00111 0.00100 0.00218 ** 0.00173 * (0.00144) (0.00146) (0.00102) (0.00101) (0.00101) (0.00098) Dummy*Daily Return of SHA (t-1) -0.12190 -0.28764 ***

(0.07777) (0.07670) Dummy*Daily Return of SHA (t-2) 0.04601 -0.00755 (0.07786) (0.07679) Dummy*Daily Return of SHB (t-1) 0.11775 0.02388 (0.16183) (0.11098) Dummy*Daily Return of SHB (t-2) 0.07524 -0.01697 （0.16178） (0.11094) Dummy*Daily Return of HKHS (t-1) -0.18175 -0.25645 ** (0.16857) (0.11453) Dummy*Daily Return of HKHS (t-2) -0.14101 -0.15312 (0.16878) (0.11468)

p value of structural change 0.0187 ** 0.0479** 0.3011 0.7783 0.0009*** 0.0242**

The effect of financial deregulation on stock markets interdependence and efficiency : evidence from Shanghai-Hong Kong Stock Connect

MSc Business Economics, Finance track

Master Thesis