Mối quan hệ giữa vàng, dầu thô, thị trường chứng khoán Mỹ và Đài Loan

Nội dung text: Mối quan hệ giữa vàng, dầu thô, thị trường chứng khoán Mỹ và Đài Loan

1. THE RELATIONSHIP AMONG GOLD, CRUDE OIL, THE US AND TAIWAN’S STOCK MARKET MỐI QUAN HỆ GIỮA VÀNG, DẦU THÔ, THỊ TRƯỜNG CHỨNG KHOÁN MỸ VÀ ĐÀI LOAN Ming-Kun Lin Hsin-Fu Chen Keng-Shen Chen Lunghwa University of Science Technology Taoyuan, Taiwan Abstract This paper tries to explore the relationship between Taiwan stock market and commodity market and American stock market. There are six variables with total 213 observations for each variable by using monthly data from the periods of October 1997 to June 2015. Model 1 examines commodity market including gold price（LLG）, crude oil price（WTI）, USD index （USDX）and Taiwan stock market. Model 2 adds American stock market, Dow Jones Industrial Average（DJIA）and NASDQA. The results on Ordinary Least Squares（OLS）showed that USDX has significant negative impact on Taiwan stock market in model 1. However, in Model 2 showed WTI and USDX have significant negative impact but NASDAQ has positive significant impact on Taiwan stock market. The results of Unit Root test demonstrated that all variables are not stationary series in its original numbers, but they become I(1) stationary series after First Difference. In addition, the results of Johansen Cointegration showed that there are no cointegration relationship on both models, but there is on-way leading relationship for Taiwan stock market on WTI, DJIA and NASDAQ from Granger Causality test. From Impulse Response Analysis, the results showed that Taiwan stock market has negative impulse response on USDX and WTI while it has positive impulse response on LLG, DJIA and NASDAQ, and the impulse lasted for 4 periods. Finally, the results of Forecast Error Variance Decomposition showed there are high self-explanation powers for all variables on both models and NASDAQ has the most influence from DJIA and Taiwan stock market in model 2. Keywords: Gold Price, Crude Oil Price, USD Index, Dow Industrial, NASDQ, Taiwan Stock Market Tóm tắt Nghiên cứu tìm hiểu mối quan hệ giữa thị trường hàng hoá và thị trường chứng khoán Đài Loan và thị trường chứng khoán Mỹ. Có 6 biến với tổng số 213 quan sát cho mỗi biến được thực hiện bằng cách sử dụng dữ liệu hàng tháng trong giai đoạn từ tháng 10 năm 1997 đến tháng 6 năm 2015. Mô hình 1 xem xét thị trường hàng hoá, gồm giá vàng (LLG), giá dầu thô (WTI), chỉ số đồng đô la Mỹ (USDX) và thị trường chứng khoán Đài Loan. Mô hình 2 có thêm thị trường chứng khoán Mỹ, chỉ số công nghiệp Dow Jones (DJIA) và NASDQA Kết quả phương pháp bình phương nhỏ nhất (OLS) cho thấy chỉ số USDX có tác động tỉ lệ nghịch đối với thị trường chứng khoán Đài Loan trong mô hình 1. Tuy nhiên, mô 243
2. hình 2 cho thấy chỉ số WTI và USDX có tác động tỉ lệ nghịch nhưng chỉ số NASDAQ có tác động tỉ lệ thuận tới thị trường chứng khoán Đài Loan. Kết quả kiểm định nghiệm đơn vị cho thấy rằng tất cả các biến đều không phải chuỗi dừng trong giá trị ban đầu, tuy nhiên đã trở thành chuỗi dừng I (1) sau vi phân bậc 1. Bên cạnh đó, kết quả kiểm định đồng liên kết Johansen cũng cho thấy rằng không có quan hệ đồng liên kết trong cả hai mô hình, nhưng có quan hệ định hướng đơn chiều đối với thị trường chứng khoán Đài Loan của chỉ số WTI, DJIA và NASDAQ từ kiểm định nhân quả Granger. Từ phân tích phản ứng xung, kết quả cho thấy thị trường chứng khoán Đài Loan có phản ứng xung tỉ lệ nghịch với USDX và WTI trong khi có phản ứng xung tỉ lệ thuận với LLG, DJIA và NASDAQ và phản ứng xung kéo dài 4 chu kỳ. Cuối cùng, kết quả phân tách phương sai sai số dự báo cho thấy có tác động tự giải thích đối với tất cả các biến số của cả hai mô hình và chỉ số NASDAQ có tác động lớn nhất từ DJIA và thị trường chứng khoán Đài Loan trong mô hình 2. Từ khoá: giá vàng, giá dầu thô, chỉ số USD, chỉ số công nghiệp Dow, NASDAQ, thị trường chứng khoán Đài Loan 1. Introduction Stock market is one of the most important economic and financial indexes for a country’s economics development. Taiwan’s stock market has been established since the year 1962 and there are total 1496 companies, 838 in list market and 658 in Over the Counter (OTC) market by the end of 2015. Gold price and crude oil price have been increased sharply since 2002, and these commodities are trading in US dollars. Therefore, this paper is setting up the first model to examine the impact of goods market, gold price (LLG) and crude oil price (WTI) with US dollar index (USDX) on Taiwan’s stock market. In addition, Taiwan is a trade-oriented country, most products are mainly for export and import and the US market is one of the major trade markets in Taiwan. Thus, in model 2, it added American stock market, Dow Jones Industrial Average (DJIA) and NASDQA into model 1, to explore the relationship between Taiwan and American stock markets. The main purpose is to explore the relationship between gold price, crude oil price, US dollar index, American stock markets and Taiwan stock market. There are six variables with total 213 observations for each variable in monthly data from October 1997 to June 2015. It tries to find out a long term relationship between Taiwan’s stock market and model 1 and model 2 by using cointegration. Further, to observe the causality among variables in Granger causality test. Finally, impulse response analysis and forecast error variance decomposition test the impact of impulse response among variables on other variables and their explanation power. The results on Ordinary Least Squares（OLS）showed that USDX has significant negative impact on Taiwan stock market in model 1. However, in Model 2 showed WTI and USDX have significant negative impact but NASDAQ has positive significant impact on Taiwan stock market. The results of Unit Root test demonstrated that all variables are 244
3. not stationary series in its original numbers, but they become I(1) stationary series after First Difference. In addition, the results of Johansen Cointegration showed that there are no cointegration relationship on both models, but there is on-way leading relationship for Taiwan stock market on WTI, DJIA and NASDAQ from Granger Causality test. From Impulse Response Analysis, the results showed that Taiwan stock market has negative impulse response on USDX and WTI while it has positive impulse response on LLG, DJIA and NASDAQ, and the impulse lasted for 4 periods. Finally, the results of Forecast Error Variance Decomposition showed there are high self-explanation powers for all variables on both models and NASDAQ has the most influence from DJIA and Taiwan stock market in model 2. 2. Literature Review Hsih (2006) used monthly data between September 1990 to January 2006 to explore the relationship among gold spot and future prices, and crude oil spot and future prices and Taiwan stock market. He found that Taiwan stock market and gold spot price has random walk trend in unit root test. In addition, gold and crude oil prices have long term stable relationship with Taiwan stock market in cointegration test. Gold price has one-way leading relationship in Taiwan stock market in granger causality test. Finally, there has a significant positive relationship between Taiwan stock market and gold price in impulse response analysis test. Chen S. (2011) used daily data between January 1992 to December 2010 to explore the relationship between gold price and Taiwan stock market. She found that gold price has negative significant impact on Taiwan stock market in OLS regression. Chang (2006) used monthly data between January 1995 to January 2005 to explore the relationship between high crude oil price and Taiwan stock market. She found out that there are neither leading relationships between WTI and Taiwan stock market in Vector Autoregression (VAR) test, nor causality in grander causality test. Finally, there is a short term impact of WTI on Taiwan stock market in impulse response analysis test. Huang H. (2008) used monthly data between January 1991 to December 2008 to explore the relationship among WTI, Shanghai stock market DJIA and Taiwan stock market. She found that WTI and Taiwan stock market become stationary series after first difference I(1) in unit root test. In addition, there have a long term stable equilibrium relationship among Shanghai stock market, DJIA, WTI and Taiwan stock market in cointegration test. In addition, there is a feedback relationship between WTI and Taiwan stock market in granger causality test. Finally, there has the same direction movement between Taiwan stock market and WTI in OLS test. Chen C. (2009) used monthly data between October 1983 to October 2008 to explore the relationship among WTI, new Taiwan Dollar/US Dollar exchange rate and Taiwan stock market. Period 1 includes October 1983 to August 1988, and period 2, August 1988 to September 2001 and period 3, September 2001 to October 2008. She found that WTI and Taiwan stock market has random walk trend and become stationary series after first difference I(1) in unit root test. WTI, Exchange rate and Taiwan stock 245
4. market have long term stable relationship in cointegration test in period 1 and 3, but not in period 2. WTI has impact on Taiwan stock market in period 3, in granger causality test. Finally, there has no impact of WTI and exchange rate on Taiwan stock market in period 1, but WTI has slight impact on Taiwan stock market in period 2 and 3 in impulse response analysis test. Huang Z. (2009) used daily data between January 2, 2006 to February 27, 2009 to explore the relationship among WTI, LLG, DJIA, German, Japan, Taiwan and Shanghai stock markets. She found that WTI, DJIA and Taiwan stock market has random walk trend and become stationary series after first difference I(1) in ADF unit root test. WTI, LLG, US dollar exchange rate and Taiwan stock market have long term stable equilibrium relationship in Johansen cointegration test. WTI and US dollar exchange rate have feedback effect on Taiwan stock market in VAR test. WTI has also feedback effect on Taiwan stock market, but US dollar exchange rate has one-way leading on Taiwan stock market in granger causality test. Shi P. (2011) used monthly data between May 1994 to November 2010 to explore the relationship among NASDAQ, Philadelphia semi-conductor index (PSCX) and Taiwan stock market. He found that NASDAQ, PSCX and Taiwan stock market have random walk trend and become stationary series after first difference I(1) in ADF unit root test. NASDAQ, PSCX and Taiwan stock market have no long term stable relationship in cointegration test. NASDAQ has most significant impact on Taiwan stock market in VAR test. NASDAQ has more impact on Taiwan’s stock market in granger causality test. Finally, there are positive relationship among NASDAQ, PSCX and Taiwan stock market in correlation teat. Lee K. (2012) used daily data between January 4, 2000 to December 28, 2011 to explore the NT$/US$ exchange rate (US dollar exchange rate), WTI and LLG. He found that US dollar exchange rate, WTI and LLG have random walk trend and become stationary series after first difference I(1) in ADF unit root test. There is no stable equilibrium relationship among Exchange rate, WTI, and LLG, in Johansen cointegration test. The impact orders are LLG less impact than WTI and US dollar exchange rate. US dollar exchange rate has most sensitive impact in VAR test. WTI and US dollar exchange rate have feedback effect. LLG has one-way leading relationship on US dollar exchange rate in granger causality test. Tien S. (2013) used monthly data between December 1999 to October 2012 to explore Mining Fund (MF), LLG and US dollar index (USDX). She found that MF, LLG and USDX have stable long term equilibrium relationship in Johansen cointegration test. LLG has one-way leading relationship with USDX in Granger causality test. 3. Research Methodology: This paper starts from Ordinary Least Square (OLS) and it is important to know that used data series are stationary or not in the empirical studies in time series models. If it is a nonstationary time series data to do regression will cause the estimate model to be 246
5. biased. Granger & Newbold（1974）94 called it is a false regression. Therefore, it uses ADF unit root test to test data series stationary, Johansen cointegration, Granger causality, impulse response, and forecast error variance decomposition. In addition, it is also important to decide the fitness of lag period by using Akaike Information Criterion (AIC) before running above tests. (1). Ordinary Least Squares In regression analysis, if there is only forecasting variable, it is called simple regression and if there are more than one, it is called multi-regression model. The models in this paper are multi-regression model, formula (3.1). (3.1) Y is dependent variable, Xs are independent variables, βs are coefficients, i s are different period of sample data number, and is error term (2). Akaike Information Criterion (AIC) There are two ways of judging the best lag period, Akaike Information Criterion (AIC）and Schwartz Bayesian Information Criterion (SIC). SIC was Schwartz（1978）95 induced from Bayesian Criterion. AIC was developed by Akaike（1973）96 from the concept of Maximum Likelihood. A. AIC Akaike（ 1974） uses formula (3.2) to choose the best lag period by using the minimum value of AIC. (3.2) P: lag period, T: valid sample numbers, : residuals of variance（maximum likelihood value） B. SBC Schwarz （1978） used Bayesian Criterion to set up SIC （Schwartz’s Bayesian Criterion）. (3.3) This paper uses Akaike（1974）AIC rules to choose the best lag period for the models (3). ADF Unit Root test Dicker and Fuller（1979）97 used unit root test to examine the stationary of data series. If the results with unit root, it implies that the data series are nonstationary. There are three different models of ADF unit root as the followings: 94 Granger, CWJ and Newbold, P. (1974). "Spurious regressions in econometrics". 95 Schwarz, Gideon E. (1978). "Estimating The Dimension of A Model". 96 Akaike, Hirotugu (1974). "A New Look at The Statistical Model Identification". 97 Dickey, DA; and Fuller, WA (1979). "Distribution of the Estimators for Autoregressive Time Series with a Unit Root". 247
6. Model 1：no intercept, no trend (3.4) Model 2：with intercept but no trend (3.5) Model 3：with intercept and trend (3.6) : first order differentiate, : intercept, : time trend, : error term, : coefficient value, and : lag period ADF unit root hypothesis： ： （with unit root，data series are not stationary） ： （without unit root, data series are stationary） According to the results, if it rejects , without unit root, it implies that original data series are stationary. If the results accept , with unit root, it implies that original data series are nonstationary and it needs to do second order differentiate until it rejects . (4). Johansen Cointegration Granger（1981）98 cointegration concept states that two nonstationary variables will become stationary after linear cointegration and it implies that they will have long- term stable relationship. Engle and Granger（1987）99 cointegration theory. It implies that if nonstationary data series become stationary after linear programming. The main purpose is to understand if there is a long-term stable relationship among data series. There are two different cointegration theories including, Engle and Granger（1987）and Johansen（1988）100, and Johansen and Juselius（1990）101, maximum likelihood cointegration. In this paper adopts maximum likelihood cointegration test. Johansen cointegration is not only to estimate all cointegration vectors but also it can complete understand long-term or short-term relationship in time series data. In addition, it can use normal distribution statistics to examine cointegration vector number（r）improved insufficient of Engle-Granger two stages 98 Granger, Clive (1981). "Some Properties of Time Series Data and Their Use in Econometric Model Specification". 99 Engle, Robert F. , Granger, Clive WJ (1987). "Co-integration and error correction: Representation, estimation and testing". 100 S. Johansen (1988). "Statistical Analysis of Cointegration Vectors". 101 S. Johansen, K. Juselius (1990) "Maximum Likelihood Estimation And Inference On Cointegration — With Applications To The Demand For Money". 248
7. cointegration. Johansen cointegration sets up under VAR system and assume a lag of k rank and k numbers of vector, its VAR model equation is as (3.7)： (3.7) : lag of k rank nonstationary variable, integration , : coefficient matrix, : constant vector, : error term Johansen and Juselius（1990）used maximum likelihood cointegration test to setup Trace Test and Maximum Eigenvalue Test, two LR （ distribution statistics to examine cointegration vector number（r）： Trace Test (3.8) （maximum r cointegration） （minimum r+1cointegration） Maximum Eigenvalue Test (3.9) （with r numbers of cointegration vectors） （with r+1numbers of cointegration vectors） T: numbers of observatories, n: series numbers, : i’s coefficient value, : all implies long-term information, and r: cointegration vector numbers Refers to the critical values of Johansen and Juselius（ 1990） to decide reject or accept . (5). Vector Autoregression (VAR） Sims（1980）102 used Vector Autoregression (VAR） to solve internal and external problem. VAR uses its own variables with one lag period as an explanation variable and see them as internal variables for all variables. It will be very clear to know how a variable change to affect another variable. General equation (3.10)： (3.10) ； : internal variable vector, : i’s one lag period vector of vector, : coefficient matrix, : expected error of vector, Σ: covariance matrix , implies time series independent for each regression , error term of current period 102 Sims, Christopher (January 1980) "Macroeconomics and reality". 249
8. (6). Granger Causality Granger（1969）103used predictability to measure the causality among variables. If there is a causality relationship, then, adds previous information of an independent variable to increase the explanation ability of dependent variable. If the previous information of variable X helps to predict dependent variable Y. Then, it can state as X variable Granger affect variable Y. In addition, if X and Y variables have mutual Granger effects, they are feedback effect between two variables. Therefore, there are four results for Granger causality to explain the relationship between variables, Granger cause, does not Granger cause, feedback and independent Granger causality models(3.11) and (3.12)： (3.11) (3.12) 、 : two irrelevant error terms. Four coefficients to determine the relationships between variables： If and ，implies Y causes X (Y Causes X)； If and ，implies X causes Y (X Causes Y)； If and ，implies there is feedback between two variables. If and ，implies there is independent between two variables. (7). Impulse Response Analysis Sims（1980）used impulse response analysis to study an unexpected change of a variable impacted by external shocks in VAR model. It implies a dynamic reaction model when a variable was impacted by external shocks. From impulse response analysis, it can know the positive or negative impact with persistence or volatility impact and periods. From Wald Decomposition Theorem transfers vector autogression (VAR) model to moving average (MA), (3.13) L: Lag Operator, : constant vector, : matrix, （unit matrix）, : vector forecast error. 103 Granger, C. W. J. (1969). "Investigating Causal Relations by Econometric Models and Cross-spectral Methods" 250
9. If has no relationship with current period, it can decide the relationship between two variables by calculating forecast error variance percentage. If has certain relationship with current period, it has to use Cholesky decomposition to choose a lower triangular matrix： (3.14) F: Non-singular, ， ， To write above equations (3.14) to: (3.15) : random term with no autocorrelation (8). Forecast Error Variance Decomposition In general empirical study, if there are too many variables in VAR model, there may exist collinear, over distribution and parameters problems to affect the results of estimation in regression model. t period forecast error equation (3.16): (3.16) : error term by using previous data of t- k period to forecast t period. It can be used coefficient D matrix of moving average (MA) in VAR model applying decomposition on covariance of k rank forecast error in each variable. Covariance matrix of k rank forecast error: (3.17) 4. Empirical Results: This paper used E-vies software to run regression. The data sample period starts from October 1997 to June 2015. There are total 213 observations for each variable by using monthly data into empirical studies. The data sources are from Taiwan Economic Journal database system (TEJ). Model 1： Model 1 represents the relationship between Taiwan stock market and goods market including WTI, LLG, and USDX. Model 2： 251
10. Model 2 represents the relationship among Taiwan stock market, goods market including WTI, LLG, and USDX and US stock markets, DJIA and NASDAQ. Table 1. Variables Explanation and Symbol Variables Variable Name Symbols Taiwan Stock Market Taiwan Capitalization Weighted Stock Index TAIEX Crude Oil Price West Texas Intermediate WTI Gold Price Loco London Gold Price LLG US Dollars Index US Dollars Index USDX Dow Jones Industrial Average Dow Jones Industrial Average DJIA National Association of Securities Dealers NASDAQ NASDAQ Automated Quotations system 4.1 OLS Results Model 1 Model 2 Coefficient t-Statistic Coefficient t-Statistic C 10.73138 10.18610 10.99142 12.28049 WTI -0.034910 -0.798655 -0.239833 -5.522188 LLG 0.073901 1.415313 0.034399 0.864964 USDX -0.499887 -2.648217 -1.241004 -7.844855 DJIA - - 0.013701 0.122880 NASDAQ - - 0.518478 7.825477 R2 0.164048 0.527588 Adjusted R2 0.152049 0.516177 D-W Statistic 0.139557 0.233087 F-Statistic 13.67149 46.23536 註： 、represents 1% significant level。 4.2 ADF Unit Root Test Results: Intercept, no trend Intercept and trend No intercept and trend TAIEX -3.296666（2） -3.348203（2） 0.219230（7） WTI -1.241910（1） -3.490660（2） 0.821409（1） LLG 0.148645（0） -2.322166（0） 1.757081（0） USDX -1.222155（2） -2.465111（2） -0.165626（2） DJIA -2.439977（6） -2.871435（6） 1.369473（6） NASDAQ -2.297691（1） -2.473811（1） 0.865147（1） Notes 1.（ ）lag period 2. logarithm statistics 3. represents 1% significant level 4.ADF t statistics to choose the best lag period. 252
11. 4.2.1 Differentiate ： Intercept, no trend Intercept and trend No intercept and trend TAIEX -5.608356（6） -5.590028（6） -5.615613（6） WTI -11.15097（0） -11.12423（0） -11.09896（0） LLG -10.99745（1） -11.11920（1） -13.88130（0） USDX -9.767361（1） -9.793694（1） -9.790116（1） DJIA -5.791962（5） -5.807383（5） -5.599625（5） NASDAQ -10.67277（0） -10.65349（0） -10.63177（0） Notes 1.（ ）lag period 2. logarithm statistics 3. represents 1% significant level 4.ADF t statistics to choose the best lag period. 4.3 Johansen Cointegration Test Results: Model 1：Johansen Cointegration Trace Test Null Hypothesis Eigenvalue Trace Test Value 5% P Value None 0.093776 44.33554 54.07904 0.2746 1 at most 0.062665 23.65708 35.19275 0.4850 2 at most 0.035770 10.06701 20.26184 0.6333 3 at most 0.011446 2.417589 9.164546 0.6940 Maximal Eigenvalue Test Null Hypothesis Eigenvalue Trace Test Value 5% P Value None 0.093776 20.67847 28.58808 0.3620 1 at most 0.062665 13.59007 22.29962 0.5001 2 at most 0.035770 7.649418 15.89210 0.5900 3 at most 0.011446 2.417589 9.164546 0.6940 註：* represents 5% significant level ，reject null hypothesis. Model 2：Johansen Cointegration Trace Test Null Hypothesis Eigenvalue Trace Test Value 5% P Value None 0.109894 85.99390 103.8473 0.4110 1 at most 0.091318 61.54686 76.97277 0.4131 2 at most 0.080432 41.43726 54.07904 0.4001 3 at most 0.047572 23.82852 35.19275 0.4738 4 at most 0.042524 13.59301 20.26184 0.3184 5 at most 0.021049 4.467454 9.164546 0.3469 Maximal Eigenvalue Test Null Hypothesis Eigenvalue Trace Test Value 5% P Value 253
12. None 0.109894 24.44704 40.95680 0.8460 1 at most 0.091318 20.10960 34.80587 0.8065 2 at most 0.080432 17.60874 28.58808 0.6091 3 at most 0.047572 10.23551 22.29962 0.8164 4 at most 0.042524 9.125551 15.89210 0.4205 5 at most 0.021049 4.467454 9.164546 0.3469 Notes：* represents 5% significant level ，reject null hypothesis. 4.4 AIC Results: Model 1 Model 2 Lag period AIC value AIC value 0 -0.490260 -3.005872 1 -12.97774 -20.46675 2 -13.08965* -20.63554* 3 -13.02778 -20.56575 4 -12.95458 -20.47278 5 -12.89462 -20.38183 6 -12.80610 -20.25834 7 -12.73125 -20.24571 8 -12.71282 -20.20535 Notes：* represents AIC minimum value 4.5 Granger Causality Results Model 1：Granger Causality Null Hypothesis Chi-sq Prob. WTI Granger TAIEX 1.3675 0.5047 LLG Granger TAIEX 0.2243 0.8939 USDX Granger TAIEX 0.2312 0.8908 TAIEX Granger WTI 14.305 0.0008 LLG Granger WTI 1.4214 0.4913 USDX Granger WTI 0.1325 0.9359 TAIEX Granger LLG 0.0528 0.9740 WTI Granger LLG 2.0893 0.3518 USDX Granger LLG 1.2299 0.5407 TAIEX Granger USDX 0.2905 0.8648 WTI Granger USDX 1.0786 0.5832 LLG Granger USDX 4.0788 0.1301 Notes: *、 、 represents 10%, 5% and 1% significant level ，reject null hypothesis. 254
13. Model 2: Granger Causality Null Hypothesis Chi-sq Prob. WTI Granger TAIEX 1.4288 0.4895 LLG Granger TAIEX 0.2566 0.8796 USDX Granger TAIEX 0.1824 0.9128 DJIA Granger TAIEX 0.2891 0.8655 NASDAQ Granger TAIEX 0.2534 0.8810 TAIEX Granger DJIA 11.146 0.0038 WTI Granger DJIA 1.7384 0.4193 LLG Granger DJIA 1.9521 0.3768 USDX Granger DJIA 0.7568 0.6849 NASDAQ Granger DJIA 0.8903 0.6407 TAIEX Granger WTI 10.635 0.0049 LLG Granger WTI 1.4887 0.4751 USDX Granger WTI 0.3235 0.8506 DJIA Granger WTI 4.7475 0.0931* NASDAQ Granger WTI 0.9634 0.6177 TAIEX Granger LLG 0.5261 0.7687 WTI Granger LLG 1.5583 0.4588 USDX Granger LLG 1.2294 0.5408 DJIA Granger LLG 0.9213 0.6309 NASDAQ Granger LLG 0.1045 0.9491 TAIEX Granger USDX 0.2536 0.8809 WTI Granger USDX 1.1077 0.5747 LLG Granger USDX 4.5629 0.1021 DJIA Granger USDX 1.8307 0.4004 NASDAQ Granger USDX 8.8326 0.0121 TAIEX Granger NASDAQ 10.951 0.0042 WTI Granger NASDAQ 0.7253 0.6958 LLG Granger NASDAQ 1.0546 0.5902 USDX Granger NASDAQ 0.0356 0.9824 DJIA Granger NASDAQ 6.0653 0.0482 Notes: *、 、 represents 10%, 5% and 1% significant level ，reject null hypothesis. 4.6 Impulse Response Analysis Model 1： .08 .06 .04 .02 .00 -.02 1 2 3 4 5 6 7 8 9 10 TAIEX WT I USDX LLG Figure 1. Model 1 of TAIEX Impulse Response 255
14. Model 2： .08 .06 .04 .02 .00 -.02 1 2 3 4 5 6 7 8 9 10 TAIEX WTI DJIA LLG NASDAQ USDX Figure 2. Model 2 of TAIEX Impulse Response 4.7 Forecast Error Variance Decomposition Results: Model 1：TAIEX Forecast Error Variance Decomposition Period TAIEX WTI USDX LLG 1 100.0000 0.000000 0.000000 0.000000 2 99.89784 0.015453 0.085048 0.001661 3 99.19198 0.465495 0.210996 0.131530 4 99.15200 0.496658 0.218988 0.132349 5 99.10574 0.541581 0.220373 0.132310 TAIEX 6 99.10345 0.543680 0.220373 0.132499 7 99.10257 0.544551 0.220377 0.132498 8 99.10255 0.544551 0.220388 0.132509 9 99.10255 0.544553 0.220389 0.132509 10 99.10255 0.544555 0.220389 0.132510 Model 2：TAIEX Forecast Error Variance Decomposition Period TAIEX WTI DJIA LLG NASDAQ USDX 1 100.0000 0.000000 0.000000 0.000000 0.000000 0.000000 2 99.84574 0.005761 0.084915 0.023475 0.001566 0.038541 3 99.07418 0.402274 0.097901 0.221296 0.110614 0.093735 4 99.01941 0.422015 0.106522 0.225253 0.129041 0.097763 5 98.91519 0.471237 0.154777 0.225102 0.129729 0.103963 TAIEX 6 98.90815 0.474601 0.156564 0.225064 0.130785 0.104839 7 98.90322 0.476762 0.158725 0.225325 0.130812 0.105155 8 98.90288 0.476763 0.158998 0.225394 0.130811 0.105155 9 98.90278 0.476779 0.159063 0.225408 0.130816 0.105156 10 98.90277 0.476781 0.159066 0.225413 0.130816 0.105157 256
15. 5. Conclusion This paper tried to explore the relationship among commodity market (gold price, crude oil price and US dollar index), American stock Market (DJIA and NASDQA) and Taiwan stock market. It also uses two models respectively to examine the relationship between commodity market and American stock market. Model 1 examines commodity market including gold price, crude oil price, USD index and Taiwan stock market. Model 2 adds American stock market, Dow Jones Industrial Average（DJIA）and NASDQA into model 1. There are six variables with total 213 observations for each variable using monthly data from the period of October 19957to June 2015. The results on Ordinary Least Squares（OLS）showed that USDX has significant negative impact on Taiwan stock market in model 1. However, in Model 2 showed WTI and USDX have significant negative impact but NASDAQ has positive significant impact on Taiwan stock market. The results of Unit Root test demonstrated that all variables are not stationary series in its original numbers, but they become I(1) stationary series after First Difference. In addition, the results of Johansen Cointegration showed that there are no cointegration relationship on both models, but there is on-way leading relationship for Taiwan stock market on WTI, DJIA and NASDAQ from Granger Causality test. From Impulse Response Analysis, the results showed that Taiwan stock market has negative impulse response on USDX and WTI while it has positive impulse response on LLG, DJIA and NASDAQ, and the impulse lasted for 4 periods. Finally, the results of Forecast Error Variance Decomposition showed there are high self-explanation powers for all variables on both models and NASDAQ has the most influence from DJIA and Taiwan stock market in model 2. REFERENCE English Albert, M.B., Avery, D., Narin, F. & McAllister,P.,(1990) ”Direct Validation of Citation Counts as Indicators of Industrially Important Patents”, Research Policy, Vol.20, Iss.2, pp251-259. Breitzman, A. F., & Narin, F., (2001), “Method and apparatus for choosing a stock portfolio, based on patent indicators”, United States Patent, 6175824. Bollerslev, T., (1986), “Generalized Autoregressive conditional heteroscedasticity,” Journal of Econometrics, vol.31, pp.307-328. Box, G.E.P. & Jenkins, G.M., (1976), Time Series Analysis： Forecasting and Control. San Francisco； Holden-Day, Press. Collins, Peter, Wyatt & Suzann, (1988), “Citation in patent to the Basic Research Literature”, Research Policy, Vol.17, Iss.2, pp65-74. Day, Theodore E., Lewis & Craig M.,” Stock Market Volatility and the Information Content of Stock Index Options”, Journal of Econometrics. Amsterdam: Apr 1992.Vol 52, Iss. 1-2; p267. Deng F., B.Lev, & Narin F., (1999), “Science and Technology as Predictors of Stock Performance,”Financial Analysts Journal, May/Jun, Vol. 55, No.3. 257
16. Dietmar H., Frederic M.S. & Katrin V., (2002), “Citations, family size, opposition and the value of patent rights” Research Policy, Vol.32, Iss.8, pp1343-1363 Dickey, D. A. & Fuller W. A., (1979), “Distribution of the Estimators for Autoregressive Time Series with a Unit Root”, Journal of American Statistical Association, 76, pp427- 431. Enders, W., (2004), Applied Econometric Time Series. New York： John Willy & Sons, Inc. Engle, R.F., (1982),“Autoregressive Conditional Heteroscedasticity with Estimates of the Variance of United Kingdom Inflation.”Econometrica, (50), pp987-1007. Engle, R.F. & B.S. Yoo (1987), “Forecasting and Testing in Cointegrated Systems”, Journal of Econometrics .35, pp143-159. Green, J. P., Stark, Andrew W & Thomas, H. M. , “UK evidence on the market valuation of research and development expenditures”, Journal of Business Finance & Accounting. Oxford: Mar 1996.Vol 23, Iss. 2; pg. 191, 26 pg Granger, C. W. J.（ 1969） . “Investigating causal relations by econometric models and cross-Spectral methods”. Econometrica, 37, pp424-438. Granger, C.W.J., & Engle, R.F., (1987), “Co-integration and Error Correction: Representation, Estimation, and Testing ”Econometrica 55 ,March, pp251-276. Granger, C. & P. Newbold, (1974), “ Spurious regressions in econometrics,” Journal of Econometrics 2, pp111-120. Hall, B., (1999),“ Innovation and Market Value,” NBER Working Paper Series, No. 6984. Hall, B. H., Jaffe A. B., & Trajtenberg M., (2001), “Market Value and Patent Citation： A First Look”, University of California, Berkeley, Dept of Economics Working Paper. Hardouvelis, G. A., (1990), “Margin Requirements, Volatility, and the Transitory Component of Stock Prices,” The American Economic Review, Vol. 80, No.4, pp736- 762. Hsieh, D.A. & M. H. Miller, (1990), “Margin Regulation and Stock Market Volatility,”The Journal of Finance, Vol. Xlv, No.1, pp3-29. Lamoureux, C. G , & Lastrapes, W. D., (1990), “Persistence in Variance, Structural Change, and the GARCH Model”, Journal of Business and Economic Statistics, 8: pp225-234. Levine, Ross & S. Zervos, (1998), “Stock Markets, Banks, and Economic Growth”, The American Economic Review 88, pp537-557. Lee, S. B. & Yoo T. Y., (1993), “Margin Regulation and Stock Market Volatility: further evidence from Japan, Korea and Taiwan,” Pacific-Basin Finance Journal, 1, pp155- 174. Maddala, G. S. & I. Kim, (1998), “Unit Roots, Cointegration, and Structural Change”. Cambridge： University Press. Markman G. D., Espina M. I. & Phan P. H, (2004),“Patent as Surrogates for Inimitable and Non-Substitutable Resources,”Journal of Management, Vol.30(4), pp529-544. Narin, France, Noma, Elliot, Perry & Ross, (1987),” Patent as Indictors of Corporate Technological Strength , ”Research Policy, Vol.16,Iss2-4, pp143-155. 258
17. Patrick Thomas, (2001), “A relationship between technology indicators and stock market performance,”Scientometrics, Vol. 51, No. 1, pp319-333. Salim M.D. & Partha, (1997), “Co-movement in International Equity Markets,”Journal of Financial Research. , Vol. 20, pp305-323. Said, E. & D. A. Dickey, (1984), “Testing for unit roots in autoregressive moving average models of unknown order,” Biometrika, Vol.71, pp599-607. Sill, D. K., (1993), “Predicting Stock—Market Volatility,” Business Review—Federal Reserve Bank of Philadelphia, pp15-p27. Sims, C.A. (1980), “Macro-ecoomics and Reality”, Econometrica, vol. 33, pp 1-48. Solnik & Bruno, (1983) “The Relation between Stock Prices and Inflationary Expectations: The International Evidence” The Journal of Finance. Cambridge: Mar 1983.Vol 38, Iss. 1; pg. 35, 14 pgs. Theodossiou, P. & U. Lee, (1993), “Mean and volatility spillovers across major national stock market: future empirical evidence”, Journal of Finance Research, 16, pp337-350. Thomas, P., Mcmillan, G S. & Abington P. S., (2001), “Using Science and Technology Indicators to Manage R&D as A Business.” , Engineering Management Journal,September,Vol.13, No.3. Yao, B & D. Ge,(2002) “Value Creation Through 〝Going Together〞： An Event Study on Market Response to Technology Allience Formation”, Working Paper. Yoon, Y. & Swales, J. (1991),”Prediction stock price performance: a neural network approach”, Proceeding of Twenty-Fourth Annual Hawaii International Conference on System Science, pp156-162 Chinese 宋嘉凌，台灣股市與主要國際股市之相關性研究，碩士論文，臺灣大學國際企業學 研究所，台北(2006)。 李冠賢，匯率、黃金與原油價格互動關係之研究- 以台灣為例，碩士論文，真理大學財經研究所，新北(2011)。 張芳倩，原油價格與大盤及類股股價指數之相關性，碩士論文，中正大學財務金融 所，嘉義(2006)。 黃姿穎，油價、金價、匯率與國際股市之關聯性研究，碩士論文，義守大學財務金 融學系碩士班，高雄(2009)。 黃慧文，原油價格對台灣股票市場之影響，碩士論文，屏東商業技術學院國際企業 所，屏東(2008)。 謝鎮州，股票、黃金與原油價格互動關係之研究- 以台灣為例，碩士論文，逢甲大學經濟學系碩士班，台中(2006)。 259