FOREX FORECASTING BY USING NGARCH MODEL GAN LONG FATT This report submitted in partial fulfillment of the requirements for the award of the degree of Master of Science (Mathematics). Faculty of Science Universiti Teknologi Malaysia November 2009 iii To my beloved family, for your love and support. To my friends, for your wits, intelligence and guidance in life. iv ACKNOWLEDGEMENT First and foremost, I would like to extend my heartfelt gratitude to my supervisor, Prof. Dr. Zuhaimy Hjh Ismail for his guidance and support that he has given me throughtout the duration of this research. My fellow postgraduate students should also be recognized for their support. My sincere appreciation also extends to all my colleagues and other who have provided assistance at various occasions. Their views and opinions are helpful indeed. Unfortunately it is impossible to list all of them in this limited space. Last but not least, I would like to thank my family members who have given me their undying support. v ABSTRACT Foreign Exchange (Forex) is the market where a nation's currency trade with another. Flexible exchange rate started in 1973 since Bretton Woods Agreement was breakdown and it because the fluctuating exchange rate moves more drastically then before. Thus, forecasting exchange rates have become very important and challenge research issue for both academic and industrial. In this study used exchange rate selling prices of RM/USD. The daily data cover the period from 03/05/2007 to 29/05/2009 exchange rate from Bank Negara Malaysia and it was volatility and moves through the times. Thus, NGARCH model will be introduced in this study to forecast the selling prices for RM/USD exchange rate in future and GARCH(1,1) as a benchmark. All the data was analyzed by using Microsoft Office Excel 2007 software. The forecast performance value of RMSE of NGARCH (1, 1) is 0.022809308 smaller than RMSE of GARCH(1, 1) which value is 0.23439891. Therefore, the NGARCH model is more accuracy than GARCH model. vi ABSTRAK Pertukaran Wang Asing (FOREX) adalah pasaran di mana sebuah negara perdagangan mata wang dengan satu negara yang lain. Pada 1973, keruntuhan Bretton Woods Agreement menyebabkan keseluruhan dunia mula menerima kadar pertukaran yang fleksibel. Oleh itu, kadar naik turun pertukaran sering bergerak lebih drastik daripada dahulu. Oleh itu, kadar-kadar pertukaran peramalan telah menjadi amat penting dan mencabar dalam kedua-dua bidang akademik dan perindustrian. Dalam kajian ini, harga jualan kadar pertukaran RM/USD telah digunakan. Tempoh data yang digunakan daripada Bank Negara Malaysia dari 03/05/2007 hingga 29/05/2009 dimana kemeruapan dan bergerak melalui masa-masa itu. Oleh itu, model NGARCH akan diperkenalkan dalam kajian ini untuk meramalkan harga jualan untuk RM/ USD kemudian hari dan GARCH(1,1) seperti satu penanda aras. Semua data itu telah dianalisis dengan menggunakan perisian Microsoft Office Excel 2007. Prestasi ramalan NGARCH (1, 1) di mana nilai RMSE adalah 0.022809308 lebih kecil daripada GARCH (1, 1) di mana value sama dengan 0.23439891. Oleh itu, model NGARCH adalah lebih tepat daripada model GARCH. ix TABLE OF CONTENTS CHAPTER I TITLE PAGE TITLE i DECLARATION ii DEDICATION iii ACKNOWLEDGEMENTS iv ABSTRACT v ABSTRAK vi TABLE OF CONTENTS vii LIST OF TABLES x LIST OF FIGURES xii LIST OF SYMBOLS xiv LIST OF APPENDICES xv INTRODUCTION 1.0 Introduction 1 1.1 History of Foreign Exchange 2 1.2 Problem Statement 4 1.3 Objective of the Study 4 1.4 Scope of the Study 5 1.5 Organization of the Research 5 ix II III IV LITERATURE REVIEW 2.0 Introduction 7 2.1 Literature of ARCH and GARCH Model 7 2.2 Literature of NGARCH Model 9 METHODOLOGY 3.0 Introduction 11 3.1 Data 11 3.2 Stylized Facts of Asset Return 12 3.3 Generalised ARCH Models 14 3.3.1 NGARCH (1, 1) model 14 3.3.2 GARCH (1, 1) model 16 3.4 Maximum likelihood method 20 3.5 3.6 Microsoft Office Excel 2007 Forecast Performance 21 25 RESULT AND DISCUSSION 4.0 Introduction 26 4.1 The Exchange Rate Series 26 4.2 Stationary Series 30 4.3 Estimation 32 4.3.1. Parameters Estimation of NGARCH (1, 1) 32 4.3.2. Parameters Estimation of GARCH (1, 1) 38 Forecasting 43 4.4 ix 4.4.1 Forecasting future value by using 44 NGARCH (1, 1) 4.4.2Forecasting future value by using GARCH 45 (1,1) 4.5 Comparison of NGARCH (1,1) and GARCH (1,1) Model 47 47 4.5.1 Comparison of Forecasting Model V CONCLUSION 5.0 Introduction 49 5.1 Result and Comparison 50 5.2 Suggestion for Further Research 51 REFERENCES 52 APPENDICES 55 x LIST OF TABLES TABLE NO. 4.1 TITLE Data analysis of first order difference exchange rate selling prices series PAGE 31 4.2 Starting value of coefficient parameters of NGARCH 32 4.3 Value of parameters coefficient for NGARCH(1,1) 33 4.4 Standardized return for NGARCH(1,1) Model 37 4.5 Starting value of parameters coefficient for GARCH(1,1) model 38 4.6 The coefficient parameters of GARCH(1,1) 39 4.7 Standardized Return for GARCH(1,1) Model 43 4.8 Comparison the real value against forecast value by using NGARCH(1,1) 44 Comparison the real value against forecast value 4.9 by using GARCH(1,1) 45 xi LIST OF TABLES TABLE NO. TITLE PAGE 4.10 RMSE value of NGARCH(1,1) 47 4.11 RMSE value of GARCH(1,1) 48 5.1 The plot of actual prices against forecast prices by NGACH(1,1) and GARCH(1,1) 53 xii LIST OF FIGURES FIGURE NO. TITLE PAGE 3.1 Estimation of unknown parameters- Step 1 Estimation of unknown parameters- Step 2(ii) 22 3.2 3.3 3.4 4.1 4.2 4.3 Estimation of unknown parameters- Step 2(iii) 23 24 Estimation of unknown parameters- Step 3(i) The dynamics of the RM/USD exchange rate selling prices in the period 2007-2009 Daily log returns for the RM/USD exchange rate The squared innovation return in the period 24 27 28 29 2007-2009 4.4 First Order Difference Selling Prices of RM/USD 30 4.5 Answer Report from Solver for NGARCH(1,1) 34 4.6 Conditional Variance Model 35 4.7 Conditional Standard NGARCH(1,1) Model 4.8 Answer Report from Solver for GARCH(1,1) 40 4.9 Conditional Variance for GARCH(1,1) Model 41 4.10 Conditional Standard Deviation for GARCH(1,1) Model 42 for NGARCH(1,1) Deviation for 36 xiii LIST OF FIGURES FIGURE NO. 4.11 5.1 TITLE The plot of actual prices against forecast prices by NGARCH(1,1) and GARCH(1,1) model PAGE 46 The plot of actual prices against forecast prices by NGARCH(1,1) and GARCH(1,1) model 50 xiv LIST OF SYMBOLS - Logarithm return for daily closing prices C - The observed assets prices , , - Conditional variance - Coefficient parameter of GARCH model - the correlation between returns and variance - iid standard normal random variable 0,1 - the constant one period riskless domestic interest rate - the constant one period riskless foreign interest rate - the constant risk premium (the reward for investing in the foreign currency) xv LIST OF APPENDICES APPENDIX TITLE PAGE Foreign Exchange Daily Data for the selling prices of RM/USD. 52 NO. A CHAPTER I INTRODUCTION 1.0 Introduction Foreign Exchange (Forex) is the market where a nation's currency trade with another. In 1973, Bretton Woods Agreement was breakdown and it makes most of countries in the world started to accept flexible exchange rate. Therefore, fluctuating of exchange rate often move more drastically than before. Thus, forecasting exchange rates have become very important and challenge research issue for both academic and industrial. Due to the fact that Forex forecasting is of practical as well as theoretical importance, a large number of methods and techniques (including linear and nonlinear) were introduced to beat random walk model in Forex market. With increasing development of time series forecasting, researchers and investors are hoping that the mysteries of the Forex market can be solve. In last two decades, extensive research has been carried out on the returns of time series data Generalized Autoregressive Conditional Heteroscedastic (GARCH) models. Many previous studies have been discussed for FOREX forecasting. In this study used exchange rate United State Dollar (USD) versus Ringgit Malaysia (RM). The daily data cover the period from 03/05/2007 to 29/05/2009 2 exchange rate from Bank Negara Malaysia and it was volatility and moves through the times. Thus, GARCH models will be introduced in this study to forecast the exchange rate in future. A GARCH process is a form of stochastic process that is commonly used in finance and economics. These models are known as univariate time series models and is very useful in analyze and forecast the volatility. These models were developing in order to model and forecast the variance of financial and economic time series over time. GARCH models are simple and easy to handle, it take care of clustered errors, nonlinearities and changes in the econometrician’s ability to forecast. GARCH is a time series modeling technique that used past variances and past variance forecasts to forecast future variances. 1.1 History of Foreign Exchange At 1870s, Western Europe was acceptance with their countries currency set par value in term of gold which is the gold standard. Since governments agreed to buy or sell gold on demand with anyone at its own fixed parity rate, therefore the value of each currency in terms of gold are fixed. Most central banks supported their currency with convertibility to gold before World War I (WWI). Although paper money could always be exchanged for gold, in reality this did not occur often, fostering the sometimes disastrous notion that there was not necessarily a need for full cover in the central reserves of the government. 3 Occasionally, inflation and political instability are causes by the ballooning supply of paper money without gold cover. Foreign exchange controls were introduced to prevent market forces from punishing monetary irresponsibility to protect local national interests. In July 1994, the Bretton Woods agreement was reached on the initiative of the USA in the latter stages of World War II (WWII). The agreement established a US dollar based monetary system. Other international institutions such as the IMF, the World Bank and GATT (General Agreement on Tariffs and Trade) were created in the same period as the emerging victors of WWII searched for a way to avoid the destabilizing monetary crises which led to the war. The Bretton Woods agreement resulted in a system of fixed exchange rates that partly reinstated the gold standard, fixing the US dollar at USD35/oz and fixing the other main currencies to the dollar - and was intended to be permanent. In 1973, Bretton Woods Agreement was breakdown and the dollar was no longer suitable as the sole international currency at a time when it was under severe pressure from increasing US budget and trade deficits. Foreign exchange trading was start develop in the following decades into all around the world. Restrictions on capital flows have been removed in most countries and most of countries in the world started to accept flexible exchange rate. 4 1.2 Problem Statement GARCH models take care of the kurtosis of the frequency and the volatility clustering of the data. These two characteristics are very important in financial time series. It can provide the accuracy of the forecasts of variances and covariance in assets returns. It was ability to model time-varying conditional variances. The exchange rate will be forecast by GARCH in this research. In fact, GARCH model can also be apply in such field like risk management, portfolio management and asset allocation, option pricing, foreign exchange and the structure of interest rates. In this research, we can easy found the highly significant of GARCH in the equity market, at present and for the future. GARCH models can use to examine the relationship between the long-term and short-term of the interest rates. Moreover, GARCH models can be using in analysis the time-varying risk premiums and the Forex market which consist the highly persistent periods of volatility and clustering. 1.3 Objectives of the Study The objectives of this study are: i. To explore NGARCH and GARCH model of forecasting exchange rate by using Microsoft Excel 2007. ii. To explore the NGARCH model in predict the Forex data. iii. To predict the daily exchange rate for RM/ USD selling prices. 5 1.4 Scope of the Study This research is focuses on examination of the Forex market by using NGARCH model. The GARCH models will be introduced as the benchmark to this study. There cover 2 years of exchange rate data will be forecast by using NGARCH model for more understanding for the reader. Hopefully this study will increase the understanding and further exploration about the GARCH model. 1.5 Organization of the Research This research is organized into five chapters. Chapter 1 gives the introduction of this study. It begins with the overview of FOREX market, GARCH models appoarch, the objectives and the scopes. Chapter 2 and 3 describes the introduction of GARCH models with its specifications. The details and principles will also be focusing. It will present the GARCH family with their consistency and the appropriate model for the data will be selected. Chapter 4 describes introduction of case study and analyze the data of Forex market which RM/USD. The data consists of daily observations of buying and selling price of exchange rate against RM (Ringgit Malaysia) over the periods 2007 – 2009. Between, this chapter will identify the variables that can influence the moving of Forex and its future. These variables will be included in the models so that the results in predict the exchange rate will more consistency and accurately. 6 Chapter 5 gives a summary of the whole study. Some conclusions are drawn and finally, some thoughts on possible directions in which future research in this area might be pursued are offered. Chapter II LITERATURE REVIEW 2.0 Introduction This chapter contains a literature review of the study of justify the stated objectives. Section 2.1 contains the literature of ARCH and GARCH model. Section 2.2 studies the literature of NGARCH model. 2.1 Literature of ARCH and GARCH Model Financial time series such as stock price or exchange rate usually are available on very high frequencies such as minute by minute. However, the econometrician uses highly aggregated data such as daily or weekly returns. Recently, many empirical studies have found different estimating approaches around the volatility problem (Bates, 1996; Kim & Kim, 2004; Watanabe, 1999). As alternative to the historical and implied approach, numerous models are devised that correspond to the stochastic volatility process characteristic. One widespread approach is ARCH or generalized ARCH (GARCH), devised by Engle (1982) and Bollerslev (1986). 8 Engle (1982) developed the Autoregressive Conditional Heteroskedasticity (ARCH) model to characterize the observed serial correlation in asset price volatility. Suppose we assume that a price follows a random walk , where ~ 0, . The variance of the error term depends upon , and the objective of the model is to characterize the way in which this variance changes over time. The ARCH model assumes that this dependence can be captured by and autoregressive process of the form where the restrictions 0, ! 0 for " 0, 1, , # ensure that the predicted variance is always nonnegative. This specification illustrates clearly how current levels of volatility will be influenced by the past, and how periods of high or low price fluctuation will tend to persist.This volatility approach is calculated using observations of historical daily asset prices and considering both the conditional and unconditional variance in the estimation process. Afterward introduced by Engle, it follow by various extensions of the ARCH model were made by many researchers. Bollerslev (1986) and Taylor (1986), independently of each other, suggested the generalized ARCH (GARCH) model. GARCH models have been found to be suitable for describing stock returns data, such as volatility clustering and thick-tailed behavior. The GARCH model has been used to characterize patterns of volatility in US dollar foreign exchange markets (Baillie and Bollerslev 1989 and 1991). However, initial investigations into the explanatory power of out of sample forecasts produced disappointing results (West and Cho, 1995). Jorion (1995) found that volatility forecasts for several major currencies from the GARCH model were outperformed by implied volatilities generated from the Black Scholes option pricing model. 9 In 1991, Nelson was developed the exponential GARCH (EGARCH) model to analyze the effects of positive and negative stock market news, with negative shocks affecting volatility differently from positive shocks of equal magnitude. Moreover, the possibility of leverage, whereby negative shocks increase volatility while positive shocks decrease volatility, is present in the EGARCH formulation. In 1998, Andersen and Bollerslev were demonstrated that one can significantly improve the forecasting power of GARCH model by measuring volatility as the sum of intraday squared returns. This measure is referred to as integrated or realized volatility. In theory, if the true underlying price path is a diffusion process it is possible to obtain progressively more accurate estimates of the true volatility by increasing the frequency of intraday observation. Of course, there are practical limits to this; micro structural effects begin to degrade accuracy beyond a certain point. Ding, Granger and Engle (1993), Glosten, Jagannathan and Runkle (1993) and Hentschel (1995) were extensions of the leverage effect and other GARCH. Bollerslev, Chou and Kroner (1992) and the work of Duan (1997) enclose the existing GARCH models into a common system known as the increased GARCH (p,q) process. 2.2 Literature of NGARCH Model GARCH option pricing model can be describe by using the standard discretetime GARCH specification. Particularly, the GARCH process known as the non-linear asymmetric GARCH(1,1) (NGARCH(1,1)) process is restrict from the first appeared in Engle and Ng (1993). Our pricing results can, nevertheless, be extended to all GARCH(1,1) specifications with little effort. Since NGARCH (1,1) process fulfill the leverage effect which is an important feature of asset returns, hence it chosen over the standard linear GARCH(1,1) model. In, other hand, GARCH models such as the EGARCH process by Nelson (1991) and the GJR-GARCH process by Glosten et al. 10 (1993) can also capture the leverage effect. According to Duan (1997) concluded that the NGARCH model is better although Engle and Ng (1993) conclude the GJR-GARCH model performs best. Since there is no conclusive evidence in favor of a particular GARCH specification, our choice of the NGARCH model is better viewed as for the demonstration purpose for the time being. In the main paper (Petra POSEDEL, 2006), the author was state that the NGARCH model could be used for option pricing which as an alternative to the Black Scholes model. The author was shows that introducing heteroscedasticity results in better fitting of the empirical distribution of foreign currency than the Brownian model. For simulation, the NGARCH model show it is possible to described the empirically observed facts from foreign currency option markets. CHAPTER III METHODOLOGY 3.0 Introduction The time series is the observations of variable Y that become available over time, which are often recorded at fixed time intervals. Monthly inventory levels, weekly exchange rate, gold rate and sequence of annual sales figures are examples of time series. Generally, time series required special methods for the analysis as the observations are related to one another which called autocorrelation. This produces patterns of variability that can be used to forecast future values. This chapter contains a methodology of the study of justify the stated objectives. We will first discuss Generalised Autoregressive Conditional Heteroskedasticity (GARCH) model which is more flexible than the approach ARCH in Section 3.1. In Section 3.2, we will discuss the Maximum Likelihood method (MLE) to estimate the unknown parameters. The software used in this study will be discuss in Section 3.3. 3.1 Data In this research, the measurements of the data from 3rd May 2007 to 29th May 2009 are used in the model development and simulation and the remainder is kept for the 12 purpose of accuracy evaluation. These data consists of daily exchange rate in Bank Negara Malaysia (BNM). The data set obtained from the website of BNM. 3.2 Stylized Facts of Asset Return Movements in asset prices or equivalently asset returns is the causes of market risk, the first step is by defining returns and then give an overview of the characteristics of typical asset returns. By defining the daily geometric or “log” return on an asset as the change in the logarithm of the daily closing price of the asset, $%& ' $%& (3.1) By using log return it can easily calculate the compounded return at the K−day horizon simply as the sum of the daily returns. :) ln&) ' ln& ) , ln&- ' ln&- -. ) , ln&- – ln&- -. (3.2) - 13 Below are the following lists of so-called stylized facts, which apply to most stochastic returns. (i.) It have a very little &0 , 1 2 0 autocorrelation in daily return. for 3 1, 2, 3, . . . , 100 In other words, returns are almost impossible to predict from their own past. (ii.) The standard deviation of returns completely dominates the mean of returns at short horizons such as daily. It is not possible to statistically reject a zero mean return. (iii.) Variance, measured, for example, by squared returns, displays positive correlation with its own past. This is most evident at short horizons such as daily or weekly. (iv.) The negative correlation value between variance and returns is the leverage effect. (v.) Correlation between assets appears to be time varying. Importantly, the correlation between assets appears to increase in highly volatile down markets and extremely so during market crashes. (vi.) As the return-horizon increases, the unconditional return distribution changes and looks increasingly like the normal distribution. 14 3.3 Generalised ARCH Models GARCH is a short form of Generalized Autoregressive Conditional Heteroscedasticity. Heteroscedasticity know as time-varying variance which means volatility. Conditional implies a dependence on the observations of the immediate past, and autoregressive describes a feedback mechanism that incorporates past observations into the present. Therefore, GARCH model is a mechanism that includes past variances in the explanation of future variances. Besides that, GARCH is a time series modeling technique that uses past variances and past variance forecasts to forecast future variances. 3.3.1 NGARCH (1, 1) model The GARCH(1,1) is the simplest and most robust of the family of volatility models. However, the model can be extended and modified in many ways. NGARCH is the modification among of GARCH model. With modify the GARCH models so that the weight given to the return depends on whether the return is positive or negative in the following simple manner: – 7 2 ' 7 which is referred to as the NGARCH (nonlinear GARCH) model. (3.3) 15 The dynamics of the time series of return is described with a nonlinear-in mean, asymmetric GARCH (1,1) model (Engle and Ng, 1993): 8 ln 9 & : & ' ' (3. 4) ' (3. 5) where are i.i.d. standard normal random variables N (0,1) and ; 0, 0, 0, and 1 < 1 (3. 6) where the constant one period riskless domestic interest rate the constant one period riskless foreign interest rate the constant risk premium (the reward for investing in the foreign currency) the correlation between returns and variance. For the tomorrow’s variance, , is known at the end of today’s time . Let us denote with = > ? the expected return in time . Therefore, the expected return and variance of the return based on the information available until time can be write as = > ? ' ' (3.7) 16 @A > ? (3.8) The unconditional variance for NGARCH can be define as ? 8 = > B CDE F (3. 9) From relation Equation (3.7) and Equation (3.8) it follows that the forecast of the variance is directly given by the model with . If observe the forecast of daily returns variance for k periods ahead, using the recursive specification of the asymmetric GARCH model (3.5) it follows that ? = >' > 1 ?- (3. 10) where ? = = >represents the expected value of the future variance for horizon G. The expression 1 is the persistence of the model. If the value of 1 is near 1 then shocks in the market persist through a long timeG H ∞. In that case the time series will say has a long memory. In other word, when 1 is small, it show that shocks in returns die out more quickly in time. 3.3.2 GARCH (1, 1) model 17 The simplest generalized autoregressive conditional heteroskedasticity (GARCH) model is GARCH (1, 1) model. The (1,1) in parentheses is a standard notation in which the first number refers to how many autoregressive lags, or ARCH terms, appear in the equation, while the second number refers to how many moving average lags are specified, which here is often called the number of GARCH terms. The dynamic of the time series of return is described with a generalized autoregressive conditional heteroskedasticity (GARCH) model can be written as 8 ln 9 & : & ' ' (3.11) (3.12) where are independently and identically (i. i. d) standard normal variables 0,1 and ; 0, ; 0, ; 0, and < 1 (3.13) so that next period forecast of variance is a blend of last period forecast and last period’s squared return. For the tomorrow’s variance, , is known at the end of today’s time . Let us denote with = > ? the expected return in time . Therefore, the 18 expected return and variance of the return based on the information available until time can be write as = > ? ' ' @A > ? (3.14) (3.15) The unconditional, or long-run average, variance, , can define as ? 8 = > = > ? = > ? so that /1 ' ' (3.16) The GARCH model, in turn, implicitly relies on . This can be seen by solving for in the long-run variance equation and substituting it into the dynamic variance equation. We get 1 ' ' ' ' (3.17) 19 Thus, tomorrow’s variance is a weighted average of the long-run variance, today’s squared return, and today’s variance. Put differently, tomorrow’s variance is the long-run average variance with something added (subtracted) if today’s squared return is above (below) its long-run average, and something added (subtracted) if today’s variance is above (below) its long-run average. It is often useful not only to forecast next period’s variance of returns, but also to make an k-step ahead forecast, especially if the goal is to price an option with k- steps to expiration using the volatility model. Again starting from the , it can derive the forecast for next period’s GARCH(1,1) equation for variance, ? = >' = >- ' 2? = >- ' ? = >- - ' ? = > - ' ? = >- ? ' , so that = >? ' 2 - = > ? ' - J – K (3.18) where ? conditional expectation, refers to taking the expectation using all the = >- information available at the end of day t , which includes the squared return on day t itself. 20 From the above equation = >? H as G H ∞ so as the forecast horizon goes to infinity, the variance forecast approaches the unconditional variance. From the G-step ahead variance forecast, as the persistence of the model determines how quickly the variance forecast converges to the unconditional variance. In Section 3.3.1 and 3.3.2, we have mentioned the domestic and foreign interest rate in Equation (3.4), (3.11) and (3.14) which are and . In this study, for illustration purposes, and are taken as constants of known values. It is possible to consider and as time series and in that case we estimate them from annual report from BNM. On the annual level the respective rates are approximately 3.3% and 2.9%. The analysis of course can be done for any other choice of these values. Since the frequency of data is at a daily level, we transform the annual interest rate into a daily interest rate by dividing by 252 the average number of business days in one year. For , this is the constant risk premium i.e. the reward for investing in the foreign currency. We notice that a negative value of decreases the mean value RM/USD exchange return which indicates the appreciation of the domestic currency. Furthermore, if we are observing the USD/RM exchange rate a positive value of the parameter will explain the appreciation of the domestic currency. 3.4 Maximum likelihood method 21 Maximum Likelihood Estimation (MLE) is a popular statistical method used for fitting a statistical model to data, and providing estimates for the model’s parameters. Let denote with 7 , , , , for GARCH (1,1) model and NGARCH (1,1) model which are sets of unknown parameters. For GARCH (1, 1) and NGARCH (1, 1) model with Normal conditional returns, the likelihood function is $ √MNO exp ' Y E E SO TU TV WNOXY NOXY NOE (3.19) and thus the joint likelihood of the model is Z[ ∏[. $ ∏[. √MNO exp ' (3.20) Y E E SO TU TV WNOXY NOXY NOE (3.21) Since the ln Z function is monotonically increasing function of Z, with maximize the log of the likelihood function Z[ ∑[ >' ln2^ [ . ' Y E 9SO _TU TV WNOXY ENOXY `: ln ' ? NOE (3.22) under condition given by constraint (3.5) which for NGARCH (1,1) model and constraint (3.16) which for GARCH (1,1) model into the Equation (3.22). 22 The maximization of the function Z [ by the parameters of the model is done by using the Solver in Microsoft Office Excel. By using the Solver we can get the value of the unknown parameters. 3.5 Microsoft Office Excel 2007 Microsoft Office Excel 2007 is an electronic spreadsheet program that can be used for storing, organizing and manipulating data. Normally spreadsheets are often used to store financial data instance performing basic mathematical operations such as summing columns and rows of figures, finding values such as profit or loss calculating repayment plans for loans or mortgages and finding the average, maximum, or minimum values in a specified range of data. In this study, we will use the add-in function in Microsoft office Excel 2007 to add in the Solver to estimate the parameters which are , , , , and . To add in the Solver in Microsoft Excel 2007, go to Excel Option>> Add-Ins >> Manage: Excel Addins >> Solver add-in >> Ok. Here is the application of Microsoft Office Excel 2007 software in discussing the parameters estimated. Firstly, insert all the data in sequences of Date. After that key in all related function to get period return, square return, conditional variance and conditional standard deviation which as shown in Figure 3.1. 23 Step 1: Insert all the data in sequences. Figure 3.1: Estimation of unknown parameters- Step 1 To estimate the unknown parameters, we need to assume a number for , , , , and under the condition given in constraint Equation (3.5) and constraint Equation (3.16). Step 2: i.) Put the cursor at Target Cell that is cell G1. ii.) Click Data>> Solver. iii.) Key in all the data as Figure 3.3. 24 Figure 3.2: Estimation of unknown parameters- Step 2(ii) 25 Figure 3.3: Estimation of unknown parameters- Step 2(iii) Step 3: (i.) Click Solve and a dialog will pop up as Figure 3.4. Figure 3.4: Estimation of unknown parameters- Step 3(i) (ii.) Click all the Report and click Ok. (iii.) All the report will show in other 3 new sheets. 3.6 Forecast Performance When reach the forecasting step, we would compare RMSE, values from NGARCH (1, 1) and GARCH(1,1) models. RMSE is the short form of Root Mean Square Error which is generating from Mean Square Error (MSE). RMSE which has the same units as the quantity being estimated; for an unbiased estimator, the RMSE is the square root of the variance, known as the standard error. In 26 statistical modelling, the RMSE is a measure of the differences between values predicted by a model or an estimator and the values actually observed from the thing being modeled or estimated. RMSE is a alternative form for the standard deviation, calculate using the formula below (Equation 3.21). ∑[.e ' ef abc= d % where e the actual value of the data ef = the predicted value of the data % number of observation Obviously, in a forecast time series, the smaller forecast performances are preferred. A model which has a small RMSE is said that is a better model then others. Chapter IV RESULT AND DISCUSSION 4.0 Introduction This chapter contains a result and discussion of the study. We will firstly discuss the results by application of Microsoft Office Excel 2007 for GARCH and NGARCH model in Section 4.1. In Section 4.2, we will discuss the results of comparison between GARCH and NGARCH model and time series methods are presented. 4.1 The Exchange Rate Series The in-sample period of exchange rate data from 3rd May 2007 to 30th April 2009 will be plotted with Microsoft Office Excel. The exchange rate series is shown in the figure below 28 Selling prices of RM/USD from 03/05/07 - 29/05/09 3.8 3.7 3.6 3.5 3.4 3.3 3.2 Selling prices of RM/USD 3.1 3 2.9 3/5/2007 1/6/2007 2/7/2007 31/7/2007 29/8/2007 28/9/2007 30/10/2007 29/11/2007 2/1/2008 5/2/2008 7/3/2008 8/4/2008 8/5/2008 9/6/2008 8/7/2008 6/8/2008 5/9/2008 8/10/2008 7/11/2008 9/12/2008 12/1/2009 16/2/2009 18/3/2009 16/4/2009 2.8 Figure 4.1: The dynamics of the RM/USD exchange rate selling prices in the period 2007-2009 In figure 4.1, the time series graph has been plotted from 3rd January 2007 to 30th April 2009. The exchange rate for RM/USD selling prices trend corresponding to the analyzed period and indicates that the selling prices of RM/USD have mainly fluctuated in the range of about RM3.4 to RM 3.7. During the period in end of the year 2007 until 29 July 2008, there occurs a breakdown. The selling prices of RM/USD are around RM3.1 to RM3.3 when the globalization economic crisis. After July 2008, the selling prices of RM/USD fluctuations are increase with quite equilibration that is around RM3.4 to RM 3.7. Next, we plot the daily log returns prices of the RM/USD exchange rate. Daily log returns for the RM/USD exchange rate 0.02000000 0.01500000 0.01000000 0.00500000 log returns for RM vs USD 0.00000000 -0.00500000 -0.01000000 -0.01500000 3/5/2007 6/6/2007 10/7/2007 13/8/2007 17/9/2007 22/10/2007 26/11/2007 2/1/2008 12/2/2008 17/3/2008 21/4/2008 27/5/2008 30/6/2008 1/8/2008 5/9/2008 13/10/2008 17/11/2008 22/12/2008 30/1/2009 10/3/2009 13/4/2009 -0.02000000 Figure 4.2: Daily log returns for the RM/USD exchange rate 30 Figure 4.2 show the returns on the RM/USD exchange rate with the average volatility. In Figure 4.2, it shows the difference with respect to first order for exchange rate selling prices series. The difference series is stationary because most of the price values are located around mean of zero. But, there are some spikes found in the figure that representing those high volatility periods. Squared innovation return in the period 2007-2009 0.00035 0.00030 0.00025 Squared innovation return 0.00020 0.00015 0.00010 0.00005 3/5/2007 6/6/2007 10/7/2007 13/8/2007 17/9/2007 22/10/2007 26/11/2007 2/1/2008 12/2/2008 17/3/2008 21/4/2008 27/5/2008 30/6/2008 1/8/2008 5/9/2008 13/10/2008 17/11/2008 22/12/2008 30/1/2009 10/3/2009 13/4/2009 0.00000 Figure 4.3: The squared innovation return in the period 2007-2009 31 Figure 4.3 describes the changes in the RM/USD exchange rate prices in the period from 2007-2009. From the figure, it can be immediately noticed that some periods have very different volatilities, which can also seen from the graph of the squared innovation returns. 4.2 Stationary Series Hereby, we provide the graph plot to show the stationary of this data series. First lagged difference from the original data time series which has giving the stationary series is shown in the figure below. First Order Difference Selling Prises 0.08 0.06 0.04 0.02 0 First Order Difference Selling Prises -0.02 -0.04 3/5/2007 1/6/2007 2/7/2007 31/7/2007 29/8/2007 28/9/2007 30/10/2007 29/11/2007 2/1/2008 5/2/2008 7/3/2008 8/4/2008 8/5/2008 9/6/2008 8/7/2008 6/8/2008 5/9/2008 8/10/2008 7/11/2008 9/12/2008 12/1/2009 16/2/2009 18/3/2009 16/4/2009 -0.06 Figure 4.4: First Order Difference Selling prices of RM/USD. 32 In Figure 4.4, it shows the difference with respect to first order for selling prices series. The difference series is stationary because most of the price values are located around mean of zero. But, there are some spikes found in the figure that representing those high volatility periods. Let us check with its normality distribution test statistics including mean, median, maximum and minimum values, standard deviation, skewness, kurtosis, range, confidence interval of first order difference exchange rate selling prices series as Table 4.1. Table 4.1: Data analysis of first order difference exchange rate selling prices series Data Analysis Mean Standard Error Median Mode Standard Deviation Sample Variance Range Minimum Maximum Sum Count Largest(1) Smallest(1) Confidence Level(95.0%) 0.0003 0.0723 0.0005 0.0000 1.0142 0.0002 0.1040 -0.0485 0.0555 0.1400 493.0000 0.0555 -0.0485 0.0013 From Table 4.1, the mean value is 0.0003 and standard deviation 1.0172 which gives a standard normal distribution of 0,1. The median is equal to 0 because most of the values after first lagged difference fall within the intervals of zero. The skewness and kurtosis values are -0.0455 and 1.1584respectively which shown the distribution is slightly asymmetric and highly leptokurtosis. 33 4.3 Estimation 4.3.1 Parameters Estimation of NGARCH (1, 1) In this section, estimation of the parameters in NGARCH (1,1) and GARCH (1,1) model will be discussed. In this study, we have five unknown parameter need to estimate which are , , , and . The parameter estimation of the model will be done by Microsoft Office Excel 2007 software. Firstly, set the unknown parameters for NGARCH (1,1) model as follow: Table 4.2: Starting value of coefficient Parameters of NGARCH(1,1) Starting value ω 6.42866 x10^-07 α 0.05000000 β 0.90000000 ρ 0.50000000 λ 0.50000000 α(1+ρ2)+β 0.96250000 MLE 1928.63450700 After define the unknown parameters, by using Solver in Microsoft Office excel the five unknown parameters was found. The result was shown as Table 4.3. 34 Table 4.3: Value of parameters coefficient for NGARCH(1, 1) NGARCH (1,1) ω 6.42866 x10-07 α 0.05000000 β 0.90000000 ρ 0.50000000 λ 0.50000000 α(1+ρ2)+β 0.96250000 MLE 1928.63450700 In Table 4.3, it shown that the values of unknown parameters are unchanged after applies by Solver. From the Figure 4.4, it was shown that the MLE =1928.63450700, 6.42866 m 10n , 0.0500, 0.9000, 0.5000, 0.5000 and 1 0.9625. All the values we get are under condition as Equation (3.14). The expression 1 is called persistence of the NGARCH (1,1) model. From expression it follows that the shocks in the market persist through a long time which define as G H ∞ since the value of 1 is near 1. In this case, we will say that the time series has a long memory. 35 Figure 4.5: Answer Report from Solver for NGARCH(1,1) From the obtained result, the optimal parameters imply the following variance dynamics: 6.42866 x10n 0.5 ' 0.5 0.9 36 The NGARCH (1, 1) model can be written into conditional mean and conditional variance as 1 = > ? 0.0131% ' 0.0115% 0.5 ' 2 @A > ? The following figures present the graphical plot for conditional variance and conditional standard deviation. The values for conditional standard deviation are taking square roots from conditional variance. Conditional Variance of selling prices RM/USD 0.000070 0.000060 0.000050 0.000040 0.000030 conditional variance prices 0.000020 0.000010 3/5/2007 7/6/2007 12/7/2007 16/8/2007 21/9/2007 29/10/2007 4/12/2007 14/1/2008 22/2/2008 31/3/2008 6/5/2008 11/6/2008 16/7/2008 20/8/2008 25/9/2008 4/11/2008 10/12/2008 19/1/2009 27/2/2009 6/4/2009 0.000000 Figure 4.6: Conditional Variance for NGARCH(1,1) model 37 In Figure 4.5, conditional variance graph is plotted. There are lesser spikes compared to conditional standard deviation graph but it does point out some of the high volatile clusters in the series. However, the extraordinary long spikes are the high volatile periods in the data series. Conditional Standard Deviation for selling pricess RM/USD 0.009 0.008 0.007 0.006 0.005 0.004 Conditional Standard Deviation 0.003 0.002 0.001 3/5/2007 1/6/2007 2/7/2007 31/7/2007 29/8/2007 28/9/2007 30/10/2007 29/11/2007 2/1/2008 5/2/2008 7/3/2008 8/4/2008 8/5/2008 9/6/2008 8/7/2008 6/8/2008 5/9/2008 8/10/2008 7/11/2008 9/12/2008 12/1/2009 16/2/2009 18/3/2009 16/4/2009 0 Figure 4.7: Conditional standard deviation for NGARCH(1,1) model A conditional standard deviation graph is drawn by Microsoft Office Excel shown in Figure 4.6. The extraordinary long spikes are the high volatile periods of the series. With NGARCH(1,1) model, the volatility clustering will be detected. 38 After we estimate the parameters, next step will be diagnostic checking on the adequacy for NGARCH (1, 1) model. It can be done through when checking on the autocorrelation and partial autocorrelation of standardized return approximately zero. Table 4.4: Standardized Return for NGARCH (1, 1) Model ACF Standardized return Lag 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 0.140728 0.037413 0.056746 0.14629 0.043647 0.044736 -0.004524 0.082708 0.015038 0.048333 -0.019805 0.019213 -0.042582 0.039118 -0.005254 -0.022726 -0.032651 -0.002665 0.132111 -0.032387 PACF Standardized return 0.140728 0.017965 0.050054 0.134068 0.004114 0.03158 -0.028206 0.067415 -0.013301 0.038724 -0.034765 0.005673 -0.052113 0.039027 -0.007224 -0.029667 -0.013452 -0.012182 0.152998 -0.076134 t-Statistics 3.12468 0.81473 1.23407 3.17162 0.92754 0.94904 -0.09579 1.75135 0.31646 1.01695 -0.41583 0.40326 -0.89345 0.81945 -0.10992 -0.47542 -0.68271 -0.05567 2.75975 -0.66634 Q-Statistics 9.8232 10.5189 12.1226 22.8026 23.7552 24.7581 24.7684 28.2103 28.3243 29.5047 29.7033 29.8905 30.8124 31.592 31.6061 31.8703 32.4169 32.4205 41.4062 41.9474 In the Table4.4, ACF and PACF of square return are approximately zero. The Ljung-Box q-statistic also provide the similar evidence with r-value that NGARCH(1,1) model is adequate. Once again, we can conclude that the model is adequate. 39 4.3.2 Parameters Estimation of GARCH (1, 1) For the GARCH (1,1) model, as the same step as NGARCH (1,1) model, the starting value was set in Microsoft Excel 2007 as Table 4.4. Table 4.5 Starting values of parameters coefficient for GARCH model Starting value ω 0.001144629 α 0.05000000 β 0.90000000 ρ 0.50000000 λ 0.50000000 α+β 0.95000000 MLE 1928.63450700 Refer to Table 4.5, we have generated the coefficient parameter of GARCH (1, 1) model by Microsoft Office Excel software. The estimated parameter coefficients by GARCH (1, 1) model gives 9.59927 m 10n , 0.049997, 0.8999, 0.5, 0.4999, and 0.95. The coefficients which estimate by using Solver are all under condition as shown at Equation (3.5). From expression it follows that the shocks in the market persist through a long time which defines as G H ∞ since the value of is near 1. In this case, we will say that the time series has a long memory. 40 Table 4.6: The coefficient parameters of GARCH (1,1) GARCH(1,1) ω 9.59927 m 10n α 0.049997095 β 0.899985239 ρ 0.5 λ 0.499998103 α+β MLE 0.95000006 1824.590093 From the obtained result, the optimal parameters imply the following variance dynamics: 9.59927 m 10'7 0.049997095 0.899985239 The GARCH (1, 1) model can be written into conditional mean and conditional variance as 1 = > ? 0.0131% ' 0.0115% 0.5 ' 2 @A > ? 41 Figure 4.8: Answer Report from Solver for GARCH(1,1) The following figures present the graphical plot for conditional variance and conditional standard deviation. The values for conditional standard deviation are taking square roots from conditional variance. 42 Conditional Variance of Selling Prices RM/USD 0.0200000000 0.0150000000 0.0100000000 0.0050000000 0.0000000000 -0.0050000000 3/5/2007 30/5/2007 26/6/2007 23/7/2007 17/8/2007 14/9/2007 11/10/2007 9/11/2007 6/12/2007 7/1/2008 6/2/2008 6/3/2008 3/4/2008 30/4/2008 29/5/2008 25/6/2008 22/7/2008 18/8/2008 15/9/2008 14/10/2008 11/11/2008 9/12/2008 8/1/2009 10/2/2009 10/3/2009 6/4/2009 -0.0100000000 coditional variance Figure 4.9: Conditional Variance for GARCH(1,1) model In Figure 4.8, conditional variance graph is plotted. There is having a consistent pattern of the graph. All the point is lay around the zero. However, the extraordinary long spikes are the high volatile periods in the data series. Next, we will plot a conditional standard deviation graph as in Figure 4.9. The extraordinary long spikes are the high volatile periods of the series. With GARCH(1,1) model, the volatility clustering will be detected. 43 Conditional Standard Deviation of Selling Prices RM/USD 0.16 0.14 0.12 0.1 0.08 0.06 0.04 0.02 3/5/2007 25/5/2007 18/6/2007 10/7/2007 1/8/2007 23/8/2007 17/9/2007 9/10/2007 1/11/2007 26/11/2007 18/12/2007 15/1/2008 12/2/2008 5/3/2008 28/3/2008 21/4/2008 14/5/2008 6/6/2008 30/6/2008 22/7/2008 13/8/2008 5/9/2008 29/9/2008 23/10/2008 17/11/2008 10/12/2008 6/1/2009 30/1/2009 25/2/2009 20/3/2009 13/4/2009 0 Conditional Standard Deviation Figure 4.10: Conditional Standard Deviation for GARCH(1,1) model After we estimate the parameters, next step will be diagnostic checking on the adequacy for GARCH (1, 1) model. It can be done through when checking on the autocorrelation and partial autocorrelation of standardized return approximately zero. 44 Table 4.7: Standardized Return for GARCH (1, 1) Model Lag 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 ACF PACF t-Statistics Q-Statistics Standardized Standardized Return Return 0.126154 0.126154 2.80108 7.8939 0.067216 0.052131 1.46924 10.1394 0.061714 0.047992 1.34311 12.0363 0.164783 0.151545 3.57318 25.5872 0.03369 -0.008346 0.71233 26.1547 0.057214 0.037489 1.20846 27.795 -0.00347 -0.0304 -0.07307 27.8011 0.097608 0.075134 2.05556 32.5949 -0.00794 -0.034447 -0.16579 32.6267 0.020398 0.006077 0.42591 32.837 -0.016821 -0.020946 -0.35109 32.9802 0.082243 0.063771 1.71616 36.4119 -0.045604 -0.058119 -0.94599 37.4692 0.035716 0.035559 0.73953 38.1191 -0.005767 -0.005564 -0.11928 38.1361 0.000185 -0.026913 0.00383 38.1361 -0.040096 -0.021409 -0.82928 38.9603 -0.021521 -0.032311 -0.44448 39.1983 0.122863 0.1547 2.53655 46.97 -0.028929 -0.079511 -0.5896 47.4018 In the Table4.6, ACF and PACF of square return are approximately zero. The Ljung-Box q-statistic also provide the similar evidence with r-value that GARCH(1,1) model is adequate. Once again, we can conclude that the model is adequate. 4.4 Forecasting 45 4.4.1 Forecasting future value by using NGARCH (1,1) Future values of selling prices of RM/USD forecast by NGARCH (1, 1) model using Microsoft Office Excel. The duration of forecast is started from 4th May 2009 to 29th May 2009. The value of forecast is according to static forecasting approach which performs the series of one-step ahead forecast of exchange rate selling prices of RM/USD. Table 4.6 was shown that the real value against forecast value by using NGARCH (1, 1) model from 04th May 2009 to 29th May 2009. Refer to Table 4.6, we can saw that the forecasted value by using Microsoft Office Excel 2007 are quite near with the real value. It was shown that the accuracy of NGARCH (1,1) model. Table 4.8: Comparison the real value against forecast value by using NGARCH (1,1) Real Value 4/5/2009 5/5/2009 6/5/2009 7/5/2009 8/5/2009 11/5/2009 12/5/2009 13/5/2009 14/5/2009 15/5/2009 18/5/2009 19/5/2009 20/5/2009 21/5/2009 22/5/2009 25/5/2009 26/5/2009 27/5/2009 28/5/2009 29/5/2009 Forecast value 3.5315 3.52 3.543 3.5225 3.535 3.503 3.525 3.513 3.555 3.545 3.57 3.543 3.5495 3.53 3.499 3.49 3.494 3.491 3.525 3.51 3.577201432 3.544552717 3.532512534 3.555118745 3.534100038 3.546214326 3.513711947 3.535397216 3.523001541 3.564776342 3.554423973 3.579181541 3.551822906 3.558066013 3.538263211 3.506952152 3.497708115 3.50150677 3.49830334 3.532188084 46 4.4.2 Forecasting future value by using GARCH (1,1) The duration of forecast is started from 4th May 2009 to 29th May 2009. Future values of selling prices of RM/USD forecast by GARCH (1, 1) model by using Microsoft Office Excel 2007. The value of forecast is according to static forecasting approach which performs the series of one-step ahead forecast of exchange rate selling prices of RM/USD. Table 4.7 was shown that the real value against forecast value by using GARCH (1, 1) model from 04th May 2009 to 29th May 2009. Refer to Table 4.7, we can saw that the forecasted value by using Microsoft Office Excel 2007 are different with the real value. It was shown the inaccuracy by using the GARCH (1,1) model. Next, the actual and forecast daily exchange rates of RM/USD by selling prices by GARCH (1, 1) model are being plotted Table 4.9: Comparison the real value against forecast value by using GARCH (1,1) Real Value 4/5/2009 5/5/2009 6/5/2009 7/5/2009 8/5/2009 11/5/2009 12/5/2009 13/5/2009 14/5/2009 15/5/2009 18/5/2009 19/5/2009 20/5/2009 21/5/2009 22/5/2009 Forecast value 3.5315 3.52 3.543 3.5225 3.535 3.503 3.525 3.513 3.555 3.545 3.57 3.543 3.5495 3.53 3.499 3.84689862 3.68975006 3.67806037 3.82447825 3.70173947 3.71452066 3.78165449 3.71898288 3.70666294 3.83559469 3.75083626 3.77658264 3.82234327 3.7645709 3.74443873 47 25/5/2009 26/5/2009 27/5/2009 28/5/2009 29/5/2009 3.77554286 3.71052522 3.71466336 3.76662235 3.75299947 3.49 3.494 3.491 3.525 3.51 3.9 3.8 3.7 3.6 real value NGARCH modle 3.5 GARCH value 3.4 4/5/2009 5/5/2009 6/5/2009 7/5/2009 8/5/2009 11/5/2009 12/5/2009 13/5/2009 14/5/2009 15/5/2009 18/5/2009 19/5/2009 20/5/2009 21/5/2009 22/5/2009 25/5/2009 26/5/2009 27/5/2009 28/5/2009 29/5/2009 3.3 Figure 4.11: The plot of actual prices against forecast prices by NGARCH(1,1) and GARCH(1,1) model In Figure 4.11, the actual and forecast daily exchange rates of RM/USD by selling prices by NGARCH (1, 1) and GARCH(1,1) model are being plotted. The trend of forecast prices has followed tight to the actual selling prices of RM/USD for 1 month out-sample period. The trend of forecast prices by using GARCH(1,1) model has far away to the actual selling prices of RM/USD for 1 month out-sample period. 48 4.5 Comparison of NGARCH (1,1) and GARCH(1,1) Model One of the study objectives is to compare the forecast performances by two univariate time series models. We will compare the NGARCH (1, 1) and GARCH (1,1) models in terms of their RMSE values in forecast performances in forecasting stage. 4.5.1 Comparison of Forecasting Model When reach the forecasting step, we would compare RMSE, values from NGARCH(1, 1) and GARCH(1,1) models. Obviously, the smaller forecast performances are preferred. When the relation of actual values and forecast values are closer to each others, a smaller forecast performance will be obtained. Table 4.10: RMSE value of NGARCH(1,1) Real Value Forecast NGARCH 3.5315 3.507201432 3.52 3.504552717 3.543 3.502512534 3.5225 3.505118745 3.535 3.544100038 3.503 3.516214326 3.525 3.503711947 3.513 3.555397216 3.555 3.503001541 3.545 3.534776342 3.57 3.554423973 3.543 3.519181541 3.5495 3.521822906 3.53 3.518066013 3.499 3.498263211 3.49 3.496952152 3.494 3.497708115 3.491 3.49150677 Error 0.024298568 0.015447283 0.040487466 0.017381255 0.009100038 0.013214326 0.021288053 0.042397216 0.051998459 0.010223658 0.015576027 0.023818459 0.027677094 0.011933987 0.000736789 0.006952152 0.003708115 0.00050677 Error Square 0.00059042 0.000238619 0.001639235 0.000302108 8.28E-05 0.000174618 0.000453181 0.001797524 0.00270384 0.000104523 0.000242613 0.000567319 0.000766022 0.00014242 5.43E-07 4.83E-05 1.38E-05 2.57E-07 49 3.51830334 3.532188084 3.525 3.51 Average RMSE 0.00669666 0.022188084 0.018281523 4.48E-05 0.000492311 0.000520265 0.022809308 Table 4.11: RMSE value of GARCH(1,1) Real Value Forecast value Error 3.5315 3.84689862 3.52 3.68975006 3.543 3.67806037 3.5225 3.82447825 3.535 3.70173947 3.503 3.71452066 3.525 3.78165449 3.513 3.71898288 3.555 3.70666294 3.545 3.83559469 3.57 3.75083626 3.543 3.77658264 3.5495 3.82234327 3.53 3.7645709 3.499 3.74443873 3.49 3.77554286 3.494 3.71052522 3.491 3.71466336 3.525 3.76662235 3.51 3.75299947 Average RMSE 0.31539862 0.16975006 0.13506037 0.30197825 0.16673947 0.21152066 0.25665449 0.20598288 0.15166294 0.29059469 0.18083626 0.23358264 0.27284327 0.2345709 0.24543873 0.28554286 0.21652522 0.22366336 0.24162235 0.24299947 0.22914837 Error Square 0.09947629 0.02881508 0.0182413 0.09119086 0.02780205 0.04474099 0.06587153 0.04242895 0.02300165 0.08444527 0.03270175 0.05456085 0.07444345 0.05502351 0.06024017 0.08153472 0.04688317 0.0500253 0.05838136 0.05904874 0.05494285 0.23439891 Table 4.10 and 4.11 shown the value of RMSE by using NGARCH(1,1) and GARCH(1,1). From the table, value of RMSE of NGARCH(1,1) is 0.022809308 while value of GARCH(1,1) is 0.23439891. From this two value, we can noticed that the value of NGARCH(1,1) is smaller than GARCH(1,1). Therefore, we conclude that NGARCH(1,1) model is a better forecast model for daily exchange rate selling prices for RM/USD than GARCH (1, 1) model. Chapter V CONCLUSION 5.0 Introduction This chapter will summarize the study and conclusions are made based on the analysis and the results of the forecasting in section 5.1. In Section 5.2, we will give suggestions for further study based on this study. 51 5.1 Result and Comparisons Table 5.1:The plot of actual prices against forecast prices by NGARCH(1,1) and GARCH(1,1) model Real Value Forecast value Forecast value NGARCH (1,1) GARCH (1,1) 4/5/2009 3.5315 3.577201432 0.00383756 5/5/2009 3.52 3.544552717 0.00368927 6/5/2009 3.543 3.532512534 0.00354839 7/5/2009 3.5225 3.555118745 0.00341464 8/5/2009 3.535 3.534100038 0.00328772 11/5/2009 3.503 3.546214326 0.00316735 12/5/2009 3.525 3.513711947 0.00305327 13/5/2009 3.513 3.535397216 0.00294522 14/5/2009 3.555 3.523001541 0.00284296 15/5/2009 3.545 3.564776342 0.00274625 18/5/2009 3.57 3.554423973 0.00265486 19/5/2009 3.543 3.579181541 0.00256856 20/5/2009 3.5495 3.551822906 0.00248714 21/5/2009 3.53 3.558066013 0.00241039 22/5/2009 3.499 3.538263211 0.00233812 25/5/2009 3.49 3.506952152 0.00227011 26/5/2009 3.494 3.497708115 0.00220619 27/5/2009 3.491 3.50150677 0.00214617 28/5/2009 3.525 3.49830334 0.00208986 29/5/2009 3.51 3.532188084 0.00203710 52 Refer to the Table 5.1, the prediction by using time series method which is NGARCH (1, 1) followed tight to the actual selling prices of RM/USD for 1 month outsample period. Whereas, GARCH (1,1) is far away to the actual selling prices. GARCH(1,1) is a model able to capture the volatility y the conditional variance of being non-constant. Whereas, Nonlinear GARCH (NGARCH) is a negative return increases variance by more than a positive return of the same magnitude is the better estimate and forecast model. 5.2 Suggestions for Further Research This study have explored two approaches that were the NGARCH (1, 1) and GARCH (1, 1) in forecasting the Foreign Exchange for the selling prices of RM/USD for daily data from 03rd May 2007 to 29th May 2009. NGARCH is a model which modify from GARCH model. It know as a GARCH model with leverage effect which known as negative return increases variance by more than a positive return of the same magnitude. NGARCH also known as a alternative to the Black Scholes Model. (Petra POSEDEL, 2006). Further research can be done by using this research as a benchmark to forecast the future values of the Foreign Exchange for the currency selling prices of RM/USD for daily data. Complex method that being done to forecast the future values can be compared to the result of this stud y so that the better method in forecasting can be identify. 53 REFERENCE List of reference Andersen, Torben and Tim Bollerslev. 1998. “Answering the skeptics: Yes, standard volatility models do provide accurate forecasts.” International Economic Review 39: 885-905. Baillie, R. and T. Bollerslev. 1989. “The Message in Daily Exchange Rates: A Conditional- Variance Tale.” Journal of Business and Economic Statistics 7: 297-305. Bera, A. K., and Higgins, M. L. 1993, “ARCH Model: Properties, Estimation and Testing,” Journal of Economic Surveys, Vol. 7, No. 4, 307-366. Baillie, R. and T. Bollerslev. 1991. “Intra-Day and Inter-Market Volatility in Foreign Exchange Rates.” Review of Economic Studies 58: 565-585. Bates, D., 1996. Jumps and stochastic volatility: Exchange rate process implicit in Deutsche mark options. Review of Financial Studies, 9, 69-107 Bollerslev, T., 1986. Generalized autoregressive conditional heteroskedasticity. Journal of Econometrics 31, 307-327.Ding, Z., Granger, C. W. J., & Engle, R. F. 1993. A long memory property of stock market returns and a new model. Journal of Empirical Finance, 1, 83-106. Duan, J.-C., 1995. The GARCH option pricing model. Mathematical Finance 5, 13-32. 54 Duan, J.-C., 1997. Augmented GARCH(p, q) process and its diffusion limit. Journal of Econometrics79, 97-127. Engle, R. Ng, V., 1993. Measuring and testing of the impact of reason volatility. Journal of Finance 48, 1749-1778. Engle, R., 1982. Autoregressive conditional heteroscedasticity with estimates of the variance of UK inflation. Econometrica 50, 987-1108. Glosten, L. R., Jagannathan, R., & Runkle, D.,1993. On the relation between the expected value and the volatility of the nominal excess return on stocks. Journal of Finance, 48, 1779-1801. Hentschel, L., 1995. All in the family: Nesting symmetric and asymmetric GARCH models. Journal of Financial Economics, 39, 71-104. Kim, I. J., & Kim, S., 2004. Empirical comparison of alternative stochastic volatility option pricing models: Evidence from Korean KOSPI 200 index options market. Pacific-Basin Finance Journal, 12, 117–142. Nelson, D. (1991). Conditional heteroskedasticity in asset returns: A new approach. Econometria, 59, 347–370. Nelson, D. B., 1991. Conditional heteroscedasticity in assets returns: A new approach. Econometrica, 55, 703-708 Petra POSEDEL, 2006. Analysis of the exchange rate and pricing foreign currency options on the Croatian Market: The NGARCH model as alternative to the Blacks Scholes Model, Financial Theory and Practice 30 (4), 347-368 Taylor, S., 1986. Modelling financial time series. New York: John Wiley & Sons. Watanabe, T., 1999. A non-linear filtering approach to stochastic volatility models with an application to daily stock returns. Journal of Applied Econometrics, 14, 101– 121. 55 West, Kenneth D. and Dongchul Cho. 1995. “The Predictive Ability of Several Models of Exchange Rate Volatility.” Journal of Econometrics 69(2): 367-91. 56 APPENDIX A Foreign Exchange Daily Data for the selling prices of RM/USD. Ringgit/USD Date 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 3/5/2007 4/5/2007 7/5/2007 8/5/2007 9/5/2007 10/5/2007 11/5/2007 14/5/2007 15/5/2007 16/5/2007 17/5/2007 18/5/2007 21/5/2007 22/5/2007 23/5/2007 24/5/2007 25/5/2007 28/5/2007 29/5/2007 30/5/2007 31/5/2007 1/6/2007 4/6/2007 5/6/2007 6/6/2007 7/6/2007 8/6/2007 11/6/2007 12/6/2007 13/6/2007 14/6/2007 15/6/2007 US Dollar Selling Prices 3.4235 3.421 3.42 3.4155 3.4105 3.4025 3.409 3.403 3.4033 3.4025 3.4015 3.403 3.393 3.3935 3.39 3.387 3.392 3.3895 3.3875 3.4005 3.406 3.3995 3.3965 3.401 3.4185 3.436 3.455 3.4595 3.4615 3.472 3.465 3.461 57 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 18/6/2007 19/6/2007 20/6/2007 21/6/2007 22/6/2007 25/6/2007 26/6/2007 27/6/2007 28/6/2007 29/6/2007 2/7/2007 3/7/2007 4/7/2007 5/7/2007 6/7/2007 9/7/2007 10/7/2007 11/7/2007 12/7/2007 13/7/2007 16/7/2007 17/7/2007 18/7/2007 19/7/2007 20/7/2007 23/7/2007 24/7/2007 25/7/2007 26/7/2007 27/7/2007 30/7/2007 31/7/2007 1/8/2007 2/8/2007 3/8/2007 6/8/2007 7/8/2007 8/8/2007 9/8/2007 10/8/2007 13/8/2007 3.4385 3.435 3.434 3.449 3.4535 3.4625 3.4665 3.49 3.472 3.4565 3.447 3.4425 3.453 3.4575 3.453 3.441 3.435 3.4515 3.456 3.4475 3.446 3.447 3.4525 3.4445 3.429 3.419 3.404 3.418 3.4305 3.4645 3.47 3.4555 3.473 3.476 3.4655 3.475 3.4665 3.4645 3.4555 3.478 3.4765 58 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 14/8/2007 15/8/2007 16/8/2007 17/8/2007 20/8/2007 21/8/2007 22/8/2007 23/8/2007 24/8/2007 27/8/2007 28/8/2007 29/8/2007 30/8/2007 3/9/2007 4/9/2007 5/9/2007 6/9/2007 7/9/2007 10/9/2007 11/9/2007 12/9/2007 13/9/2007 14/9/2007 17/9/2007 18/9/2007 19/9/2007 20/9/2007 21/9/2007 24/9/2007 25/9/2007 26/9/2007 27/9/2007 28/9/2007 1/10/2007 2/10/2007 3/10/2007 4/10/2007 5/10/2007 8/10/2007 9/10/2007 10/10/2007 3.4775 3.493 3.509 3.5205 3.5025 3.4975 3.5005 3.47 3.492 3.48 3.4985 3.506 3.5055 3.499 3.506 3.5085 3.5105 3.5035 3.517 3.5185 3.5065 3.492 3.4815 3.4865 3.493 3.457 3.4475 3.4465 3.434 3.437 3.4275 3.421 3.4185 3.397 3.4 3.4105 3.424 3.407 3.39 3.403 3.388 59 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 11/10/2007 12/10/2007 16/10/2007 17/10/2007 18/10/2007 19/10/2007 22/10/2007 23/10/2007 24/10/2007 25/10/2007 26/10/2007 29/10/2007 30/10/2007 31/10/2007 1/11/2007 2/11/2007 5/11/2007 6/11/2007 7/11/2007 9/11/2007 12/11/2007 13/11/2007 14/11/2007 15/11/2007 16/11/2007 19/11/2007 20/11/2007 21/11/2007 22/11/2007 23/11/2007 26/11/2007 27/11/2007 28/11/2007 29/11/2007 30/11/2007 3/12/2007 4/12/2007 5/12/2007 6/12/2007 7/12/2007 10/12/2007 3.379 3.3735 3.387 3.388 3.374 3.364 3.3795 3.38 3.3715 3.3635 3.351 3.3425 3.349 3.3435 3.332 3.3455 3.3495 3.354 3.3335 3.323 3.341 3.354 3.343 3.363 3.3845 3.368 3.3815 3.369 3.384 3.376 3.357 3.37 3.377 3.373 3.36 3.363 3.3545 3.3445 3.3445 3.333 3.326 60 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 11/12/2007 12/12/2007 13/12/2007 14/12/2007 17/12/2007 18/12/2007 19/12/2007 21/12/2007 24/12/2007 26/12/2007 27/12/2007 28/12/2007 31/12/2007 2/1/2008 3/1/2008 4/1/2008 7/1/2008 8/1/2008 9/1/2008 11/1/2008 14/1/2008 15/1/2008 16/1/2008 17/1/2008 18/1/2008 21/1/2008 22/1/2008 24/1/2008 25/1/2008 28/1/2008 29/1/2008 30/1/2008 31/1/2008 4/2/2008 5/2/2008 6/2/2008 11/2/2008 12/2/2008 13/2/2008 14/2/2008 15/2/2008 3.3165 3.326 3.313 3.3175 3.3325 3.351 3.35 3.352 3.34 3.342 3.341 3.326 3.308 3.3105 3.308 3.285 3.285 3.274 3.274 3.258 3.2595 3.258 3.2595 3.283 3.2835 3.274 3.302 3.2705 3.244 3.245 3.238 3.241 3.237 3.231 3.2335 3.233 3.239 3.2455 3.236 3.2395 3.232 61 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 18/2/2008 19/2/2008 20/2/2008 21/2/2008 22/2/2008 25/2/2008 26/2/2008 27/2/2008 28/2/2008 29/2/2008 3/3/2008 4/3/2008 5/3/2008 6/3/2008 7/3/2008 10/3/2008 11/3/2008 12/3/2008 13/3/2008 14/3/2008 17/3/2008 18/3/2008 19/3/2008 21/3/2008 24/3/2008 25/3/2008 26/3/2008 27/3/2008 28/3/2008 31/3/2008 1/4/2008 2/4/2008 3/4/2008 4/4/2008 7/4/2008 8/4/2008 9/4/2008 10/4/2008 11/4/2008 14/4/2008 15/4/2008 3.222 3.222 3.222 3.222 3.2161 3.2175 3.215 3.2045 3.202 3.19 3.2015 3.192 3.183 3.17 3.175 3.204 3.203 3.185 3.164 3.171 3.192 3.182 3.1685 3.1955 3.201 3.2015 3.182 3.1975 3.21 3.189 3.182 3.189 3.191 3.1945 3.189 3.188 3.185 3.154 3.154 3.163 3.1635 62 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 16/4/2008 17/4/2008 18/4/2008 21/4/2008 22/4/2008 23/4/2008 24/4/2008 25/4/2008 28/4/2008 29/4/2008 30/4/2008 2/5/2008 5/5/2008 6/5/2008 7/5/2008 8/5/2008 9/5/2008 12/5/2008 13/5/2008 14/5/2008 15/5/2008 16/5/2008 20/5/2008 21/5/2008 22/5/2008 23/5/2008 26/5/2008 27/5/2008 28/5/2008 29/5/2008 30/5/2008 2/6/2008 3/6/2008 4/6/2008 5/6/2008 6/6/2008 9/6/2008 10/6/2008 11/6/2008 12/6/2008 13/6/2008 3.1645 3.152 3.1485 3.1455 3.143 3.1325 3.1365 3.151 3.1565 3.155 3.1595 3.163 3.165 3.149 3.1535 3.209 3.21 3.2175 3.22 3.253 3.275 3.255 3.255 3.231 3.213 3.218 3.224 3.25 3.245 3.2405 3.245 3.228 3.227 3.2345 3.262 3.26 3.265 3.271 3.273 3.2755 3.2755 63 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 16/6/2008 17/6/2008 18/6/2008 19/6/2008 20/6/2008 23/6/2008 24/6/2008 25/6/2008 26/6/2008 27/6/2008 30/6/2008 1/7/2008 2/7/2008 3/7/2008 4/7/2008 7/7/2008 8/7/2008 9/7/2008 10/7/2008 11/7/2008 14/7/2008 15/7/2008 16/7/2008 17/7/2008 18/7/2008 21/7/2008 22/7/2008 23/7/2008 24/7/2008 25/7/2008 28/7/2008 29/7/2008 30/7/2008 31/7/2008 1/8/2008 4/8/2008 5/8/2008 6/8/2008 7/8/2008 8/8/2008 11/8/2008 3.2755 3.257 3.244 3.262 3.26 3.2655 3.267 3.2655 3.258 3.2665 3.268 3.267 3.275 3.267 3.269 3.267 3.261 3.254 3.2445 3.248 3.237 3.216 3.225 3.23 3.244 3.238 3.243 3.243 3.245 3.253 3.2656 3.26 3.264 3.264 3.263 3.267 3.27 3.278 3.285 3.298 3.3185 64 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 12/8/2008 13/8/2008 14/8/2008 15/8/2008 18/8/2008 19/8/2008 20/8/2008 21/8/2008 22/8/2008 25/8/2008 26/8/2008 27/8/2008 28/8/2008 29/8/2008 2/9/2008 3/9/2008 4/9/2008 5/9/2008 8/9/2008 9/9/2008 10/9/2008 11/9/2008 12/9/2008 15/9/2008 16/9/2008 17/9/2008 18/9/2008 19/9/2008 22/9/2008 23/9/2008 24/9/2008 25/9/2008 26/9/2008 29/9/2008 30/9/2008 3/10/2008 6/10/2008 7/10/2008 8/10/2008 9/10/2008 10/10/2008 3.335 3.318 3.332 3.343 3.337 3.3385 3.333 3.344 3.342 3.369 3.392 3.374 3.3775 3.391 3.424 3.441 3.431 3.4595 3.421 3.454 3.46 3.471 3.46 3.45 3.455 3.449 3.462 3.4645 3.42 3.404 3.423 3.43 3.4345 3.444 3.46 3.4775 3.4845 3.485 3.5 3.505 3.514 65 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 13/10/2008 14/10/2008 15/10/2008 16/10/2008 17/10/2008 20/10/2008 21/10/2008 22/10/2008 23/10/2008 24/10/2008 28/10/2008 29/10/2008 30/10/2008 31/10/2008 3/11/2008 4/11/2008 5/11/2008 6/11/2008 7/11/2008 10/11/2008 11/11/2008 12/11/2008 13/11/2008 14/11/2008 17/11/2008 18/11/2008 19/11/2008 20/11/2008 21/11/2008 24/11/2008 25/11/2008 26/11/2008 27/11/2008 28/11/2008 1/12/2008 2/12/2008 3/12/2008 4/12/2008 5/12/2008 9/12/2008 10/12/2008 3.5125 3.493 3.511 3.526 3.527 3.525 3.527 3.541 3.573 3.5775 3.595 3.575 3.555 3.565 3.523 3.544 3.523 3.5455 3.555 3.541 3.58 3.591 3.599 3.594 3.599 3.601 3.611 3.626 3.63 3.632 3.6235 3.6235 3.6225 3.62 3.631 3.6415 3.6415 3.6385 3.639 3.631 3.62 66 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 11/12/2008 12/12/2008 15/12/2008 16/12/2008 17/12/2008 18/12/2008 19/12/2008 22/12/2008 23/12/2008 24/12/2008 26/12/2008 30/12/2008 31/12/2008 2/1/2009 5/1/2009 6/1/2009 7/1/2009 8/1/2009 9/1/2009 12/1/2009 13/1/2009 14/1/2009 15/1/2009 16/1/2009 19/1/2009 20/1/2009 21/1/2009 22/1/2009 23/1/2009 28/1/2009 29/1/2009 30/1/2009 3/2/2009 4/2/2009 5/2/2009 6/2/2009 10/2/2009 11/2/2009 12/2/2009 13/2/2009 16/2/2009 3.59 3.552 3.57 3.557 3.5335 3.485 3.47 3.4825 3.487 3.477 3.482 3.485 3.4665 3.474 3.482 3.501 3.5 3.5315 3.545 3.5775 3.58 3.576 3.595 3.582 3.572 3.608 3.613 3.614 3.626 3.612 3.6025 3.61 3.62 3.614 3.623 3.61 3.594 3.615 3.613 3.618 3.6185 67 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 17/2/2009 18/2/2009 19/2/2009 20/2/2009 23/2/2009 24/2/2009 25/2/2009 26/2/2009 27/2/2009 2/3/2009 3/3/2009 4/3/2009 5/3/2009 6/3/2009 10/3/2009 11/3/2009 12/3/2009 13/3/2009 16/3/2009 17/3/2009 18/3/2009 19/3/2009 20/3/2009 23/3/2009 24/3/2009 25/3/2009 26/3/2009 27/3/2009 30/3/2009 31/3/2009 1/4/2009 2/4/2009 3/4/2009 6/4/2009 7/4/2009 8/4/2009 9/4/2009 10/4/2009 13/4/2009 14/4/2009 15/4/2009 3.635 3.659 3.6605 3.672 3.667 3.673 3.671 3.676 3.695 3.7275 3.7115 3.721 3.722 3.722 3.703 3.695 3.692 3.698 3.698 3.675 3.6685 3.658 3.647 3.649 3.625 3.633 3.631 3.617 3.637 3.6495 3.651 3.63 3.59 3.555 3.585 3.634 3.619 3.619 3.618 3.598 3.6195 68 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 16/4/2009 17/4/2009 20/4/2009 21/4/2009 22/4/2009 23/4/2009 24/4/2009 27/4/2009 28/4/2009 29/4/2009 30/4/2009 4/5/2009 5/5/2009 6/5/2009 7/5/2009 8/5/2009 11/5/2009 12/5/2009 13/5/2009 14/5/2009 15/5/2009 18/5/2009 19/5/2009 20/5/2009 21/5/2009 22/5/2009 25/5/2009 26/5/2009 27/5/2009 28/5/2009 29/5/2009 3.59 3.605 3.6325 3.65 3.644 3.643 3.597 3.605 3.62 3.601 3.5635 3.5315 3.52 3.543 3.5225 3.535 3.503 3.525 3.513 3.555 3.545 3.57 3.543 3.5495 3.53 3.499 3.49 3.494 3.491 3.525 3.51