LV Thạc sỹ_Non linear model predictability of Vietnam stock market price

NATIONAL ECONOMICS UNIVERSITY HANOI INSTITUTE OF SOCIAL STUDIES THE HAGUE VIETNAM – NETHERLANDS CENTER FOR DEVELOPMENT ECONOMICS AND PUBLIC POLICY Non-linear model predictability of Vietnam stock market price A thesis presented by NATIONAL ECONOMICS UNIVERSITY HANOI INSTITUTE OF SOCIAL STUDIES THE HAGUE VIETNAM – NETHERLANDS CENTER FOR DEVELOPMENT ECONOMICS AND PUBLIC POLICY Non-linear model predictability of Vietnam stock market price A thesis presented by In partial fulfillment of the requirements for the obtaining the Degree of MASTER OF ARTS IN DEVELOPMENT ECONOMICS CERTIFICATION "I certify that the substance of this dissertation has not already been submitted for any degree and is not being currently submitted for any other degree I certify that to the best of my knowledge any help received in preparing this dissertation, and all sources used, have been acknowledged in this dissertation" LIST OF CONTENTS ACKNOWLEDGEMENTS ABBREVIATION LIST OF TABLES LIST OF FIGURES CHAPTER 1: INTRODUCTION 1.1 Problem Statement 1.2 Reseach Objective .8 1.3 Research questions 1.4 Literature review .8 1.5 Thesis Methodology 11 1.6 Thesis Structure .11 CHAPTER 2: ANALYTICAL FRAMWORK 12 2.1 Concept and Definition 12 2.2 Models for predicting stock market price 13 CHAPTER 3: PERFORMANCE OF VIETNAM STOCK MARKET 19 3.1 Overview of Vietnam Stock Market 19 3.2 Performance of five Blue Chips in therecent time 27 CHAPTER 4: MODEL SPECIPICATION AND DATA ANALYSIS 32 4.1 Description of data 32 4.2 Estimated results 32 CHAPTER 5: CONCLUSIONS 46 REFERENCE .48 APPENDIX 51 ACKNOWLEDGEMENTS I would like to express my special thanks Ph.D…, at National Economic University in Hanoi for his valuable instructions, comments, criticism, and correction during the development of this thesis My deepest gratitude goes to the Vietnam - Netherlands Project for MA Program in Development Economics where I have learned and been able to write this thesis I would like to thank all lecturers and the staff in the Project for their long, kind encouragement and timely assistance And I wish to express my warmest thanks to my family and my friends for their encouragement and constructive suggestions throughout my course Without their spiritual and support, the thesis would have been made impossible ABBREVIATION HSX Hochiminh Stock Exchange HNX Hanoi Stock Exchange LSTR Logistic Smooth Transition Regression STR Smooth Transition Regression GBM Geometric Brownian Motion MR Mean Reversion Model AR Autoregressive Process ARIMA Autoregressive integrated moving average ADF Augmented Dickey-Fuller VAR Vector auto regression GSO General Statistics Office LIST OF TABLES Table 1: Capitalization level, proportion, and growth rate of the stock market 20 Table 2: The growth rate of the number of stock companies and asset management companies 22 Table 3: Business performance of Blue Chips 27 Table 4: Testing linear or non-linear of data series for suggesting model 32 Table 5: Augmented Dickey-Fuller (ADF) Unit Root Test results .37 Table 6: Estimation results of LSTR1 model .40 LIST OF FIGURES Figure 1: Size of listing on Vietnam’s Stock Trading Center .21 Figure 2: VN-Index in 2010 .24 Figure 3: Upcom-Index and tranding value in Upcom market 26 Figure 4: Performances of Blue chips on the stock market up to 31 October, 2011 .30 Figure 5: Defining initial values of γ and c by using the method of GRID SEARCH .35 Figure 7: ADF Test for residual of error terms with one lag 40 Figure 8: The graph of STR model .42 Figure 9: VN-Index, HN-Index, and five blue chips forecast 45 CHAPTER 1: INTRODUCTION 1.1 Problem Statement Vietnam’s stock market has operated for nearly ten years and played an important role on Vietnam’s economy Nowadays, most of financial organizations and investors have paid more their attentions to predict stock market, notably by analyzing random-walk behavior of stock time series, in order to take benefits from investing in the stock market However, there are lacking in qualitative studies in forecasting Vietnam’s stock market The main reason is that Vietnamese forecasters are lacking in professional knowledge and qualitative analysis skills Moreover, forecasting stock market has been becoming more difficulty due to the data of the economy, enterprises, and the market are not enough long time series and less confidence Nowadays, many scientists and researchers have paid more attention to develop an adaptive analysis method for non-linear time series A number of time series forecasting models have been developed and applied in analysing and forecasting the stock market and stock price moves such as Smoothing Exponential Regression, Threshold Regression, Artificial Neural Network, and Smooth Transition Regression (STAR), Logistic Smooth Transition Regression (LSTR), etc In fact, there are only some typical studies in analyzing Vietnam’s stock market such as Hoang Dinh Tuan (2008) and most of current studies used ARIMA model to forecast stock index in short-term Therefore, this study is going to predict stock index, particularly VN-index, HNIndex and some of stock index with large capitalization, by applying non-linear model (Logistic Smooth Transition Regression – LSTR) in order to provide useful information for investors and financial organizations The result of this study hopefully contributes an effective method in analyzing and forecasting the stock market and stock prices in Vietnam 1.2 Research Objectives The objective of this study is predicting the stock market and stock price moves, particularly VN-index and HN-Index and blue-chip stock price of stock with large capitalization, by applying non-linear model (Logistic Smooth Transition Regression – LSTR)in order to provide useful information for investors as well as financial organizations Suggesting recommendations and policy implications for developing Vietnam’s stock market in the near future is also an objective of this study 1.3 Research questions  Question 1: Whether non-linear model will be used to predict VN-Index and HNIndex?  Question 2: Whether non-linear model will be better than linear model in predicting Vietnam stock market price?  Question 3: How does blue-chip stock price of HSX (Hochiminh Stock Exchange) and HNX (Hanoi Stock Exchange) look like by using non-linear model? 1.4 Literature review Analyzing and forecasting Vietnam’s stock market always attracts more concern of financial organizations and investors in the world, especially in term of stock market prices Bachelier (1900) shows that stock market prices were changed consecutively with an augmentation: S(t+Δ)-S(t)=Δ1/2 at time t In 1923, Alber Einstein given Brown W(t) moving process as studying kinematic elements and then the stock market price is a component of Brown W(t) According to L.Savege, stock market prices were never negative Brownian Motion Model was given in order to describe kinematics regarding to the market price of different kinds of stocks (P.Samuelson, 1965), for instance: S(t) = [22] Samuelson, Paul A (1965), “Rational Theory of Warrant Pricing” Indus Management rev (Spring 1965): 13-31 [23] Sarantis N (2001), “Nonlinearities, cyclical behavior and predictability in stock markets: international evidence”, International Journal of Forecasting 17, 459-482 [24] Terasvirta T and H.M Anderson (1992), “Characterizing nonlinearities in business cycles using smooth transition autoregressive models”, Journal of Applied Econometrics 7, S119-S136 [25] Terasvirta, Timo (1994), “Specification, estimation and evaluation of smooth transition autoregressive models”, Journal of the American Statistical Association 89, 208-218 [26] Tong H (1990), Non-Linear Time Series, A Dynamical System Approach, Oxford University Press, Oxford In Vietnamese [27] Hoang Dinh Tuan (2008), “Using some random process of analyzing stock price moves on Vietnam’s stock market”, paper of the workshop: the issues of financial economics and applied mathematics, National Economic University [28] Nguyen Khac Minh (2009), “Smooth Transition and GDP growth in Vietnamese industrial sector”, Journal of Economics & Development [29] Nguyen Viet Hung (2010), “Short term prediction of VN INDEX and HNX-INDEX”, Journal of Economics & Development, N0 152, pp 36-42 52 APPENDIX I STR Estimation Results VN-Index variables in AR part: restriction theta=0: restriction phi=0: restriction phi=-theta: transition variable: sample range: transition function: number of iterations: variable CONST trend(t) str_resids(t-1) str_resids(t-1) trend(t) [3, 706], T = 704 LSTR1 start Estimate SD t-stat p-value - linear part -CONST trend(t) str_resids(t-1) nonlinear part -CONST trend(t) Gamma C1 5.74781 -0.03734 0.96211 5.75850 0.0096 597.3638 -0.02929 0.0018 -16.3076 0.96132 0.0102 94.3639 0.63501 0.62309 0.03685 0.02880 7.33518 7.54421 27.24138 35.34717 0.0108 0.0018 0.0902 2.0821 57.8146 16.0476 83.6655 16.9769 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 AIC: -8.1920e+00 SC: -8.1467e+00 HQ: -8.1745e+00 R2: 9.9224e-01 adjusted R2: 0.9923 variance of transition variable: 41360.0000 SD of transition variable: 203.3716 variance of residuals: 0.0003 SD of residuals: 0.0166 HN-Index variables in AR part: restriction theta=0: restriction phi=0: restriction phi=-theta: transition variable: sample range: transition function: number of iterations: CONST HNX_trend(t) str_resids(t-1) CONST HNX_trend(t) [3, 386], T = 384 LSTR1 17 53 variable start - linear part -HNX_trend(t 0.13597 ) str_resids(t-1) 23.25871 nonlinear part -CONST 3.65529 HNX_trend(t -0.13167 ) str_resids(t-1) -22.43561 Gamma 10.00000 C1 49.37931 estimate SD t-stat p-value 0.54385 0.0717 7.5813 0.0000 15.95064 2.0522 7.7724 0.0000 4.02354 -0.54087 0.0041 972.2327 0.0000 0.0717 -7.5401 0.0000 -15.00868 2.0578 -7.2935 6.92011 0.2759 25.0803 -36.74764 5.3743 -6.8377 AIC: SC: HQ: R2: adjusted R2: variance of transition variable: SD of transition variable: variance of residuals: SD of residuals: 0.0000 0.0000 0.0000 -7.5366e+00 -7.4645e+00 -7.5080e+00 9.9448e-01 0.9945 12320.0000 110.9955 0.0005 0.0229 AGF variables in AR part: restriction theta=0: restriction phi=0: restriction phi=-theta: transition variable: sample range: transition function: number of iterations: variable CONST AGF_Trend(t) str_resids(t-1) AGF_Trend(t) [3, 336], T = 334 LSTR1 start estimate SD t-stat pvalue - linear part -CONST 3.26166 3.28000 -0.02275 244.357 -14.7903 0.0000 AGF_Trend(t) -0.01892 str_resids(t-1) 0.42364 0.47072 0.013 0.001 0.109 4.2865 0.0000 nonlinear part -CONST 0.26586 0.26217 18.2955 0.0000 AGF_Trend(t) 0.01822 0.02199 0.014 0.001 14.3873 0.0000 0.0000 54 str_resids(t-1) Gamma 0.52029 7.33518 0.47036 6.54484 C1 48.0689 44.0540 0.1143 4.1162 0.199 32.7448 1.460 30.1737 0.0000 0.0000 0.0000 55 AIC: SC: HQ: R2: adjusted R2: variance of transition variable: SD of transition variable: variance of residuals: SD of residuals: -7.2738e+00 -7.1825e+00 -7.2374e+00 9.7913e-01 0.9792 9324.1667 96.5617 0.0007 0.0260 BMC variables in AR part: restriction theta=0: restriction phi=0: restriction phi=-theta: transition variable: sample range: transition function: number of iterations: variable CONST BMC_trend(t) str_resids(t-1) str_resids(t-1) BMC_trend(t) [3, 506], T = 504 LSTR1 start - linear part -CONST 4.08092 BMC_trend(t ) str_resids(t-1) estimate SD t-stat pvalue 4.06188 0.005 0.000 0.013 763.260 -15.2747 0.0000 69.7742 0.0000 0.015 0.000 0.165 1.106 -81.1180 0.0000 28.7178 0.0000 32.7368 0.0000 200.198 0.0000 -0.00157 -0.00121 0.95250 0.95238 nonlinear part -CONST -1.31311 -1.28970 BMC_trend(t ) Gamma 0.00255 0.00217 5.96602 5.42040 C1 227.9310 221.5940 AIC: SC: HQ: R2: adjusted R2: variance of transition variable: SD of transition variable: variance of residuals: 0.0000 -7.0533e+00 -6.9947e+00 -7.0303e+00 9.9396e-01 0.9940 21210.0000 145.6365 0.0009 56 SD of residuals: 0.0292 FPT variables in AR part: restriction theta=0: restriction phi=0: restriction phi=-theta: transition variable: sample range: transition function: number of iterations: variable CONST FPT_trend(t) str_resids(t-1) str_resids(t-1) FPT_trend(t) [3, 463], T = 461 LSTR1 start estimate SD t-stat pvalue - linear part -CONST 4.05179 4.04913 0.00133 1336.916 42.9584 0.0000 FPT_trend(t) 0.003 0.000 0.022 39.1528 0.0000 0.026 0.000 0.161 1.150 -12.9987 0.0000 -12.7104 0.0000 31.4332 0.0000 232.0655 0.0000 0.00130 str_resids(t0.87429 1) nonlinear part -CONST -0.43944 0.87490 FPT_trend(t) -0.00053 -0.00078 Gamma 4.85242 5.07363 C1 272.2413 267.0849 -0.34523 AIC: SC: HQ: R2: adjusted R2: variance of transition variable: SD of transition variable: variance of residuals: SD of residuals: 0.0000 -7.8659e+00 -7.8031e+00 -7.8411e+00 9.8000e-01 0.9800 17748.5000 133.2235 0.0004 0.0194 PVD variables in AR part: CONST PVD_trend(t) str_resids(t-1) 57 restriction theta=0: restriction phi=0: restriction phi=-theta: transition variable: sample range: transition function: number of iterations: variable str_resids(t-1) PVD_trend(t) [3, 400], T = 398 LSTR1 start estimate SD t-stat pvalue - linear part -CONST 4.03507 4.03920 -0.00326 919.915 -35.8396 0.0000 PVD_trend(t) 0.004 0.000 0.023 37.9652 0.0000 0.013 0.000 0.151 0.997 39.2338 0.0000 9.2443 0.0000 32.4183 0.0000 164.945 0.0000 -0.00318 str_resids(t0.88923 1) nonlinear part -CONST 0.55234 0.88953 PVD_trend(t) 0.00064 0.00079 Gamma 4.85242 4.91804 C1 166.6896 164.5430 AIC: SC: HQ: R2: adjusted R2: variance of transition variable: SD of transition variable: variance of residuals: SD of residuals: 0.52879 0.0000 -7.7716e+00 -7.7015e+00 -7.7438e+00 9.7053e-01 0.9706 13233.5000 115.0370 0.0004 0.0204 SSI variables in AR part: restriction theta=0: restriction phi=0: restriction phi=-theta: transition variable: sample range: transition function: number of iterations: variable start CONST TREND(t) str_resids(t-1) str_resids(t-1) TREND(t) [3, 433], T = 431 LSTR1 estimate SD t-stat pvalue 58 - linear part -CONST 3.72445 3.73305 TREND(t) -0.00237 -0.00228 0.003 1167.5230 0.0000 0.000 -107.3580 0.0000 0.0119 80.5657 0.0000 str_resids(t0.87259 1) nonlinear part -CONST -1.75402 TREND(t) 0.00460 0.95621 Gamma 10.00000 6.10053 C1 316.6896 359.1268 -3.10019 0.00802 0.1141 0.000 0.422 2.653 -27.1820 28.6032 0.0000 0.0000 14.4360 0.0000 135.3687 0.0000 AIC: -7.3790e+00 SC: -7.3130e+00 HQ: -7.3529e+00 R2: 9.9328e-01 adjusted R2: 0.9933 variance of transition variable: 15516.0000 SD of transition variable: 124.5632 variance of residuals: 0.0006 SD of residuals: 0.0248 II ADF Test for reidual of data series VN-Index ADF Test for series: u_resids sample range: [1961 Q2, 2136 Q2], T = 701 lagged differences: no intercept, no time trend asymptotic critical values reference: Davidson, R and MacKinnon, J (1993), "Estimation and Inference in Econometrics" p 708, table 20.1, Oxford University Press, London 1% 5% 10% -2.56 -1.94 -1.62 value of test statistic: -14.9965 regression results: variable coefficient t-statistic x(-1) -0.8389 -14.9965 dx(-1) 0.0938 1.9932 dx(-2) 0.0238 0.6291 RSS 0.1785 59 OPTIMAL ENDOGENOUS LAGS FROM INFORMATION CRITERIA sample range: [1963 Q2, 2136 Q2], T = 693 optimal number of lags (searched up to 10 lags of differences): Akaike Info Criterion: Final Prediction Error: Hannan-Quinn Criterion: Schwarz Criterion: PORTMANTEAU TEST with 12 lags Portmanteau: p-Value (Chi^2): Ljung & Box: p-Value (Chi^2): 13.1505 0.3582 13.3328 0.3453 60 HN-Index ADF Test for series: str_resids sample range: [1961 Q2, 2056 Q2], T = 381 lagged differences: intercept, time trend asymptotic critical values reference: Davidson, R and MacKinnon, J (1993), "Estimation and Inference in Econometrics" p 708, table 20.1, Oxford University Press, London 1% 5% 10% -3.96 -3.41 -3.13 value of test statistic: -7.5356 regression results: variable coefficient t-statistic x(-1) -0.4830 -7.5356 dx(-1) -0.2923 -4.7918 dx(-2) -0.1829 -3.6783 constant -0.0003 -0.3343 trend 0.0000 0.5795 RSS 0.1419 OPTIMAL ENDOGENOUS LAGS FROM INFORMATION CRITERIA sample range: [1963 Q2, 2056 Q2], T = 373 optimal number of lags (searched up to 10 lags of differences): Akaike Info Criterion: Final Prediction Error: Hannan-Quinn Criterion: Schwarz Criterion: PORTMANTEAU TEST with 12 lags Portmanteau: p-Value (Chi^2): Ljung & Box: p-Value (Chi^2): 17.4597 0.1331 17.9926 0.1159 61 AGF ADF Test for series: str_resids sample range: [1961 Q2, 2043 Q4], T = 331 lagged differences: intercept, time trend asymptotic critical values reference: Davidson, R and MacKinnon, J (1993), "Estimation and Inference in Econometrics" p 708, table 20.1, Oxford University Press, London 1% 5% 10% -3.96 -3.41 -3.13 value of test statistic: -11.3531 regression results: variable coefficient t-statistic x(-1) -1.0166 -11.3531 dx(-1) 0.1477 2.0425 dx(-2) 0.0591 1.0761 constant -0.0001 -0.1055 trend 0.0000 0.1914 RSS 0.2112 OPTIMAL ENDOGENOUS LAGS FROM INFORMATION CRITERIA sample range: [1963 Q2, 2043 Q4], T = 323 optimal number of lags (searched up to 10 lags of differences): Akaike Info Criterion: Final Prediction Error: Hannan-Quinn Criterion: Schwarz Criterion: PORTMANTEAU TEST with 12 lags Portmanteau: p-Value (Chi^2): Ljung & Box: p-Value (Chi^2): 16.8228 0.1564 17.3452 0.1371 BMC 62 ADF Test for series: str_resids sample range: [1961 Q2, 2086 Q2], T = 501 lagged differences: intercept, time trend asymptotic critical values reference: Davidson, R and MacKinnon, J (1993), "Estimation and Inference in Econometrics" p 708, table 20.1, Oxford University Press, London 1% 5% 10% -3.96 -3.41 -3.13 value of test statistic: -11.1740 regression results: variable coefficient t-statistic x(-1) -0.7437 -11.1740 dx(-1) -0.0591 -1.0271 dx(-2) -0.0445 -0.9926 constant -0.0000 -0.0109 trend 0.0000 0.0259 RSS 0.4038 OPTIMAL ENDOGENOUS LAGS FROM INFORMATION CRITERIA sample range: [1963 Q2, 2086 Q2], T = 493 optimal number of lags (searched up to 10 lags of differences): Akaike Info Criterion: Final Prediction Error: Hannan-Quinn Criterion: Schwarz Criterion: PORTMANTEAU TEST with 12 lags Portmanteau: p-Value (Chi^2): Ljung & Box: p-Value (Chi^2): 8.2828 0.7627 8.4350 0.7503 63 FPT ADF Test for series: str_resids sample range: [1961 Q2, 2075 Q3], T = 458 lagged differences: intercept, time trend asymptotic critical values reference: Davidson, R and MacKinnon, J (1993), "Estimation and Inference in Econometrics" p 708, table 20.1, Oxford University Press, London 1% 5% 10% -3.96 -3.41 -3.13 value of test statistic: -12.1304 regression results: variable coefficient t-statistic x(-1) -0.9223 -12.1304 dx(-1) -0.0263 -0.4109 dx(-2) 0.0370 0.7919 constant -0.0002 -0.2039 trend 0.0000 0.3560 RSS 0.1673 OPTIMAL ENDOGENOUS LAGS FROM INFORMATION CRITERIA sample range: [1963 Q2, 2075 Q3], T = 450 optimal number of lags (searched up to 10 lags of differences): Akaike Info Criterion: Final Prediction Error: Hannan-Quinn Criterion: Schwarz Criterion: PORTMANTEAU TEST with 12 lags Portmanteau: p-Value (Chi^2): Ljung & Box: p-Value (Chi^2): 6.4791 0.8900 6.6231 0.8815 64 PVD ADF Test for series: str_resids sample range: [1961 Q2, 2059 Q4], T = 395 lagged differences: no intercept, no time trend asymptotic critical values reference: Davidson, R and MacKinnon, J (1993), "Estimation and Inference in Econometrics" p 708, table 20.1, Oxford University Press, London 1% 5% 10% -2.56 -1.94 -1.62 value of test statistic: -11.6700 regression results: variable coefficient t-statistic x(-1) -0.9661 -11.6700 dx(-1) 0.0758 1.1196 dx(-2) 0.0169 0.3332 RSS 0.1589 OPTIMAL ENDOGENOUS LAGS FROM INFORMATION CRITERIA sample range: [1963 Q2, 2059 Q4], T = 387 optimal number of lags (searched up to 10 lags of differences): Akaike Info Criterion: Final Prediction Error: Hannan-Quinn Criterion: Schwarz Criterion: PORTMANTEAU TEST with 12 lags Portmanteau: p-Value (Chi^2): Ljung & Box: p-Value (Chi^2): 12.7482 0.3876 13.0099 0.3683 65 SSI ADF Test for series: str_resids sample range: [1961 Q2, 2068 Q1], T = 428 lagged differences: no intercept, no time trend asymptotic critical values reference: Davidson, R and MacKinnon, J (1993), "Estimation and Inference in Econometrics" p 708, table 20.1, Oxford University Press, London 1% 5% 10% -2.56 -1.94 -1.62 value of test statistic: -10.8424 regression results: variable coefficient t-statistic x(-1) -0.8635 -10.8424 dx(-1) -0.0248 -0.3804 dx(-2) -0.0764 -1.5778 RSS 0.2369 OPTIMAL ENDOGENOUS LAGS FROM INFORMATION CRITERIA sample range: [1963 Q2, 2068 Q1], T = 420 optimal number of lags (searched up to 10 lags of differences): Akaike Info Criterion: Final Prediction Error: Hannan-Quinn Criterion: Schwarz Criterion: PORTMANTEAU TEST with 12 lags Portmanteau: p-Value (Chi^2): Ljung & Box: p-Value (Chi^2): 7.0971 0.8511 7.2809 0.8385 66 ... the formulation of LSTR1, LSTR2, and ESTR models 19 CHAPTER 3: PERFORMANCE OF VIETNAM STOCK MARKET 3.1 Overview of Vietnam Stock Market Despite the birth of Vietnamese Stock Market had its start... non -linear model will be better than linear model in predicting Vietnam stock market price?  Question 3: How does blue-chip stock price of HSX (Hochiminh Stock Exchange) and HNX (Hanoi Stock. .. 12 2.2 Models for predicting stock market price 13 CHAPTER 3: PERFORMANCE OF VIETNAM STOCK MARKET 19 3.1 Overview of Vietnam Stock Market 19 3.2 Performance of five Blue