FORECASTING DOMESTIC TOURISM: APPLICATION TO JOHOR TOURISM DATA NURUL FARHANA BINTI ZOLKIPLI UNIVERSITI TEKNOLOGI MALAYSIA FORECASTING DOMESTIC TOURISM: APPLICATION TO JOHOR TOURISM DATA NURUL FARHANA BINTI ZOLKIPLI A thesis submitted in fulfilment 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 Abah, Mak, Yaya and Ail iv ACKNOWLEDGEMENT Assalamualaikum warahmatullahi wabarakatuh… Firstly, I would like to thank Allah because of His blessings; I would be able to successfully complete this thesis. Besides that, I also want to convey my deepest appreciation to my supervisor, Prof. Dr. Zuhaimy bin Hj. Ismail, for guiding and advising me all the way through. Not forgotten, thousand of thanks to my beloved parents for giving me their strongest and continuous support every time I drowned in troubles during the process of doing this thesis. Profuse thanks also to all my fellow friends for lending me their hands every time I asked their favour. I am very grateful and thankful because finally and successfully, I manage to complete this thesis after trudging along the difficult road. Thanks a lot. v ABSTRACT Serious tourism forecasting spans over three decades. A number of tourism forecasters have noted that complex forecasting models are seldom more accurate in predicting the future than simple ones. Hence, in this study, it shows through a Johor tourism forecasting comparison using the naïve model, moving average and exponential smoothing. An approach to selecting the method appropriate to Johor tourism data is based on criteria which are Mean Absolute Percentage Error (MAPE) and Mean Square Error (MSE). The naïve model, moving average and exponential smoothing are relatively simple to apply, requiring no more than a data series and a computer spreadsheet program. This study also explores the need of Tabu Search method in implying forecast. Tabu Search will be used to ensure the pattern of the studied data by searching the value of weighted constants , and . The confirmation is performed by Automated Computerized Forecasting (ACF) System. vi ABSTRAK Ramalan dalam industri pelancongan telah dijalankan secara serius menjangkau lebih daripada tiga puluh tahun. Beberapa peramal menyedari bahawa model ramalan yang kompleks tidak selalunya memberi keputusan yang lebih tepat daripada model yang ringkas dalam meramal masa depan. Oleh itu, kajian ini menunjukkan perbandingan ramalan industri pelancongan negeri Johor menggunakan model naif, Purata Bergerak dan Pelancar Eksponen. Pendekatan untuk memilih kaedah yang sesuai bagi data pelancongan Johor dengan berpandukan kepada kriteria yang mana ianya adalah Min Peratusan Ralat Mutlak (MAPE) dan Min Kuasa Dua Ralat (MSE). Model naif, Purata Bergerak dan Pelancar Eksponen agak mudah diaplikasi, memerlukan hanya siri data dan boleh menggunakan program komputer berlembaran. Kajian ini juga meneroka perlunya kaedah Carian Tabu dalam pelaksanaan ramalan. Carian Tabu akan digunakan sebagai pengesahan kepada corak data yang dikaji dengan mencari nilai pemberat bagi pemalar , dan . Pengesahan ini dilakukan melalui sistem Pengkomputeran Automatik Ramalan (ACF). vii TABLE OF CONTENTS CHAPTER 1 TITLE PAGE DECLARATION ii DEDICATION iii ACKNOWLEDGEMENTS iv ABSTRACT v ABSTRAK vi TABLE OF CONTENTS vii LIST OF TABLES x LIST OF FIGURES xi LIST OF SYMBOLS xiii LIST OF APPENDICES xiv INTRODUCTION 1 1.1 Introduction 1 1.2 Background of the Problem 2 1.3 Statement of the Problem 4 1.4 Objectives of the Study 5 1.5 Scope of the Study 6 1.6 Significance of the Study 6 1.7 Dissertation Organization 7 viii 2 3 LITERATURE REVIEW 8 2.1 Introduction 8 2.2 Tourism Forecasting 8 2.3 History of Johor 10 2.4 Time Series Forecasting 12 2.5 Types of Forecasting Methods 14 2.6 The Forecasting Programme 15 2.7 The Naïve Forecasting Method 16 2.8 Moving Average Method 18 2.9 Single Exponential Smoothing Method 19 2.10 Tabu Search 21 2.11 Automated Computerized Forecasting System (ACF) 21 2.12 Summary 22 RESEARCH METHODOLOGY 23 3.1 Introduction 23 3.2 Forecast Accuracy 23 3.3 Exponential Smoothing 26 3.4 Holt’s Method 31 3.4.1 Applications of Holt’s Method 33 Holt-Winter’s Method 33 3.5.1 Applications of Holt-Winter’s 35 Tabu Search Method 36 3.6.1 A Basic Tabu Search Procedure 38 3.6.2 Tabu Search Application 39 Summary 40 3.5 3.6 3.7 ix 4 EXPERIMENTATION 41 4.1 Introduction 41 4.2 The Naïve Model 42 4.2.1 The Naïve 1 Model 42 4.2.2 The Naïve 2 Model 44 4.2.3 The Seasonal Naïve Model 46 4.3 Moving Average 48 4.4 Exponential Smoothing 51 4.4.1 Implementation of Exponential Smoothing 52 4.5 Holt’s Model 55 4.6 Tabu Search Application 58 4.6.1 Forecasting Graph Using Tabu Search 59 Summary 61 4.7 5 CONCLUSION AND RECOMMENDATION 62 5.1 Introduction 62 5.2 Conclusion 62 5.3 Recommendation 65 REFERENCES 66 Appendices A-G 72-93 x LIST OF TABLES TABLE NO. TITLE PAGE 1.1 Number of Visitors in Johor 72 4.1 Value of MAPE and MSE when m=4 and m=12 47 4.2 Value of MAPE and MSE Using 3-month Moving Average 50 And 6-month Moving Average 4.3 Value of , MSE and MAPE for Exponential Smoothing 53 4.4 Calculation for Johor Tourism Data using Naïve 1 Model 76 4.5 Calculation for Johor Tourism Data using Naïve 2 Model 79 4.6 Calculation for Johor Tourism Data using Seasonal Naïve Model 82 4.7 Calculation for Johor Tourism Data using 3-month Moving 85 Average 4.8 Calculation for Johor Tourism Data using Exponential Smoothing 88 ( 0.37 ) 4.9 Calculation for Johor Tourism Data using Holt’s Model 91 ( 0.36, 0 and a 1236616, b 6712.383 ) 5.1 Comparison of the Selected Models to Forecast Johor Tourism For 120 periods 64 xi LIST OF FIGURES FIGURE NO. 4.1 TITLE Number of Visitors to Johor for 120 periods (January 1999 PAGE 42 -December 2008) 4.2 Actual and Forecast Value Using Naïve 1 Model for 44 Johor Tourism 4.3 Actual and Forecast Value Using Naïve 2 Model 46 for Johor Tourism 4.4 Actual and Forecast Value Using 3-month Moving Average 49 for Johor Tourism 4.5 Actual and Forecast Value Using 6-month Moving Average 50 for Johor Tourism 4.6 Actual and Forecast Value Using 3-month Moving Average And 6-month Moving Average for Johor Tourism 51 xii 4.7 Actual and Forecast Value Using Exponential Smoothing 54 with 0.05 , 0.20 and 0.37 4.8 The Forecasting Graph Using the Tabu Search in Finding MSE 59 4.9 The Forecasting Graph Using the Tabu Search in Finding MAPE 60 xiii LIST OF SYMBOLS MAPE - Mean Absolute Percentage Error MSE - Mean Square Error - Weighted Constant (between 0 and 1) - Weighted Constant (between 0 and 1) - Weighted Constant (between 0 and 1) AFC - Automated Computerized Forecasting ADLM - Autoregressive Distributed Lag Model ECM - Error Correction Model VAR - Vector Autoregressive TVP - Time Varying Parameter AIDS - Almost Ideal Demand System BSM - Basic Structural Model SES - Single Exponential Smoothing ARIMA - Autoregressive Moving Average N ( s) - Neighbourhood ST - Smoothed Estimate or Smoothed Statistic - Random Component xiv LIST OF APPENDICES APPENDIX TITLE PAGE A Johor Tourism Data 69 B Calculation for Johor Tourism Data Using Naïve 1 Model 73 C Calculation for Johor Tourism Data Using Naïve 2 Model 76 D Calculation for Johor Tourism Data Using Seasonal Naïve 79 (when m=4) E Calculation for Johor Tourism Data Using 3-month Moving 82 Average F Calculation for Johor Tourism Data Using Exponential 85 Smoothing ( 0.37 ) G Calculation for Johor Tourism Data Using Holt’s Model ( 0.36, 0 and a 1236616, b 6712.383 ) 88 CHAPTER 1 INTRODUCTION 1.1 Introduction Predictions of future events and conditions are called forecasts and the act of making such predictions is called forecasting. Forecasting can be broadly considered as a method or a technique for estimating many future aspects of a business or other operation. Planning for the future is a critical aspect of managing any organization and small business enterprise are no exception. Indeed, their typically modest capital resources make such planning particularly important. In fact, the long-term success of both small and large organizations is closely tied to how well the management of the organization is able to foresee its future and to develop appropriate strategies to deal with likely future scenarios. Intuition, good judgment and an awareness of how well the industry and national economy is going may give the manager of a firm a sense of future market and economic trends. Nevertheless, it is not easy to convert a feeling about the future into a precise and useful number. Perfect accuracy (in forecasting) is not obtainable (Brealy and Myers, 1988). If it were, the need for planning would be much less. Still the firm must 2 do the best it can. Forecasting cannot be reduced to a mechanical exercise. Naïve extrapolation or fitting trends to past data is off limited value. It is because the future is not likely to resemble the past that planning is needed. Forecasters rely on a variety of data sources and forecasting methods. In other cases, the forecasters may employ statistical techniques for analyzing and projecting time series. A time series is simply a set of observations measured at successive points in time or over successive periods of time. As far as forecasting method are concerned, they can generally be classified into quantitative and qualitative (Archer, 1980; Uysal & Crompton, 1985). The quantitative approach gives more accurate forecasts than judgment forecasts. These methods are further divided into causal models and time series approaches. Their main distinction is that causal models attempt to identify and measure both economic and non-economic variables affecting other variables such as price and quantity while time series approaches identify stochastic components in each time series. All forecasting methods can be divided into two broad categories: qualitative and quantitative. Division of forecasting methods into qualitative and quantitative categories is based on the availability of historical time series data. Clearly, forecasting is very important in many type of organizations since predictions of the future must be incorporated into decision-making process. In particular, any organizations must be able to make forecasts in order to make intelligent decisions. 1.2 Background of the Problem Some say that travel and tourism is the ‘world’s largest industry and generator of quality jobs’ (World Travel and Tourism Council, 1995: 1). They estimate that 3 tourism directly and indirectly contributes nearly 11 percent of the gross world product, the most comprehensive measure of the total value of the goods and services in the world’s economies produce. The World Travel and Tourism Council estimated that in 2008, gross world product both directly and indirectly related to travel and tourism would total about US$944 billion, up from US$ 857 billion in 2007. This activity is buttressed by more than US$8 trillion invested in world plan, equipment and infrastructure related to travel and tourism. While these estimates may be controversial, there is no doubt that tourism activities, encompassing travel away from home for business or pleasure, comprise a substantial part of lifestyles of the world’s residents, or that a very large industry has grown up to serve these travellers. Futurist John Naisbitt (1944), in his bestselling book Global Paradox, subscribes to the concept that tourism will be one of three industries that will drive the world of economy into twenty-first century. However, the rate of growth has varied immensely from year to year, and while most countries have enjoyed steady increase in their tourist arrivals, some experienced declines in numbers at certain times. Planning under these circumstances is exceptionally difficult and important. But in attempt to plan successfully, accurate forecasts of tourism is required. This point has been noted by several authors in tourism field. For example, Wandner and Van Erden (1980, p. 381) point out that since governments and private industry must plan for expected tourism demand and provide tourism investment goods and infrastructure, the availability of accurate estimates of international tourism demand has important economic consequences. Archer (1987, p. 77) emphasizes the particular necessity for accuracy in tourism forecasting. The tourism industries and those interested in their success in contributing into to the social and economic welfare of citizenry, need to reduce the risk of decisions which is reduce the chances that a decision will fail to achieve desired objectives. Future events important to tourism operations are somewhat predictable and somewhat changeable. Future time, on the other hand, includes time that has passed for which we 4 do not yet have reasonably complete and accurate data, as well as time not yet encountered. Some time series of interest to tourism forecasters may run three or six months behind actual time, so that we may not know what happened to tourism industry in two or three years ahead. Planners and others in public agencies use tourism forecasts to predict the economic, social/ cultural or environmental consequences of visitors. In short, sound tourism forecasts can reduce the risks of decisions and the costs of attracting and serving the travelling public. It is difficult to imagine the tourism industry without forecasting to deal with the risks inherent in saving and investing. 1.3 Statement of the Problem In this research, the tourism forecasting will be done by using basic extrapolative models (naïve and simple moving average). These are seldom used on their own to produce forecasts of tourism, but rather are the foundation for developing and applying more sophisticated models. The naïve models are especially useful as benchmarks for evaluating more complex forecasting techniques. Then, we can test a number of alternative quantitative forecasting models, including extrapolative methods. We look for the model that best simulates the data, using criterion of error magnitude such as Mean Square Error (MSE) and Mean Absolute Percentage Error (MAPE). This should be the best candidate for forecasting future periods. Once we have determined this model, we can produce the forecasts we need. We will also examine intermediate extrapolative forecasting methods, specifically single exponential smoothing, double exponential smoothing and triple 5 exponential smoothing. These methods are intermediate in their complexity among time series forecasting methods. Single exponential smoothing can be used to forecast from stationary time series while double exponential smoothing is designed for series showing a linear trend or a trend with seasonality removed. Triple exponential smoothing can be used for series including both trend and seasonal components. Intermediate time series methods are somewhat more complex than their basic brethren, but can still be handled in modern computer spreadsheets. However, as complexity arises, more data are required in time series models to obtain satisfactory results. Finally, the use of Tabu Search will be applied to verify the pattern of the data by searching the value of , and . This is because Tabu Search technique is a metaheuristic that guides a local heuristic search procedure to the solution space beyond optimality. It uses adaptive memory (memory-based strategies) and takes advantage of historical information of past solutions to form memories which are useful to help the search to explore other regions via diversification and intensification whenever necessary. 1.4 Objectives of the Study Objectives of this study are: i. to forecast tourism using basic and intermediate extrapolative methods in Microsoft Excel spreadsheets. ii. to use Tabu Search for the confirmation of data pattern based on the value of , and found. iii. to compare the obtained results from those methods and select the best model in forecasting Johor tourism.. 6 1.5 Scope of the Study In this study, the basic and intermediate extrapolative methods will be used in determining the best model to forecast tourism. The naïve method will be introduced as the basic of this research, followed by the moving average and exponential smoothing method. At the mean time, the Tabu Search technique will be used to ensure the data pattern based on the value of , and found using ACF system. This study concentrates only on univariate model where the time series forecasting based on the past value. Measurement of error used is the MSE and MAPE. The data that will be used in this study is the Johor tourism data which is taken from Majlis Tindakan Pelancongan Negeri Johor (MTPNJ). The data covers the period for ten years from January 1999 to December 2008. The data are listed in the Appendix A. 1.6 Significance of the Study The need for quick, reliable and simple forecasts of various time series is often encountered in economic and business environments. This study is an introduction of scientific approach in making decision. The result from this study is practical for other forecasters as a guide in forecasting especially in using extrapolative methods. A tourist forecaster or policy maker can at least make a good estimate of visitor arrivals in the absence of structural data by just using the available time series data. Relatively simple methodologies can be used on a spreadsheet and can give reasonable estimates albeit a short period time into the future. Moreover, these models are such that can be easily updated as time goes on; hence a constant track can be kept of the expected number of visitor arrivals. These models also can be extended to cover more regions or even predict the total expected number of visitor arrivals. Further works need to be done to generalize the findings to other tourist receiving destinations. That is, future studies can employ the same forecasting models but with different data series, for forecasting 7 accuracy testing. Moreover, additional forecasting techniques can be incorporated into future studies. 1.7 Dissertation Organization Chapter 1 is the framework of the study that includes brief introduction, problem background, problem statement, objectives of the study, scope of the study, significance of the study and research outline. Chapter 2 looks at the literature review involving tourism forecasting, history of Johor, time series forecasting, the types of forecasting methods, the naïve methods, the moving average methods, the single exponential smoothing method, the double exponential smoothing method, the triple exponential smoothing method, the Tabu Search technique and the Automated Computerized Forecasting (ACF) system. Chapter 3 explores more about those methods mentioned above. The forecast errors which are MSE and MAPE will be discussed in forecast accuracy. This chapter will elaborate about the approach of the Tabu Search. Chapter 4 presents the implementation of forecasting methods. Here, it shows the calculation using each basic and intermediate extrapolative method on Johor tourism data. Chapter 5 discusses the comparison of each forecasting method. The best model among these methods will be selected in order to forecast. CHAPTER 2 LITERATURE REVIEW 2.1 Introduction This chapter briefly discusses time series forecasting which includes the component of time series and its applications. Explanation of naïve method, moving average and single exponential smoothing as types of forecasting methods will also be discussed. This is followed by the introduction of Tabu Search as a method to ensure the pattern of the data. 2.2 Tourism Forecasting Forecasting is a process designed to predict future events. In the realm of tourism, an event may be the number of visitors to a destination, the number of roomnights sold in a hotel or a group of hotels, the number of passengers flying between two points or the number of brochures requested by potential visitors. A valid forecast 9 event has two characteristics: a specific time and a specific outcome. These outcomes are often precise volumes of demand. However, they may be stated as ranges or even qualitative conditions, such as visitor demand is forecast to be greater next year than this year. Tourism forecasting has become an important component in tourism research and different approaches have been used to generate forecasts of tourism. By the end of the 1950s, the serious study of international tourism had begun. By virtue of the accumulating record of international level at this time, researchers began to study international tourism demand, one area of which has been forecasts. Over the last three to four decades, a substantial number of such study have been undertaken. A large portion of these have yielded empirical information and a considerable majority has reported the accuracy of forecasts. Witt and Witt (1955) provided a comprehensive review of the early tourism forecasting literature. Together with the rapid development of modern econometrics, many researches have now applied these recent developments in forecasting tourism demand in various settings. Li, Song and Witt (2005) reviewed eighty-four-post-1990 empirical studies of international tourism demand modeling and forecasting and gave an extensive and detailed view on issues such as data types and frequencies, independent and dependant variables, estimation methods and reported diagnostic test statistics. Their review suggested that the most frequently used forecasting methods in tourism are the static regression model, the autoregressive distributed lag model (ADLM), the error correction model (ECM), the vector autoregressive (VAR) models, time varying parameter (TVP) model, almost ideal demand system (AIDS) and basic structural model (BSM). Song, Witt and Li (2003) used the general-to-specific modeling approach to obtain ex ante forecasts of the demand for Thai tourism. Song and Witt (2006) used the VAR modeling technique to forecast the demand for Macau tourism over the period 10 2003-2008. Kulendran and King (1997) considered four time series models and econometric model when predicting quarterly tourist flows into Australia from four major tourist markets. Song, Witt and Jensen (2003) compared the forecasting performance of the ECM, ADLM, TVP and VAR models with those generated by two univariate time series models in forecasting the demand for Denmark tourism and found that the TVP model generates the most accurate one-year-ahead forecasts. Li, Song and Witt (2006) reported the forecasts of tourist expenditure by United Kingdom (UK) residents in a number of Western European countries using TVP and constant parameter linear AIDS models. 2.3 History of Johor Johor, the southern gateway and third largest state in Peninsula Malaysia is the keeper of many national and natural treasures untold. It is a delightful mixture of the traditional and modern, and natural and ultra-modern urban jungles. Johor gets its name from the Arabic word Jauhar meaning 'precious stones'. Prior to that, it was known as Ujung Tanah which mean 'land's end' in Malay, due to its location at the end of the Malay Peninsula. It is also an apt name, as the southern-most tip of Asia is located at Tanjung Piai, Johor. Johor has a rich and illustrious ancient history - its sultanate was established in the early 16th century, commanding the southern Peninsula and Riau islands. Upon Malacca's defeat to the Portuguese in 1511, the son of the last Sultan of Malacca, Sultan Mahmud Shah, had established a monarchy in Johor and had posed a constant threat to the Portuguese. Johor is the only state, apart from Melaka, that grew into an empire. During its peak, the whole of Pahang and the present day Indonesian territories of Riau Archipelago and part of Sumatra Island were under Johor's rule. (Source: http://tourismjohor.com/about/) 11 With its 3.17 million populations, Johor is the most populous state in Malaysia. Besides the cultural expressions practiced by the local Malays, Chinese and Indians, culture in Johor is also influenced by visitors and traders throughout history. Johor is divided into eight districts: Batu Pahat, Johor Bahru, Kluang, Kota Tinggi, Mersing, Muar, Pontian and Segamat, the largest of these being Johor Bahru which is the state capital. Visitors commonly head for Johor Bahru's royal palaces and lively night markets, considered as one of the state’s best attractions. Johor also has eight small islands located off its coast: Tinggim, Rawa, Sibu, Tengah, Pemanggil, Aur, Besar dan Lima. Most of these islands are sparsely populated, with Tinggi being the most significance island and a popular tourist attraction which can be visited by boat from the mainland. (Source: http://www.hoteltravel.com/malaysia/johor/guides/overview.htm) Johor is increasingly popular for eco tourism as it is endowed with rich natural heritage. The Endau-Rompin Park, one of the world’s oldest tropical rain forests – offers great natural beauty for the adventurous traveler. Furthermore, Johor’s beautiful and natural forest parks also being developed into centers for eco tourism. The beacon of new growth, Iskandar Malaysia will spur economic developments that actuate Malaysia's global potential to greater heights. Sprawling 2,217 sq. km, it is 3 times the size of Singapore and is comprised of most of Johor's major hubs including Johor Bahru, Danga Bay, world-class ports Tanjung Pelepas and Pasir Gudang, amongst others. (Source: http://tourismjohor.com/about/) 12 2.4 Time Series Forecasting A time series is a chronological sequence of observations on a particular variable. In forecasting, we are trying to estimate how the sequence of observations will continue into the future. Forecasting the future value of an observed time series is an important problem in many areas, including economics, production planning, sales forecasting and stock control. Forecast can be based entirely on past observations in a given time series, by fitting a model to the data and extrapolating, which is called univariate (Chatfield, 1984). Forecasting is done at all levels and events. Education, job decision and long-life partner are made on personal prediction of future happenings. In formal level, economists and politicians are continually forecasting economics and politic trends. Many different forecasting techniques are used with varying degrees of success (Runyon et. al, 1982). Types of forecasting procedure can be classified into three categories (Chatfield, 1984): a) Subjective- forecast can be made on subjective basis using judgment, intuition, commercial knowledge and any other relevant information. For example is Delphi method. b) Univariate- forecast can be based entirely on past observations in a given time series, by fitting a model to the data and extrapolating. For example, forecast of future sales of a product based entirely on past sales. c) Multivariate- forecast can be made by taking observations on other variables into account. For example, sales may depend on stocks. Although there are many different forecasting techniques, this study focuses on the most commonly used forecasting technique, the univariate time series model. In this model, past data are analyzed over a given time period and patterns are estimated. 13 A key assumption in the univariate time series model is that patterns which have occurred in the past will continue in the future (Runyon et. al, 1982). An important in selecting an appropriate forecasting method is to consider the type of data patterns, so that the methods most appropriate to those patterns can be utilized. Four types of time series data patterns can be distinguished (Makridakis, Wheelwright, Hyndman, 1998): seasonal, cyclical, trend and irregularity. a) Seasonal- exists when a series is influenced by seasonal factors (e.g., the quarter of the year, the month or day of the week). Sales of product such as soft drinks, ice creams and household electricity consumption all exhibit this type of pattern. Seasonal series are sometimes called “periodic” although they do not exactly repeat themselves over each period. b) Cyclical- exists when the data exhibit rises and falls that are not of a fixed period. For economic series, these are usually due to economic fluctuations such as those associated with the business cycle. The sales of products such as automobiles, steel and major appliances exhibit this type of pattern. The major distinction between a seasonal and cyclical pattern is that the seasonal is of a constant length and recurs on a regular periodic basis while the cyclical is varies in length. Moreover, the average length of cycle is usually longer than seasonal and the magnitude of a cycle is usually more variable than seasonal. c) Trend- exists when there is a long-term increase or decrease in the data. The sales of companies, the gross national product (GNP) and many other business or economic indicators follow a trend pattern in their movement over time. d) Irregularity- exhibit erratic movements in a time series that follow no recognizable or regular pattern. Such movements represent what is ‘left over” in a time series after seasonal, cyclical, trend have been accounted for. Many irregularities in time series are caused by “unusual” events that cannot be 14 forecasted such as earthquake, accidents, hurricanes, wars, etc. This factor is the source of error in forecasting. Many data series include combinations of the preceding patterns. For example, the sales of product shows trend, seasonality and cyclical behaviour. One of the things that make forecasting interesting and challenging is the huge variety of patterns that commonly occur in the real time series data. Forecasting methods that are capable of distinguishing each of the patterns must be employed if a separation of the component patterns is needed. Similarly, alternative methods of forecasting can be used to identify the pattern and to best fit the data, so that future values can be forecasted. 2.5 Types of Forecasting Methods Forecasting methods fall into two major categories: quantitative and qualitative. Quantitative methods organize past information about a phenomenon by mathematical rules. These rules take advantage of underlying patterns and relationships in the data of interest to the forecaster. Objective numerical measurements consistent over some historical period are required in these methods. These methods also assume that at least some elements of past patterns will continue into the future (Frechtling, 2001). There are two major subcategories of quantitative methods: extrapolative and causal. Extrapolative methods also called ‘time series methods’, assume that the variable’s past course is the key to predicting its future. Patterns in the data during the past are used to project or extrapolate future values. Causal relationships are ignored. 15 The other subcategory of quantitative forecasting methods is causal methods. These attempt to mathematically simulate cause-and-effect relationships. Determining the causal variables (better called ‘explanatory variables’) that affect the forecast variable and the appropriate mathematical expression of this relationship is the central objective. These methods have the advantage over time series methods of explicitly portraying cause-and-effect relationships. This is crucial in certain forecasting situations, such as when management wants to know how much impact on demand an increased advertising budget will have. Likewise, tourism policy forecasting requires causal models. However these methods are more costly and time-consuming to construct than time series models and are often considerably less accurate. Qualitative methods are also called ‘judgmental methods’. Past information about the forecast variable is organized by experts using their judgment rather than mathematical rules. These are not necessarily cheaper or easier to apply than quantitative methods, but they have the advantage of not requiring historical data series. (Frechtling, 2001). 2.6 The Forecasting Programme The objective of the forecasting programme is to establish a system for periodically producing forecasts required by management, termed a tourism demand forecasting programme. There are four major phases of developing this forecasting programme, focusing on building a system that will be used repeatedly over a year or 16 more to produce forecasts: design phase, specification phase, implementation phase and evaluation phase. The design phase guides the forecaster in choosing the appropriate forecasting method to employ. This phase examines the problem, the resources and the relationships that help determine a preliminary choice of method. The specification phase includes determining the relationships that will comprise the appropriate forecasting model and selecting an appropriate model. The implementation phase comprises employing the selected model to generate forecasts and preparing these forecasts for presentation to management. The evaluation phase covers monitoring the forecasts over time to determine if adjustments should be made in the forecasting model and making the appropriate adjustments to secure the most accurate series of forecasts. Following these steps ensures that the forecaster systematically develops a valid strategy for solving his or her forecasting challenge. This help ensures that we do not waste time and money in determining the shapes of the tourism features we are interested in. 2.7 The Naïve Forecasting Method The naïve forecasting method simply states the value for the period to be forecast is equal to the actual value of the last period available. More formally, 17 Naïve 1 model: Ft At 1 (2.7.1) where F = forecast value A = actual value t = some time period. This is the simplest forecasting model. As such, it is frequently used as a benchmark to compare other forecast models against. It is not unusual for more elaborate models to produce higher MAPEs than the naïve model and are thus not worth the time and money to operate. Witt and Witt (1992: 99-123) present a number of these situations among international tourism demand series and conclude that more complex forecasting models are less accurate than the naïve model for many series. There are two other versions of the naïve concept that are sometimes used as a benchmark forecast. The Naïve 2 model defines the forecast value as the current value multiplied by the growth rate between the current value and the previous value. This might be a useful benchmark for a series that trends upward or downward. Naïve 2 model: Ft At 1 * At 1 At 2 where F = forecast value A = actual value t = some time period The ‘seasonal’ naïve can be used with a seasonal data and postulates that the next period’s value is equal to the value of the same period in the previous year. Seasonal naïve model: Ft At m 18 where F = forecast value A = actual value t = some time period m = number of periods in a year (for example, four quarters, twelve months) 2.8 Moving Average Method The moving average method is the second in simplicity only to the naïve method. We can average any number of periods to produce a forecast through the moving average model. The general equation for the model is: Ft At 1 At 2 At n n where F = forecast value A = actual value t = some time period n = number of periods The moving average method allows some past values to determine forecast values and all have the same influence on the forecast value. The more past values included in the model, the smoother it becomes. This is because the more values (the higher n is), the less influence any one values has on the average; rather, the values tend to offset each other to provide a smooth forecast series. 19 If a time series shows wide variations around a trend, then the longer the moving average, the better it will pick up the trend. However, long moving averages are slow to pick up recent changes in trend because so many past values are affecting it. The moving average method is more accurate in forecasting a series with very little variation around its trend than one with significant seasonality or volatility. 2.9 Single Exponential Smoothing Method Smoothing method was first developed in the late 1950s by operation researches. It is unclear whether Holt (1957) or Brown (1956) was the first to introduce exponential smoothing, or perhaps it was Magee (1958). Most of the important development work on exponential smoothing was completed in the 1950s and published by the early 1960s. This work included that done by Brown (1956) and Holt (1957) and subsequent work by Magee (1958), Brown (1959), Hot et al. (1960), Winters (1960), Brown and Meyer (1961), and Brown (1963). Since that time, the concept of exponential smoothing has grown and become a practical method with wide application, mainly in the forecasting of inventories. Single exponential smoothing (SES) is a forecasting technique that attempts to “track” changes in a time series by using newly observed time series values to “update” the estimates of parameters describing the time series. It is also known as simple exponential smoothing. Simple smoothing is used for short-range forecasting, usually just one month into the future. The model assumes that the data fluctuates around a reasonably stable mean (no trend or consistent pattern of growth). SES allows us to vary the important of recent values to the forecast and includes all of the information past values can provide us. 20 The logic of the SES is the evident in its general equation: Ft Ft 1 ( At 1 Ft 1 ) where F = forecast value = smoothing constant between 0 and 1 A = actual value t = some time period Another way of writing the above equation is: Ft At 1 (1 ) Ft 1 According to Witt, Newbould and Watkins (1992:38), exponential smoothing tourism forecasting models ‘tend to perform well, with accuracy levels comparable to more complex and statistically sophisticated forecasting methods which require considerable user understanding to employ them successfully’. They further maintain that while this method is not prominent in forecasting literature, in actual practice it, along with the moving average method, is the most popular. The major disadvantage of this method is that it cannot take into account factors affecting the series other than its past values. It does not explain the relationships between such factors and the series of interest. Should events occur that can radically change tourism behaviour, such as pestilence, terrorism, entertainment mega-events or natural disasters, time series methods fail. However, they can indicate the values that should have been achieved in the absence of these catastrophes and thus measure the magnitude of their impact on tourism. 21 2.10 Tabu Search Tabu Search has its antecedents in methods designed to cross boundaries of feasibility or local optimality normally treated as barriers and systematically to impose and release constraints to permit exploration of otherwise forbidden regions. Early examples of such procedures include heuristic based on surrogate constraint methods and cutting plane approaches that systematically violate feasibility conditions. The modern form of Tabu Search derives from F. Glover. Seminal ideas of the method are also developed by P. Hansen in a steepest ascent/mildest descent formulation. Additional contributions are shaping the evolution of the method and are responsible for its growing body of successful applications. Webster’s Dictionary defines tabu or taboo as ‘set apart as charged with a dangerous supernatural power and forbidden to profane use or contact…’ or ‘banned on grounds of morality or taste or as constituting a risk…’. Tabu Search scarcely involves reference to supernatural or moral considerations, but instead is concerned with imposing restrictions to a guide a search process to negotiate otherwise difficult regions. These restrictions operate in several forms, both by direct exclusion of certain search alternatives classed as ‘forbidden’ and also by translation into modified evaluations and probabilities of selection. 2.11 Automated Computerized Forecasting System (ACF) The ACF system has been developed by Mohd Nizam (2003) specifically as an interactive system for solving Winter’s method of forecasting. It is built in order to search for the smoothing parameter ( , and ) within the range 0 to 1 using Genetic Algorithm (and Tabu Search). The results are output in the form of Graph that show the 22 actual value and the forecast value using the constant parameter ( , and ) found earlier. This software is written using Borland Delphi 5 Enterprise. There are two parts in this system which are the integration of Genetic Algorithm into Winter’s method and the Tabu Search as a refinement from Genetic Algorithm search result. In this study, we only focus on the second part. The Tabu Search properties Windows consist of Neighbourhood Size, Tabu List Size and Number of Iteration. For neighbourhood size, it allows the user to specify the number of neighbour. A larger number of neighbourhoods would lead to a slower processing time (a more time is required to calculate the MSE or MAPE). As for Tabu List Size, user can specify the size of interest. Tabu list is used to prevent the search going back to the point visited previously. Lastly, in Convergence, it allows the user to choose the number of iteration for the Tabu Search procedure. 2.12 Summary The literature review provides an in-depth understanding of the area of forecasting. The tourism industry, time series forecasting and the types of forecasting methods have been discussed thoroughly. Tabu Search technique described here is chosen to obtain the value of , and . The ACF also has been explained before its implementation in chapter 4. CHAPTER 3 RESEARCH METHODOLOGY 3.1 Introduction In the previous chapter, literature review was given on tourism forecasting, time series forecasting, types of forecasting methods, Tabu Search and the Automated Computerized Forecasting System. This chapter will discuss the research methodology which consists of forecast accuracy and application on Exponential Smoothing, Holt’s method, Holt-Winter’s method and Tabu Search. All equations involved are given in this chapter. 3.2 Forecast Accuracy More accurate forecasts reduce the risks of decisions more than do less accurate ones. Accuracy is the most important forecast evaluation criterion (Frechtling, 2001). The word “accuracy” refers to “goodness of fit,” which in turn refers to how well the 24 forecasting model is able to reproduce the data that are already known. To the consumer of forecasts, it is the accuracy of the future forecast that is most important (Makridakis, Wheelwright, Hyndman, 1998). The most familiar concept of forecasting accuracy is called ‘error magnitude accuracy’ and relates to forecast error associated with a particular forecasting model. This is defined as: et At Ft where t = some time period, such as month, quarter or year e = forecast error A = actual value of the variable being forecast F = forecast value If the actual value is greater than the forecast value at time, t, then the forecasting error is positive. If less than the forecast value, then the forecasting error is negative. There are quite a few ways to summarize the error magnitude accuracy of a forecasting model. Some of these compute measures of absolute error and are thus subject to the units and time period over which the model is tested. They are often difficult to interpret and compare across different models. Other error magnitude measures compute percentage errors relative to the values in the historical series. These allow comparing several different models across different time periods. One of the most useful of these, due to its simplicity and intuitive clarity, is the MAPE: MAPE 1 n et n t 1 At 25 where n = number of periods e = forecast error A = actual value of the variable being forecast t = some time period The MAPE is the sum of the absolute errors for each time period divided by the actual value for the period; this sum is divided by the number of periods to obtain a mean value. Then, by convention, this is multiplied by 100 to state it in percentage terms. This is a simple measure permitting comparison across different forecasting models with different time periods and numbers of observations and weighting all percentage error magnitudes the same. Lower MAPE values are preferred to higher ones because they indicate a forecasting model is producing smaller percentage errors. Moreover, its interpretation is intuitive. The MAPE indicates, on the average, the percentage error a given forecasting model produces for a specified period. One author has suggested the following interpretation of MAPE values: less than 10 percent is highly accurate forecasting between 10 and 20 percent is good forecasting between 20 and 50 percent is reasonable forecasting greater than 50 percent is inaccurate forecasting Such a standard can be quite misleading because it ignores the change characteristics of the time series being forecast. The MSE is calculated using this equation: MSE 1 n 2 et n t 1 26 where n = number of periods t = some time period e = forecast error Here, the errors are made positive by squaring each one and then the squared errors are averaged. The MSE has the advantage of being easier to handle mathematically and so it is often used in statistical optimization. There is no consensus among statisticians as to which measure is preferable. The decision is which one to use depends on the make up of the data; if there are only one or two large errors, these will be magnified by using MSE, thus another forecast error such as MAPE will be used. But if all the error is similar in magnitude, the MSE will be used. 3.3 Exponential Smoothing Exponential smoothing is a procedure for continually revising a forecast in the light of more recent experience. Exponential smoothing assigns exponentially decreasing weights as the observation get older. In other words, recent observations are given relatively more weight in forecasting than the older observations. To begin the exponential smoothing approach, let now suppose that at the end of a particular time period, which we shall call time period T-1, we have obtained a set of observations for the time series, which we shall denote A1, A2,..., AT 1 . Given these observations we wish to estimate , the average level of the time series. As just stated, 27 the least squares estimate of , which we now denote as b (T 1) to emphasize the fact that the most recent observation in the time series corresponds to time period T-1, is T 1 b (T 1) A At t 1 (T 1) So, the estimate of is simply the average of the observations in the time series through period T-1. Given this estimate, the forecast for any future time period, say period T 1 , where is a positive integer, is b (T 1) . Thus the forecast for any future period is the average of the observations in the time series through period T-1. Now suppose that we obtain a new observation AT at the end of the next time period, period T. We would like to incorporate this new observation into the estimate of . That is, we wish to obtain an updated estimate of that is based on the new observation AT as well as the old observations A1, A2,..., AT 1 . We shall call this new estimate b (T ) to indicate that the most recent observation in the time series corresponds to time period T. Regression may be used to obtain such an estimate. This estimate is given by T b (T ) A t 1 At T So this new estimate can be found by simply recalculating A , which is now the average of the observations in the time series through period T . This new estimate b (T ) is then the new forecast for the average level of the time series in any future time period. Another way to incorporate this new observation into the estimate of is known as simple exponential smoothing. This approach generates the new estimate 28 b (T ) in a different a different way. It seems intuitive to change the old estimate b (T 1) by some fraction of the forecast error which resulted when the old estimate was used to forecast the value of the time series for the present period. This forecast error is given by eT AT b (T 1) and is the difference between the observed value in period T and the forecast made for period T in period T-1. If the fraction we use is , then the updated estimate is given by b (T ) b (T 1) [ AT b (T 1)] Then the new estimate is partly based on the old estimate b (T 1) . If the old estimate yielded a forecast for period T that was too low, then the new estimate is higher. If the old estimate yielded a forecast for period T that was too high, then the new estimate is lower. The magnitude of the adjustment up or down is determined by the magnitude of the forecast error. A large error leads to a large adjustment while a small error leads to a small adjustment. In order to simplify notation at this point, we shall define. So the equation we use to update the estimate can be written as ST ST 1 ( AT ST 1 ) ST 1 AT ST 1 ST AT (1 ) ST 1 This equation defines the updating procedure called simple exponential smoothing. We call ST the smoothed estimate or smoothed statistic. The fraction is called the smoothing constant. 29 Examining the smoothing equation ST AT (1 ) ST 1 we see that the smoothed estimate is simply an estimate based on the observations A1, A2,..., AT . This is true since ST 1 or b (T 1) is the average of the observations A1, A2,..., AT 1 . Now, let us change the time origin so that the initial estimate of is assumed to be generated in time period zero. We shall call this initial estimate S . In practice, the estimate S would be obtained by calculating the average of an initial set of the observations in the time series. If such an initial set of observations is not available, S is commonly set to the first observed value of the time series. Let us also assume that the smoothing equation ST AT (1 ) ST 1 has been used to update the estimates for each time period t from period 1 to period T, the present period. In this situation, the smoothed estimate for period T, that is ST , can be shown to be a linear combination of all past observations. To see this, we consider the smoothed estimate ST AT (1 ) ST 1 from which we can see that ST 1 AT 1 (1 )ST 2 Substitution therefore gives us ST AT (1 )[ AT 1 (1 ) ST 2 AT (1 ) AT 1 (1 ) 2 ST 2 30 Again, we can see that ST 2 AT 2 (1 ) ST 3 and substituting again we have ST AT (1 ) AT 1 (1 ) 2 [ AT 2 (1 ) ST 3 ] AT (1 ) AT 1 (1 )2 AT 2 (1 )3 ST 3 Substituting recursively for ST 3 , ST 4,..., S 2 and S1 , we obtain ST AT (1 ) AT 1 (1 ) 2 AT 2 ... (1 )T 1 A1 (1 )T A Thus we see that ST , the estimate of in time period T, can be expressed in terms of the observations A1, A2,..., AT and the initial estimate S . The coefficients of the observations, , (1 ), (1 )2 ,..., (1 )T 1 measure the contributions that the observations AT , AT 1, AT 2,..., A1 make the most recent estimate ST . It can be seen that these coefficients decrease geometrically with the age of the observations. For example, if the smoothing constant is 0.1, then these coefficients are 0.1, 0.09, 0.081 and so on. The updating procedure we have described is called simple exponential smoothing because these coefficients decrease exponentially. Since these coefficients are decreasing, the most recent observation AT makes the largest contribution to the current estimate of . Older observations make smaller and smaller contributions to the current estimate of at each successive time point. Thus, remote observations are “dampened out” of the current estimate of as time advances. The rate at which remote observations are dampened out depends on the 31 smoothing constant . For values of near 1, remote observations are dampened out quickly, while for values of near 0, remote observations are dampened out more slowly. For example, for 0.09 , we obtain coefficients 0.9,0.09,0.009,0.0009,… while for 0.1 , we obtain coefficients 0.1,0.09,0.081,0.0081,… So, the choice of the smoothing constant has a great bearing on the estimate ST . In general, when the time series quite volatile, that is when the random component t has a large variance, we would select a small smoothing constant, so that the smoothed estimate ST will weight ST 1 , the smoothed estimate for the previous period, to a greater degree than it weights the observation AT . For more a stable time series, in which the random component t has a smaller variance, we would select a larger smoothing constant. In making forecast, suppose we are in period T and the current estimate of is ST b (T ) . We wish to forecast the time series for a future period T . Since the model is At t , the forecast is simply FT (T ) ST , the current estimate of . Here the use of the hat F indicates that FT (T ) is a forecast value rather than an observation of the time series. The notation of FT (T ) is used to emphasize that this forecast is being made for period T and is being made in period T. 3.4 Holt’s Method This method was developed to deal with time series showing a linear trend over time. In essence, the Holt’s model computes a smoothed level and trend at each data point. These values for the last point in the time series can be used to forecast one or two points ahead in the future. The level is a smoothed estimate of the value of the data 32 at the end of each period. The trend is a smoothed estimate of average growth at the end of each period. The equations for this model are: Level : At (1 )(at 1 bt 1 ) Trend : bt (at at 1 ) (1 )bt 1 Forecast : Ft m at mbt where a = level of the series = level smoothing constant between 0 and 1 A= actual value b= trend of the series trend smoothing constant between 0 and 1 t = some time period m = number of time periods ahead to be forecast We need to develop two initial estimates to begin the Holt’s process. Makridakis, Wheelwright and Hyndman (1998: 159) suggest the following: Level initialization : a1 A1 Trend initialization : b1 A2 A1 A variation on this Holt’s method is double exponential smoothing method (Makridakis, Wheelwright and Hyndman, 1998:158). Here, the smoothing constant, , in the level equation is replaced by an independent smoothing constant, , in the trend equation. These two values are varied between zero and one to produce the equation with the lowest MAPE or other measure of error. 33 The advantages of Holt’s method are that it is relatively simple, captures linear trends up or down well and can forecast several periods ahead (unlike single exponential smoothing). Its disadvantages are that it is cannot track non-linear trends well, fails to simulate a stepped series well and cannot deal with seasonality. Finally, in common with all time series methods, it does not incorporate any causal relationship that may be important to management. 3.4.1 Applications of Holt’s Method Sheldon (1993) tested eight models for forecasting international visitor expenditures in six developed countries and found that Holt’s method was the second most accurate model in terms of MAPE. The most accurate model was the Naïve 1 model of no change from the previous year. Martin and Witt (1989) compared the accuracy of seven forecasting methods for simulating visitor flows among twenty-four origin-destination pairs and, like Sheldon, found that exponential smoothing was the second most accurate model in terms of MAPE, after Naïve 1. 3.5 Holt-Winter’s Method The Holt-Winter’s trend and seasonality method employs triple exponential smoothing: one equation for the level, one for the trend and one for the seasonality. The 34 equations associated with each of these elements are as follows (Makridakis, Wheelwright and Hyndman, 1998:165): yt (1 )(at 1 bt 1 ) St s Level : at Trend : bt (at at 1 ) (1 )bt 1 Seasonality : St Forecast At (1 ) St s at : Ft m (at mbt ) St s m where a = level of the series = level smoothing constant between 0 and 1 A = actual value s = number of seasonal periods in a year (for example, four quarters, twelve months) b = trend of the series = trend smoothing constant between 0 and 1 S = seasonal component = seasonal smoothing constant between 0 and 1 t = some time period m = number of time periods ahead to be forecast Initialization is more complex than the earlier Holt’s method. Values must be sought for as , bs and S s , that is at the end of the first complete season. One recommended approach is the following (Makridakis, Wheelwright and Hyndman, 1998:168): 35 A1 A2 ... As s Initial level as Initial trend A As 1 A A A A2 bs s 1 1 s 2 ... s s s s s s Seasonal indices for the first year S1 A A1 A , S 2 2 ,..., S s s as as as The level is initialized by the average of the first season of values. The trend is initialized by the average of each of the s estimates of trend over the first season. Finally, the seasonal indices are initially set as the ratio of the first year’s values to the mean of the first year, as . 3.5.1 Applications of Holt-Winter’s Method Chu (1998) tested six forecasting methods in forecasting monthly visitor arrivals in ten Asia-Pacific nations over the 1975-94 period: Naïve 1, Naïve 2 time trend regression, sine wave regression, autoregressive/ moving average (ARIMA) and HoltWinters. Using the MAPE as his measure in eighteen-month ex post forecasts, he found Holt-Winter’s models were the second only to ARIMA in producing superior forecasts for nine of the ten countries and led ARIMA in forecasting New Zealand visitor arrivals. Turner, Kulendran and Fernando (1997) compared six different methods for forecasting quarterly tourism flows to each of Japan, Australia and New Zealand from seven originating countries over the 1978-95 period. They grouped the quarters into four ‘periodic series’ for each origin-destination pair and applied Holt’s exponential 36 smoothing method and the autoregressive method to the series to develop ex post forecasts. They tested these against Winter’s exponential smoothing method and the Box-Jenkins ARIMA for the straight seasonalized series. They found the models applied to the periodic series generally proved less accurate in ex post forecasting than the ARIMA model or the naïve model applied to the seasonalized series. The Winter’s models proved generally less accurate than the Naïve 1 model for the seasonalized series. 3.6 Tabu Search Tabu Search is a meta-heuristic approach designed to find a near-optimal solution for combinatorial optimization problems. Like other traditional methods, Tabu Search needs strong domain knowledge in order to utilize the procedure. Below are some terminologies used in implementing Tabu Search: A move - A transition from a current solution to its neighbouring (or another) solution. Tabu list – A list of moves that are currently tabu (a list of forbidden exchanges to avoid cycling between the same solutions endlessly) Tabu list size – The number of iterations for which a recently accepted move is not allowed to be reserved, Ts Termination – Tabu Search stops when there is no more move available or until it reaches a preset maximum iteration. Parameter settings – Since Tabu Search is problem-oriented, it requires fine-tuning of its parameters such as searching strategy, sopping rules, initial solution, length of tabu list and maximum iteration. 37 Tabu Search can be thought of as an iterative method. An initial solution is randomly generated and a neighbourhood around that solution is examined. If a new solution in the neighbourhood that is preferred in the original, then the new solution replaces the old and the process repeats. If no new solution is found that improves upon the old function evaluation, then unlike the gradient descent procedure which would stop at that point (a local minimum), the Tabu Search algorithm may continue by accepting a new value that is worse than the old value. Therefore, a collection of solution in a given neighbourhood is generated and the final solution would be based on the best solution so far for that particular neighbourhood. To keep from cycling, an additional step is included that prohibits solutions from recurring for a user defined number of solutions. This Tabu list is generated by adding the last solution to the beginning of the list and discarding the oldest solution from the list. During this procedure, the best solution found so far is retained. From the subset of acceptable neighbourhoods the best solution is chosen. In a problem in which there is no known solution, a given value for a maximum number of iteration can be used to terminate the process. When the number of iterations has taken place with no improvement over the best solution, the algorithm will terminate. According to Hertz and De-Werra (1990), the easiest way to introduce Tabu Search is to think of an iterative descent method used for finding in a (finite or infinite) set X of feasible solutions, s which minimizes a real valued objective function, f. A neighbourhood N ( s) should be defined for each solution s in X; the procedure would then start from an initial feasible solution s. Whenever a feasible solution s has been reached, the neighbourhood N(s) is examined. If a solution s’ in N(s) with f(s’) <f(s) has been found, one move to s’ and repeats the step. If no such s’ is found, the descent procedure stops. In such a case, the method has found a local minimum with respect to the topology chosen on X for defining the neighbourhood N. 38 In contrast to this descent method, Tabu Search may try to continue from there: in order to avoid being trapped in local minima, a movement from an s to an s’ in N(s) even if f(s’) f ( s) is allowed. We therefore generate, when we are at s, a collection of solution si in N(s); we then move to the best si generated. Call it s ; this may, of course introduce cycling in the algorithm. In order to prevent cycling, we do not allow going back to a solution visited in the last k iterations (for a given k). More precisely, we introduced a so-called Tabu list T of length T =k (fixed or variable) which is used as a queue as follows: whenever a move s s ' is made. We should introduce s at the end of T and we should remove the oldest solution introduced into T (in the case where T is fixed). Thus, all moves back to s are now forbidden for the next T iterations, s is a Tabu Solution. Any moving back to s is a Tabu move. 3.6.1 A Basic Tabu Search Procedure 1. Initialization Generate an initial solution, say x (which is not necessarily feasible) Set the best current solution xbest x Evaluate all the moves in the neighbourhood N ( x) Set values for tabu list size and the data structure for candidate list of solutions. Set counters: iteration=0 (current iteration) and best iteration=0 (iteration for which the xbest is found) 39 2. Candidate list of solutions Determine strategically, by the use of a special data structure, the candidate list of the best moves in the neighbourhood N ' ( x ) N ( x) 3. 4. Update the data structure if necessary Selection strategy: Forbidding strategy and freeing strategy Choose the best admissible solution x ' N ' ( x ) Set x x ' and iteration = iteration +1 If F ( x ' ) F ( xbest ) , set xbest x ' and best iteration = iteration Update the tabu list Stopping criterion If the stopping criterion is met, go to step 5, otherwise go to step 2. 5. Diversification (optional) Apply some forms of diversification on well-defined solutions and go to step 2, or else stop. 3.6.2 Tabu Search Application Tabu Search techniques have enjoyed success in a variety of problem settings such as scheduling, transportation, layout and circuit design, electronic circuit design, graphs, probabilistic logic and expert systems, neural networks and others (Reeves, 1993). Tabu Search has the ability to employ any procedure to arrive at a best solution. Furthermore, it has the ability to escape from a local optimum, thereby increasing the chances of reaching global optimum. 40 3.7 Summary This chapter discussed about the forecast accuracy used in forecasting which are MAPE and MSE. Each method also has been elaborated and all equations involved are given in this chapter. Their applications also have been included to give more clearly understanding. CHAPTER 4 EXPERIMENTATION 4.1 Introduction This chapter presents calculation and implementation of forecasting methods which have been discussed in the previous chapter. The calculation begins with Naïve model, followed by Moving Average, Exponential Smoothing and Holt’s model. Tabu Search will be the last to be implemented. Figure 4.1 below shows the number of visitors (in million) to Johor for ten years (120 periods) which covers from January 1999 until December 2008. The graph shows a series with irregular patterns. It follows no recognizable or regular pattern. 42 4,000,000 3,500,000 3,000,000 visitors 2,500,000 2,000,000 1,500,000 1,000,000 500,000 0 1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106 113 120 t Figure 4.1 Number of visitors to Johor for 120 periods (January 1999-December 2008) 4.2 The Naïve Model 4.2.1 The Naïve 1 Model For the Naïve 1 model, the equation that is being used is: Ft At 1 where F = forecast value A = actual value 43 t = some time period To use this equation, the last value is the best forecast of the next value. For example, the value of F2 is the value of the previous actual value, A1 . Thus, F2 A1 F3 A2 F120 A119 which are F2 A1 1213874 F3 A 2 1107735 F120 A119 2217748 From this model, we obtained MAPE = 0.213690271 and MSE = 2.44245138814E11. 44 4,000,000 3,500,000 3,000,000 visitors 2,500,000 2,000,000 1,500,000 1,000,000 500,000 0 1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106 113 120 t Actual Data At Figure 4.2 Forecast value Ft Actual and Forecast Value Using Naïve 1 Model for Johor Tourism 4.2.2 The Naïve 2 Model This model uses an equation: Ft At 1 At 1 At 2 where F = forecast value A = actual value 45 t = some time period The forecast value is obtained by multiplying the current value by the growth rate between the current value and the previous value. For example, F3 A2 A2 A1 F4 A3 A3 1151772 A2 F5 A4 A4 852033 A3 Thus, F120 A119 A119 2079311 A118 MAPE and MSE for this model are 0.421844828 and 1.172248109134E12 respectively. 46 9,000,000 8,000,000 7,000,000 visitors 6,000,000 5,000,000 4,000,000 3,000,000 2,000,000 1,000,000 0 1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106 113 120 t Actual Data At Figure 4.3 Forecast value Ft Actual and Forecast Value Using Naïve 2 Model for Johor Tourism 4.2.3 The Seasonal Naïve Model We used an equation which is: Ft At m where F = forecast value A = actual value t = some time period 47 m = number of periods in a year For this model, we used m 4 and m 12 which represent 4 quarters and 12 months respectively. For example, when m 4 , the calculation becomes: F5 A1 1213874 F6 A2 1107735 F120 A116 2225317 while for m 12 , F13 A1 1213874 F14 A2 1107735 F120 A108 2145665 Table 4.1 Value of MAPE and MSE when m 4 and m 12 m MAPE MSE m=4 0.287878460 3.60123345545E11 m=12 0.254703814 3.18035348453E11 From the table above, we can see that the seasonal model for m=12 gives much smaller MAPE and MSE than when m=4. 48 4.3 Moving Average We can average any number of periods to produce a forecast through moving average model. The general equation we used is: Ft At 1 At 2 At n n where F = forecast value A = actual value t = some time period n = number of past period For this model, we used the average of the previous three values and previous six values to serve as our forecast for the next period. Example for these models is as follows: Calculation example for 3-month moving average model A1 A2 A3 1213874 1107735 1129539 1150383 3 3 A A3 A4 1107735 1129539 981022 F5 2 1072765 3 3 A A118 A119 1866076 2365402 2217748 F120 117 2149742 3 3 F4 49 4,000,000 3,500,000 3,000,000 visitors 2,500,000 2,000,000 1,500,000 1,000,000 500,000 0 1 9 17 25 33 41 49 57 65 73 81 89 97 105 113 t Actual Data At Figure 4.4 Forecast value Ft Actual and Forecast Value Using 3-month Moving Average for Johor Tourism Calculation example for 6-month moving average model A1 A2 A6 1213874 1107735 1240579 1146404 6 6 A A3 A7 1107735 1129539 1334745 F8 2 1166549 6 6 A A115 A119 1770749 1914976 2217748 F120 114 2060045 6 6 F7 50 4,000,000 3,500,000 3,000,000 visitors 2,500,000 2,000,000 1,500,000 1,000,000 500,000 0 1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106 113 120 t Actual Data At 6-month moving average Figure 4.5 Actual and Forecast Value Using 6-month Moving Average for Johor Tourism Table 4.2 Value of MAPE and MSE using 3-month moving average and 6-month moving average Moving average MAPE MSE 3-month moving average 0.202699029 1.89307879049E11 6-month moving average 0.220517655 2.13774356448E11 The smallest MAPE and MSE among these two models is produced by 3-month moving average with MAPE=0.202699029 and MSE=1.89307879049E11. 51 4,000,000 3,500,000 3,000,000 visitors 2,500,000 2,000,000 1,500,000 1,000,000 500,000 0 1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106 113 120 t Actual Data At Figure 4.6 4.4 3-month moving average 6-month moving average Actual and Forecast Value of 3-month Moving Average and 6-month Moving Average for Johor Tourism Exponential Smoothing Exponential smoothing is a technique that used all the preceding observation to determine a smoothed value for a particular time period. Single exponential smoothing that is being used is Ft At 1 (1 ) Ft 1 where F = forecast value A = actual value t = some time period = smoothing constant between 0 and 1 52 The one-step ahead forecast Ft represents a weighted average of all past observations. The starting value for the exponentially weighted average is the number of visitors in the first month of year 1999 (January) which is 1213874. Thus, F1 A1 Then, F2 A1 (1 ) F1 F3 A2 (1 ) F2 Fn An 1 (1 ) Fn 1 for n = 1,2,…,120 For example, when =0.35, F1 A1 1213874 F2 A1 (1 ) F1 0.35(1213874) (1 0.35)(1213874) 1213874 F3 A2 (1 ) F2 0.35(1107735) (1 0.35)(1213874) 1176725 F120 A119 (1 ) F119 0.35(2217748) (1 0.35)(2108095) 2146474 4.4.1 Implementation of Exponential Smoothing Different value of has been used to find the best value of to minimize the value of MAPE and MSE by using the exponential smoothing method. Possible values for the smoothing constant, in 0.05 increments in finding the one that minimizes the MSE and MAPE are shown in the table below. 53 Table 4.3 Values of , MSE and MAPE for Exponential Smoothing Smoothing Constant, MSE MAPE 0.05 2.19519581800E11 0.227142 0.10 1.98058992400E11 0.218336 0.15 1.89277642000E11 0.213353 0.20 1.85128160100E11 0.209261 0.25 1.82917659200E11 0.205530 0.30 1.81694205300E11 0.201835 0.35 1.81156299600E11 0.199749 0.40 1.81244894900E11 0.197842 0.45 1.81983353400E11 0.197005 0.50 1.83416091700 E11 0.196779 0.55 1.85587696400 E11 0.197236 0.60 1.88538524400 E11 0.198176 0.65 1.92306673000 E11 0.199154 0.70 1.96932085000 E11 0.200304 0.75 2.02461112300 E11 0.201745 0.80 2.08951002000 E11 0.203417 0.85 2.16474284900 E11 0.205170 0.90 2.25123289500 E11 0.207034 0.95 2.35015131700 E11 0.210168 1.00 2.46297619000 E11 0.215486 54 From Table 4.3, we can see that the value of MSE decreases until =0.35 and starts to increase when =0.40. For MAPE, the value also decreases until =0.50 and increases when =0.55. The smallest MSE and MAPE from this table is when =0.35 and =0.50 respectively. The search then continues until it found that the best value for MSE is 1.81116159000E11 when =0.37. While for MAPE, the best value is 0.19673 when =0.48. 4,000,000 3,500,000 visitors 3,000,000 2,500,000 2,000,000 1,500,000 1,000,000 500,000 0 1 8 15 22 29 36 43 50 57 64 71 78 85 92 99 106 113 120 t Actual Data At Figure 4.7 alpha=0.05 alpha=0.20 alpha=0.37 Actual and Forecast Value Using Exponential Smoothing With 0.05 , 0.20 and 0.37 55 4.5 Holt’s Model When a time series contains a trend component, exponential smoothing involves two updating steps, one for the smoothed estimate and one for a trend estimate. The updating formula for the smoothed estimate is as follows: at At (1 )(at 1 bt 1 ) where at = the smoothed estimate for the current period t at 1 = the smoothed estimate for the preceding period t-1 At = the actual data point for the current period t = smoothing constant between 0 and 1 While the updating formula for the trend estimate is: bt (at at 1 ) (1 )bt 1 where bt = the trend estimate for the current period t bt 1 = the trend estimate for the preceding period t-1 = the smoothing constant between 0 and 1 In Holt’s method, we begin by calculating the initial value, a and b . Equations below are used to calculate the initial trend value, b and overall smoothed value, a respectively. b Am A1 (m 1) L where L = the period 56 m = the number of year A1 = the average of data for year 1 Am = the average of data for year m b is calculated as follows: b Am A1 A10 A1 2001828 1276890 6712.383 (m 1) L (10 1)12 108 Next, calculate the initial overall smoothed value, a : a A1 L 12 b 1276890 (6712.383) 1236616 2 2 Using the initial value a =1236616 and b =6712.383, the update estimates, a1 and b1 as shown below: at At (1 )(at 1 bt 1 ) a1 0.36 *(1213874) (1 0.36) *(1236616 6712.383) a1 1232724.8 bt (at at 1 ) (1 )bt 1 b1 0 * (1232724.8 1236616) (1 0) * 6712.383 b1 6712.383 Then, the forecast made in F2 is F2 a1 b1 F2 1239437 57 Next, we can obtain the updated estimates, a2 and b2 using the same way we did to a1 and b1 which is as follows: at At (1 )(at 1 bt 1 ) a2 0.36 * (1107735) (1 0.36)* (1232724.8 6712.383) a2 1192024.4 bt (at at 1 ) (1 )bt 1 b2 0 *(1192024.4 1232724.8) (1 0) * (6712.383) b2 6712.383 Since we have calculated a2 and b2 , the forecast F3 is F3 a2 b2 F3 11920.24 6712.383 F3 1198737 The procedure continuous through the entire 120 periods of historical data. The forecast value for time period t+1 is the current smoothed values plus the current smoothed trend value. An equation below is used to forecast m number of time periods into the future. Ft m at mbt (m=1, 2, 3…) Hence, the forecast for one period ahead is, for F121 : F1201 a120 (1) * (b120 ) F121 1989607.3 (1) * (6712.383) F121 1996320 58 For F122 : F120 2 a120 (2)* (b120 ) F122 1989607.3 (2)* (6712.383) F122 2003032 and so on. Referring to Appendix G , we knew that by using 0.36 and 0.00 , the value of MSE and MAPE are 1.79321039826E11 and 0.201822197 respectively. The search continues and found that the best value of and are 0.48 and 0.00 for MSE. For MAPE, 0.36 and 0.00 are the best value found. However, the value of from the results indicates that the data does not show a linear trend. Thus, Holt’s method is not an appropriate method for the data. 4.6 Tabu Search Application The following characteristics will be examined to get the best value of , and to minimize MSE and MAPE: i. an initial solution ii. the tabu list size, Ts iii. the neighbourhood size In this study, the neighbourhood size used is 1 and the tabu list size is 100. The Automated Computerized Forecasting System has been run to obtain the best value of , and to minimize MSE and MAPE. 59 4.6.1 Forecasting Graph Using Tabu Search Figure 4.8 The Forecasting Graph Using Tabu Search in Finding MSE Figure 4.7 shows the parameter values of , and found using Tabu Search which are =0.49, =0.00 and =0.00 with MSE=1.52074940277814E11. Since the values of and are zeroes, it clearly shows that Holt-Winter’s method is not suitable to be applied into Johor tourism data. 60 Figure 4.9 The Forecasting Graph Using Tabu Search in Finding MAPE The same also goes for MAPE. The parameter values of =0.69, =0.00 and =0.00 with MAPE=0.182282 are found which is shown in Figure 4.9 above. Thus, the result strictly declines that the data cannot be applied using Holt-Winter’s method because the data pattern does not show any linear trend and seasonality. 61 4.7 Summary In this chapter, the forecast has been done using the naïve model, moving average and single exponential smoothing. While from the result of Holt’s method and Tabu Search, we can ensure that the data shows irregular patterns (no linear trend and seasonality). CHAPTER 5 CONCLUSION AND RECOMMENDATION 5.1 Introduction This chapter summarized the results and findings obtained in Chapter 4. From these observations, conclusion is drawn. Finally, some recommendations for future research in this area which might be pursued are offered. 5.2 Conclusion In this study, we have examined four time series forecast which are the naïve model, moving average, single exponential smoothing and Holt’s method. Tabu Search also has been applied in order to gain the objectives of this study. Each method has advantages and disadvantages in dealing with particular time series. 63 From Table 5.1 below, the minimum MAPE and MSE for Naïve model equal to 0.213690271 and 2.44245138814E11 (both belong to Naïve 1). For seasonal Naïve, it produces the smallest MAPE and MSE when m=12 i.e. MAPE=0.254703814 and MSE=3.18035348453E11. While for moving average, the minimum MAPE=0.202699029 and MSE=1.89307879049, which come from 3-month moving average. For exponential smoothing, when =0.48, it shows the smallest MAPE and when =0.37, it shows the smallest MSE with the value are 0.196703000 and 1.81116159000E11 respectively. For Holt’s method, its initial overall value, a 1236616 and trend, b 6712.383 . From this method, we obtained the smallest MAPE=0.197288000 when =0.48 and =0.00 while the smallest MSE=1.79328268900 when =0.36 and =0.00. Nevertheless, the Holt’s method cannot be used as the forecast method for this data since the value of =0.00 which means the data has no linear trend. Lastly, using Tabu Search, we get the smallest MAPE=0.182282000 when =0.69, =0.00 and =0.00 and when =0.49, =0.00 and =0.00, it gives the smallest MSE=1.52074940277. From the value of and , it certainly confirmed that the data has no linear trend and seasonality. Hence, Holt-Winter’s is also not an appropriate method to be applied. Comparison of the MAPE and MSE among the naïve model, moving average and single exponential smoothing clearly indicates that single exponential smoothing gives the best fit while the Naïve 2 model gives the poorest fit. 64 Table 5.1 Comparison of the Selected Model to Forecast Johor Tourism for 120 periods Method Characteristics MAPE MSE 1. Naïve 1 0.213690271 2.44245138814E11 2. Naïve 2 0.421844828 1.172248109134E12 m=4 0.287878460 3.60123345545E11 m=12 0.254703814 3.18035348453E11 3-month 0.202699099 1.89307879049E11 0.220517655 2.13774356448E11 =0.35 0.199749000 1.81156299600E11 Exponential =0.37 0.198978000 1.81116159000E11 Smoothing =0.48 0.196703000 1.82756737700E11 =0.50 0.196779000 1.83416091700E11 3. Seasonal Naïve moving average 4. Moving Average 6-month moving average 5. Single 65 5.3 Recommendation In this study, it focuses on forecasting technique i.e. the Naïve, moving average, single exponential smoothing and Holt’s method. For further studies, these models can be extended to cover more regions. Better method should also be used as a comparison to these models such as Box-Jenkins, ARIMA, Neural Network and others. Besides that, combining quantitative and qualitative forecasting technique could be an especially effective way of achieving convergent validity. Related to this concept is the value of combining forecasts from different models into a single forecast. Current and previous values of other related factors which could provide additional useful information, such as economic variables, also could be incorporated into forecast of visitors in future studies. This study also focuses on Tabu Search which helps to ensure the pattern of the data. Other heuristic method could be involved in the future such as Simulated Annealing, Neural Network or Evolutionary Algorithm. 66 REFERENCES Archer, B. H. (1980). Forecasting Demand: Quantitative and Intuitive Technique. International Journal of Tourism Management, 1(1), 5-12. Archer, B. H. (1987). Demand Forecasting and Estimation. In J. R. B. Ritchie, and C. Goeldner (Eds.), Travel Tourism and Hospitality Research. New York: Wiley. Billah, B., King, M. L., Synder, R.D. & Koehler, A. B. (2006). Exponential Smoothing Model Selection for Forecasting. International Journal of Forecasting, 22, 239-247. Bowerman, B. L. and O’Connell, R. T. (1993). Forecasting and Time-Series. Duxbury Press. Brealy, R. A. and Myers, S. C. (1988). Principles of Corporate Finance. 3rd ed. New York: McGraw-Hill. Brown, R. G. (1956). Exponential Smoothing for Predicting Demand. Presented At the Tenth National Meeting of the Operations Research Society of America, San Francisco, 16 November 1956. Brown, R. G. (1963). Smoothing, Forecasting and Prediction. Englewood Cliffs, N. J.: Prentice Hall. 67 Brown, R. G. and Meyer, R. F. (1961). The Fundamental Theorem of Exponential Smoothing. Operations Research, 9, No. 5, 673-685. Burger, C. J. S. C., Dohnal, M., Kathrada, M. and Law, R. (2001). A Practitioners Guide to Time-Series Methods for Tourism Demand Forecasting- A Case Study of Durban, South Africa. Tourism Management, 22, 403-409. Chatfield, C. (1984). The Analysis of Time Series: An Introduction. 3rd ed. London: Chapman & Hall. Chu, F,-L. (1998). Forecasting Tourism Demand in Asian-Pacific Countries. Annals of Tourism Research, 25(3), 597-615. Dalrymple, K. and Greenidge, K. (1999). Forecasting Arrivals to Barbados. Annals of Tourism Research, 26(1), 188-218. De Gooijer, J. G. and Franses, P. H. (1997). Forecasting and Seasonality. International Journal of Forecasting, 13, 303-305. Faulkner, B. and Valerio, P. (1995). An Integrative Approach to Tourism Demand Forecasting. Tourism Management, 16(1), 29-37. Frechtling, D. C. (2001). Forecasting Tourism Demand: Methods and Strategies. Butterworth Heinemann. Gardner E. S., Jr. (2006). Exponential Smoothing: The State of Art-Part II. International Journal of Forecasting, 22, 637-666. Gardner E. S., Jr. and Diaz-Saiz, Jr. (2008). Exponential Smoothing in the Telecommunications Data. International Journal of Forecasting, 24, 170174. 68 Hertz, A. and De-Werra, D. (1990). The Tabu Search Metaheuristics: How We Used It. Annals of Mathematics and Artificial Intelligence. 111-121. Holt, C. C., Modigliani, F., Muth, J. F. & Simon, H. A. (1960). Planning Production Inventories and Work Force. Englewood Cliffs, N. J.: Prentice Hall. Holt, C.C. (1957). Forecasting Seasonal and Trends By Exponentially Weighted Moving Averages. Office of Naval Research, Research Memorandum. Li, G., Song, H. and Witt, S. F. (2005). Recent Developments in Econometric Modeling and Forecasting. Journal of Travel Research, 44, 82-99. Li, G., Song, H. and Witt, S. F. (2006). Time Varying Parameter and Fixed Parameter Linear AIDS: Application to Tourism Demand Forecasting. International Journal of Forecasting, 22, 57-71. Lim, C. and McAleer, M. (2001). Forecasting Tourist Arrivals. Annals of Tourism Research, 28(4), 965-977. Lim, C. and McAleer, M. (2003). Time-Series Forecasts of International Travel Demand for Australia. Tourism Management, 23, 389-396. M. A. Umar. (2007). Comparative Study of Holt-Winter, Double Exponential and the Linear Trend Regression Models, with Application to Exchange Rates of the Naira to the Dollar. Research Journal of Applied Science, 2(5), 633-637. Magee, J. F. (1958). Production Planning and Inventory Control. New York: McGraw Hill. 69 Makridakis, S., Wheelwright, S. C. and Hyndman, R. I. (1998). Forecasting: Methods and Applications. 3rd ed. John Wiley & Sons, Inc. Martin, C. A. and Witt, S. F. (1989). Accuracy of Econometric Forecasts of Tourism. Annals of Tourism Research, 16, 407-428. Mohd Nizam, Mohmad Kahar. (2003). Genetic Algorithm Approach in Improving Winter’s Method of Forecasting. Universiti Teknologi Malaysia: Tesis Sarjana. N. Kulendran and King, L. K. (1997). Forecasting International Quarterly Tourists Flows Using Error-Connection and Time-Series Models. International Journal of Forecasting, 13, 319-327. Naisbitt, J. (1994). Global Paradox. Morrow. Nazem, S. M. (1988). Applied Time-Series Analysis for Business and Economic Forecasting. Marcel Dekker, Inc. Nurul Ain Zhafarina, Muhammad. (2009). Modeling of Sales and Operations Planning (S & OP). Universiti Teknologi Malaysia: Thesis, M. Sc. Reeves, C. R. (1993). Modern Heuristic Techniques for Combinatorial Problems. Blackwell Scientific Publications. Rosas, A. L. and Guerrero, V. M. (1994). Restricted Forecasts Using Exponential Smoothing Techniques. International Journal of Forecasting, 10, 515-527. Runyon, R. P. and Haber, A. (1982). Business: Statistics. Homewood III: Richard D Irwin. 70 Sheldon, P. (1993). Forecasting Tourism: Expenditures versus Arrivals. Journal of Travel Research. 22(1), Summer, 13-20. Song, H. and Witt, S. F. (2006). Forecasting International Tourists Flows to Macau. Tourism Management, 27, 214-224. Song, H., Witt, S. F. and Jensen, T. C. (2003). Tourism Forecasting: Accuracy of Alternative Econometric Models. International Journal of Forecasting, 19, 123-141. Song, H., Witt, S. F. and Li, G. (2003). Modeling and Forecasting the Demand for Thai Tourism. Tourism Economics 9, 363-387. Study of Durban, South Africa. Tourism Management, 22, 403-409. Turner, L. W., Kulendran, N. and Fernando, H. (1997). Univariate Modeling Using Periodic and Non-Periodic Analysis: Inbound Tourism to Japan, Australia and New Zealand Compared. Tourism Economics, 3(1), March, 39-56. Uysal, M. and Crompton, J. L. (1985). An Overview of Approaches Used to Forecast Tourism Demand. Journal of Travel Research, 23, 7-15. Wandner, S. A. and Van Erden, J. D. (1980). Estimating the Demand for International Tourism Using Time Series Analysis. In D. E. Hawkins, E. L. Shafer & J. M. Rovelstad (Eds.). Tourism Planning and Development Issues. Washington DC: George Washington University. Winters, P. R. (1960). Forecasting Sales By Exponentially Weighted Moving Averages. Management Science, 6, 324-342. Witt, S. F. and Witt, C. A. (1992). Modeling and Forecasting Demand in Tourism. Academic Press. 71 Witt, S. F. and Witt, C. A. (1995). Forecasting Tourism Demand: A Review of Empirical Research. International Journal of Forecasting, 11, 447-475. Witt, S. F., Newbould, G. D. and Watkins, A. J. (1992). Forecasting Domestic Tourism Demand: Application to Las Vegas Arrivals Data. Journal of Travel Research, 31(1), Summer, 36-41. Wong, K. F., Song, H., Witt, S. F. & Wu, D. C. (2007). Tourism Forecasting: To Combine or Not To Combine? Tourism Management, 28, 1068-1078. World Tourism Organization (WTO). (1995). Concepts, Definitions and Classifications for Tourism Statistics: Technical Manual No.1. WTO. 72 APPENDIX A Johor Tourism Data (Data is taken from Majlis Tindakan Pelancongan Negeri Johor) Table 1.1 Period 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 Number of Visitors in Johor Year Month Jan Feb Mar Apr May 1999 June July Aug Sept Oct Nov Dec Jan Feb Mar Apr May 2000 June July Aug Sept Oct Nov Dec Jan Feb No. of Visitors 1,213,874 1,107,735 1,129,539 981,022 1,205,673 1,240,579 1,334,745 1,514,357 1,213,478 1,474,035 1,514,357 1,393,287 1,088,378 1,080,564 1,109,080 1,049,116 1,013,572 1,205,808 1,109,254 2,146,602 953,087 952,753 930,598 1,193,784 1,080,693 944,095 73 27 28 29 30 31 32 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 2001 2002 2003 2004 Mar Apr May June July Aug Sept Oct Nov Dec Jan Feb Mar Apr May June July Aug Sept Oct Nov Dec Jan Feb Mar Apr May June July Aug Sept Oct Nov Dec Jan Feb Mar Apr May June 1,305,077 1,217,528 1,190,167 1,345,568 1,120,333 1,393,845 981,729 969,556 962,730 1,239,244 1,234,351 1,148,229 1,333,280 985,363 914,062 1,035,614 1,270,599 1,911,344 1,505,878 2,590,348 1,747,456 3,450,757 1,057,455 1,171,256 1,087,495 1,083,396 1,277,491 1,274,229 1,133,624 1,103,639 2,103,175 2,188,585 1,148,499 1,545,169 1,333,513 1,145,025 1,270,344 889,574 919,763 1,129,101 74 67 68 69 70 71 72 73 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 2005 2006 2007 July Aug Sept Oct Nov Dec Jan Feb Mar Apr May June July Aug Sept Oct Nov Dec Jan Feb Mar Apr May June July Aug Sept Oct Nov Dec Jan Feb Mar Apr May June July Aug Sept Oct 870,424 923,640 994,139 2,846,747 906,309 1,068,855 1,492,337 840,330 1,016,850 887,480 1,003,125 907,830 979,130 859,112 794,946 600,394 814,820 1,073,763 949,066 1,173,758 1,212,861 1,649,596 2,204,547 2,528,855 2,136,125 1,750,787 1,452,891 975,078 1,302,898 1,364,959 1,544,982 1,863,259 1,806,980 1,806,074 2,229,541 2,002,932 2,147,831 2,164,777 2,057,634 2,224,213 75 107 108 109 110 111 112 113 114 115 116 117 118 119 120 2008 Nov Dec Jan Feb Mar Apr May June July Aug Sept Oct Nov Dec 1,979,464 2,145,665 2,345,687 1,609,402 2,087,459 1,927,879 2,019,837 1,770,749 1,914,976 2,225,317 1,866,076 2,365,402 2,217,748 1,671,398 76 APPENDIX B Table 4.4 Calculation for Johor Tourism Data using Naïve 1 Model (Assume F2 A1 ) Period Actual Data Forecast value Error t 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 33 At 1,213,874 1,107,735 1,129,539 981,022 1,205,673 1,240,579 1,334,745 1,514,357 1,213,478 1,474,035 1,514,357 1,393,287 1,088,378 1,080,564 1,109,080 1,049,116 1,013,572 1,205,808 1,109,254 2,146,602 953,087 952,753 930,598 1,193,784 1,080,693 944,095 1,305,077 1,217,528 1,190,167 1,345,568 1,120,333 1,393,845 981,729 Ft et Square Error Absolute Percentage Error 1,213,874 -106,139 11265487321 0.095816238 1,107,735 21,804 475414416 0.01930345 1,129,539 -148,517 22057299289 0.151390081 981,022 224,651 50468071801 0.1863283 1,205,673 34,906 1218428836 0.028136862 1,240,579 94,166 8867235556 0.070549805 1,334,745 179,612 32260470544 0.118606115 1,514,357 -300,879 90528172641 0.247947635 1,213,478 260,557 67889950249 0.176764459 1,474,035 40,322 1625863684 0.026626482 1,514,357 -121,070 14657944900 0.086895234 1,393,287 -304,909 92969498281 0.280149911 1,088,378 -7,814 61058596 0.007231409 1,080,564 28,516 813162256 0.0257114 1,109,080 -59,964 3595681296 0.057156692 1,049,116 -35,544 1263375936 0.035068056 1,013,572 192,236 36954679696 0.159425049 1,205,808 -96,554 9322674916 0.087044085 1,109,254 1,037,348 1076090873104 0.483251204 2,146,602 -1,193,515 1424478055225 1.252262385 953,087 -334 111556 0.000350563 952,753 -22,155 490844025 0.023807272 930,598 263,186 69266870596 0.220463668 1,193,784 -113,091 12789574281 0.104646741 1,080,693 -136,598 18659013604 0.144686711 944,095 360,982 130308004324 0.276598239 1,305,077 -87,549 7664827401 0.071907176 1,217,528 -27,361 748624321 0.022989211 1,190,167 155,401 24149470801 0.115491005 1,345,568 -225,235 50730805225 0.201042904 1,120,333 273,512 74808814144 0.196228419 1,393,845 -412,116 169839597456 0.419785908 77 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 74 75 76 77 78 79 969,556 962,730 1,239,244 1,234,351 1,148,229 1,333,280 985,363 914,062 1,035,614 1,270,599 1,911,344 1,505,878 2,590,348 1,747,456 3,450,757 1,057,455 1,171,256 1,087,495 1,083,396 1,277,491 1,274,229 1,133,624 1,103,639 2,103,175 2,188,585 1,148,499 1,545,169 1,333,513 1,145,025 1,270,344 889,574 919,763 1,129,101 870,424 923,640 994,139 2,846,747 906,309 1,068,855 1,492,337 840,330 1,016,850 887,480 1,003,125 907,830 979,130 981,729 -12,173 148181929 0.012555231 969,556 -6,826 46594276 0.007090254 962,730 276,514 76459992196 0.223131199 1,239,244 -4,893 23941449 0.003964026 1,234,351 -86,122 7416998884 0.075004202 1,148,229 185,051 34243872601 0.138793802 1,333,280 -347,917 121046238889 0.353085107 985,363 -71,301 5083832601 0.078004555 914,062 121,552 14774888704 0.117371917 1,035,614 234,985 55217950225 0.184940331 1,270,599 640,745 410554155025 0.335232695 1,911,344 -405,466 164402677156 0.269255544 1,505,878 1,084,470 1176075180900 0.418658034 2,590,348 -842,892 710466923664 0.482353776 1,747,456 1,703,301 2901234296601 0.493602128 3,450,757 -2,393,302 5727894463204 2.263266049 1,057,455 113,801 12950667601 0.097161509 1,171,256 -83,761 7015905121 0.077021963 1,087,495 -4,099 16801801 0.003783473 1,083,396 194,095 37672869025 0.151934534 1,277,491 -3,262 10640644 0.002559979 1,274,229 -140,605 19769766025 0.124031425 1,133,624 -29,985 899100225 0.02716921 1,103,639 999,536 999072215296 0.47525099 2,103,175 85,410 7294868100 0.039025215 2,188,585 -1,040,086 1081778887396 0.90560462 1,148,499 396,670 157347088900 0.256716256 1,545,169 -211,656 44798262336 0.158720612 1,333,513 -188,488 35527726144 0.164614746 1,145,025 125,319 15704851761 0.098649657 1,270,344 -380,770 144985792900 0.428036341 889,574 30,189 911375721 0.032822586 919,763 209,338 43822398244 0.185402369 1,129,101 -258,677 66913790329 0.29718505 870,424 53,216 2831942656 0.057615521 923,640 70,499 4970109001 0.070914631 994,139 1,852,608 3432156401664 0.650780698 2,846,747 -1,940,438 3765299631844 2.141033577 906,309 162,546 26421202116 0.152074884 1,068,855 423,482 179337004324 0.283771025 1,492,337 -652,007 425113128049 0.775893994 840,330 176,520 31159310400 0.173594926 1,016,850 -129,370 16736596900 0.145772299 887,480 115,645 13373766025 0.115284735 1,003,125 -95,295 9081137025 0.104970094 907,830 71,300 5083690000 0.072819748 78 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 859,112 794,946 600,394 814,820 1,073,763 949,066 1,173,758 1,212,861 1,649,596 2,204,547 2,528,855 2,136,125 1,750,787 1,452,891 975,078 1,302,898 1,364,959 1,544,982 1,863,259 1,806,980 1,806,074 2,229,541 2,002,932 2,147,831 2,164,777 2,057,634 2,224,213 1,979,464 2,145,665 2,345,687 1,609,402 2,087,459 1,927,879 2,019,837 1,770,749 1,914,976 2,225,317 1,866,076 2,365,402 2,217,748 1,671,398 979,130 859,112 794,946 600,394 814,820 1,073,763 949,066 1,173,758 1,212,861 1,649,596 2,204,547 2,528,855 2,136,125 1,750,787 1,452,891 975,078 1,302,898 1,364,959 1,544,982 1,863,259 1,806,980 1,806,074 2,229,541 2,002,932 2,147,831 2,164,777 2,057,634 2,224,213 1,979,464 2,145,665 2,345,687 1,609,402 2,087,459 1,927,879 2,019,837 1,770,749 1,914,976 2,225,317 1,866,076 2,365,402 2,217,748 -120,018 -64,166 -194,552 214,426 258,943 -124,697 224,692 39,103 436,735 554,951 324,308 -392,730 -385,338 -297,896 -477,813 327,820 62,061 180,023 318,277 -56,279 -906 423,467 -226,609 144,899 16,946 -107,143 166,579 -244,749 166,201 200,022 -736,285 478,057 -159,580 91,958 -249,088 144,227 310,341 -359,241 499,326 -147,654 -546,350 14404320324 4117275556 37850480704 45978509476 67051477249 15549341809 50486494864 1529044609 190737460225 307970612401 105175678864 154236852900 148485374244 88742026816 228305262969 107465952400 3851567721 32408280529 101300248729 3167325841 820836 179324300089 51351638881 20995720201 287166916 11479622449 27748563241 59902073001 27622772401 40008800484 542115601225 228538495249 25465776400 8456273764 62044831744 20801427529 96311536281 129054096081 249326454276 21801703716 298498322500 0.139700062 0.080717432 0.324040547 0.263157507 0.241154705 0.131389176 0.191429579 0.032240298 0.264752703 0.251730174 0.128243019 0.1838516 0.220094163 0.205036716 0.490025413 0.251608338 0.0454673 0.116521099 0.170817369 0.031145336 0.000501641 0.18993461 0.113138639 0.067462943 0.007828058 0.052070971 0.074893457 0.123644077 0.07745897 0.085272246 0.457489801 0.229013839 0.082774904 0.045527436 0.140668158 0.075315304 0.139459232 0.192511452 0.211095619 0.066578349 0.326882047 79 APPENDIX C Table 4.5 Calculation for Johor Tourism Data using Naïve 2 Model Period Actual Data Forecast value Error t 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 33 34 35 At 1,213,874 1,107,735 1,129,539 981,022 1,205,673 1,240,579 1,334,745 1,514,357 1,213,478 1,474,035 1,514,357 1,393,287 1,088,378 1,080,564 1,109,080 1,049,116 1,013,572 1,205,808 1,109,254 2,146,602 953,087 952,753 930,598 1,193,784 1,080,693 944,095 1,305,077 1,217,528 1,190,167 1,345,568 1,120,333 1,393,845 981,729 969,556 962,730 Ft et 1010877 1151772 852033 1481768 1276496 1436059 1718139 972379 1790539 1555782 1281896 850196 1072806 1138349 992394 979232 1434504 1020431 4154053 423169 952419 908958 1531403 978315 824763 1804083 1135852 1163421 1521260 932800 1734131 691463 957534 118,662 -170,750 353,640 -241,189 58,249 78,298 -504,661 501,656 -276,182 -162,495 -193,518 230,368 36,274 -89,233 21,178 226,576 -325,250 1,126,171 -3,200,966 529,584 -21,821 284,826 -450,710 -34,220 480,314 -586,555 54,315 182,147 -400,927 461,045 -752,402 278,093 5,196 Square Error Absolute Percentage Error 14080763556 29155622955 125061461434 58172318620 3392994948 6130630216 254682485885 251658801182 76276267973 26404625704 37449339895 53069535584 1315795782 7962445307 448506006 51336579994 105787462751 1268260050428 10246184717954 280459505242 476161149 81125744908 203139184120 1171040866 230701744584 344047061328 2950110054 33177576106 160742319330 212562354634 566108364505 77335882563 26999042 0.105053826 0.174053362 0.293313609 0.194416787 0.04364086 0.051704018 0.415879615 0.340328458 0.18237548 0.116627086 0.177804328 0.213192611 0.032706297 0.085054975 0.020894382 0.187903689 0.293214941 0.524629402 3.358524685 0.555846348 0.023448489 0.238590746 0.417056134 0.036246855 0.368035154 0.481759146 0.045636382 0.135368207 0.357863979 0.330771966 0.766404711 0.286825411 0.005397214 80 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 74 75 76 77 78 79 80 81 1,239,244 1,234,351 1,148,229 1,333,280 985,363 914,062 1,035,614 1,270,599 1,911,344 1,505,878 2,590,348 1,747,456 3,450,757 1,057,455 1,171,256 1,087,495 1,083,396 1,277,491 1,274,229 1,133,624 1,103,639 2,103,175 2,188,585 1,148,499 1,545,169 1,333,513 1,145,025 1,270,344 889,574 919,763 1,129,101 870,424 923,640 994,139 2,846,747 906,309 1,068,855 1,492,337 840,330 1,016,850 887,480 1,003,125 907,830 979,130 859,112 794,946 955952 1595178 1229477 1068116 1548154 728234 847920 1173330 1558903 2875208 1186426 4455808 1178839 6814320 324048 1297304 1009724 1079312 1506359 1270975 1008534 1074447 4007964 2277464 602695 2078841 1150849 983179 1409379 622935 950977 1386084 671010 980110 1070019 8151746 288538 1260554 2083603 473187 1230450 774569 1133839 821588 1056030 753805 283,292 -360,827 -81,248 265,164 -562,791 185,828 187,694 97,269 352,441 -1,369,330 1,403,922 -2,708,352 2,271,918 -5,756,865 847,208 -209,809 73,672 198,179 -232,130 -137,351 95,105 1,028,728 -1,819,379 -1,128,965 942,474 -745,328 -5,824 287,165 -519,805 296,828 178,124 -515,660 252,630 14,029 1,776,728 -7,245,437 780,317 231,783 -1,243,273 543,663 -342,970 228,556 -226,009 157,542 -196,918 41,141 80254324780 130196105636 6601289405 70312039817 316733944099 34531930111 35228906248 9461260375 124214626246 1875063521070 1970996490925 7335168835368 5161612756492 33141491354415 717761310469 44019822916 5427553309 39274737679 53884311153 18865387669 9044945636 1058281050439 3310138955302 1274560848286 888256623645 555514384263 33924424 82463619734 270196955531 88106785955 31728334634 265905560896 63821971598 196826284 3156762377670 52496355357759 608893898162 53723576042 1545728303887 295569427783 117628318049 52237727020 51080236111 24819530010 38776630180 1692553402 0.228600617 0.292321207 0.070759682 0.198881087 0.571151148 0.20329878 0.181239004 0.076553665 0.184394308 0.909323058 0.541981936 1.549882619 0.658382581 5.444075366 0.723332858 0.192928717 0.068000925 0.155131073 0.182172863 0.121161275 0.086173938 0.4891309 0.831303663 0.982991281 0.609948603 0.558920965 0.00508676 0.226052782 0.584329947 0.322722128 0.157757802 0.592424285 0.273515773 0.014112191 0.624125712 7.99444435 0.730049012 0.155315768 1.479505934 0.534654052 0.38645361 0.22784373 0.24895561 0.160900139 0.22921089 0.051752765 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 600,394 814,820 1,073,763 949,066 1,173,758 1,212,861 1,649,596 2,204,547 2,528,855 2,136,125 1,750,787 1,452,891 975,078 1,302,898 1,364,959 1,544,982 1,863,259 1,806,980 1,806,074 2,229,541 2,002,932 2,147,831 2,164,777 2,057,634 2,224,213 1,979,464 2,145,665 2,345,687 1,609,402 2,087,459 1,927,879 2,019,837 1,770,749 1,914,976 2,225,317 1,866,076 2,365,402 2,217,748 1,671,398 735572 453456 1105827 1414996 838850 1451646 1253267 2243593 2946193 2900872 1804386 1434961 1205682 654404 1740931 1429976 1748748 2247103 1752401 1805168 2752298 1799355 2303212 2181857 1955794 2404278 1761647 2325821 2564355 1104229 2707518 1780498 2116181 1552379 2070950 2585952 1564829 2998338 2079311 -135,178 361,364 -32,064 -465,930 334,908 -238,785 396,329 -39,046 -417,338 -764,747 -53,599 17,930 -230,604 648,494 -375,972 115,006 114,511 -440,123 53,673 424,373 -749,366 348,476 -138,435 -124,223 268,419 -424,814 384,018 19,866 -954,953 983,230 -779,639 239,339 -345,432 362,597 154,367 -719,876 800,573 -780,590 -407,913 18273220976 130584011537 1028071958 217090695652 112163256301 57018266125 157076920750 1524623188 174170645566 584837242782 2872829809 321500794 53178174824 420544982121 141354698618 13226344511 13112759223 193708490967 2880803284 180092057579 561548748573 121435256689 19164385657 15431279416 72048806674 180466650146 147469973329 394670957 911935882795 966741914227 607837156772 57282958783 119323480372 131476771546 23829094144 518221254889 640917803245 609320781494 166392944341 0.225149615 0.44348948 0.029860929 0.49093522 0.285329542 0.196877448 0.240258408 0.017711767 0.165030248 0.358006447 0.03061411 0.012341217 0.236497937 0.497732283 0.275445396 0.074438308 0.061457348 0.243568422 0.029718115 0.190340768 0.374134301 0.162245362 0.063949078 0.060371621 0.120680478 0.214610453 0.178973975 0.0084693 0.593359111 0.4710178 0.404402517 0.11849401 0.195076947 0.1893482 0.069368432 0.385769851 0.338451317 0.351974174 0.244054924 82 APPENDIX D Table 4.6 Calculation for Johor Tourism Data using Seasonal Naïve Model (when m=4) Period Actual Data Year t 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 33 At 1,213,874 1,107,735 1,129,539 981,022 1,205,673 1,240,579 1,334,745 1,514,357 1,213,478 1,474,035 1,514,357 1,393,287 1,088,378 1,080,564 1,109,080 1,049,116 1,013,572 1,205,808 1,109,254 2,146,602 953,087 952,753 930,598 1,193,784 1,080,693 944,095 1,305,077 1,217,528 1,190,167 1,345,568 1,120,333 1,393,845 981,729 1 1 1 1 2 2 2 2 3 3 3 3 4 4 4 4 5 5 5 5 6 6 6 6 7 7 7 7 8 8 8 8 9 Forecast value Error Ft et 1,213,874 1,107,735 1,129,539 981,022 1,205,673 1,240,579 1,334,745 1,514,357 1,213,478 1,474,035 1,514,357 1,393,287 1,088,378 1,080,564 1,109,080 1,049,116 1,013,572 1,205,808 1,109,254 2,146,602 953,087 952,753 930,598 1,193,784 1,080,693 944,095 1,305,077 1,217,528 1,190,167 -8,201 132,844 205,206 533,335 7,805 233,456 179,612 -121,070 -125,100 -393,471 -405,277 -344,171 -74,806 125,244 174 1,097,486 -60,485 -253,055 -178,656 -952,818 127,606 -8,658 374,479 23,744 109,474 401,473 -184,744 176,317 -208,438 Square Error Absolute Percentage Error 67256401 17647528336 42109502436 284446222225 60918025 54501703936 32260470544 14657944900 15650010000 154819427841 164249446729 118453677241 5595937636 15686059536 30276 1204475520196 3658435225 64036833025 31917966336 907862141124 16283291236 74960964 140234521441 563777536 11984556676 161180569729 34130345536 31087684489 43446399844 0.00680201 0.107082258 0.153741726 0.352185779 0.006431925 0.158378872 0.118606115 0.086895234 0.114941684 0.364134841 0.365417283 0.328058098 0.073804328 0.103867282 0.000156862 0.511266644 0.063462202 0.265603992 0.191979781 0.798149414 0.118077937 0.009170687 0.286940158 0.01950181 0.09198205 0.298366935 0.164900971 0.126496849 0.212317248 83 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 74 75 76 77 78 79 969,556 962,730 1,239,244 1,234,351 1,148,229 1,333,280 985,363 914,062 1,035,614 1,270,599 1,911,344 1,505,878 2,590,348 1,747,456 3,450,757 1,057,455 1,171,256 1,087,495 1,083,396 1,277,491 1,274,229 1,133,624 1,103,639 2,103,175 2,188,585 1,148,499 1,545,169 1,333,513 1,145,025 1,270,344 889,574 919,763 1,129,101 870,424 923,640 994,139 2,846,747 906,309 1,068,855 1,492,337 840,330 1,016,850 887,480 1,003,125 907,830 979,130 9 9 9 10 10 10 10 11 11 11 11 12 12 12 12 13 13 13 13 14 14 14 14 15 15 15 15 16 16 16 16 17 17 17 17 18 18 18 18 19 19 19 19 20 20 20 1,345,568 1,120,333 1,393,845 981,729 969,556 962,730 1,239,244 1,234,351 1,148,229 1,333,280 985,363 914,062 1,035,614 1,270,599 1,911,344 1,505,878 2,590,348 1,747,456 3,450,757 1,057,455 1,171,256 1,087,495 1,083,396 1,277,491 1,274,229 1,133,624 1,103,639 2,103,175 2,188,585 1,148,499 1,545,169 1,333,513 1,145,025 1,270,344 889,574 919,763 1,129,101 870,424 923,640 994,139 2,846,747 906,309 1,068,855 1,492,337 840,330 1,016,850 -376,012 -157,603 -154,601 252,622 178,673 370,550 -253,881 -320,289 -112,615 -62,681 925,981 591,816 1,554,734 476,857 1,539,413 -448,423 -1,419,092 -659,961 -2,367,361 220,036 102,973 46,129 20,243 825,684 914,356 14,875 441,530 -769,662 -1,043,560 121,845 -655,595 -413,750 -15,924 -399,920 34,066 74,376 1,717,646 35,885 145,215 498,198 -2,006,417 110,541 -181,375 -489,212 67,500 -37,720 141385024144 24838705609 23901469201 63817874884 31924040929 137307302500 64455562161 102585043521 12682138225 3928907761 857440812361 350246177856 2417197810756 227392598449 2369792384569 201083186929 2013822104464 435548521521 5604398104321 48415841296 10603438729 2127884641 409779049 681754067856 836046894736 221265625 194948740900 592379594244 1089017473600 14846204025 429804804025 171189062500 253573776 159936006400 1160492356 5531789376 2950307781316 1287733225 21087396225 248201247204 4025709177889 12219312681 32896890625 239328380944 4556250000 1422798400 0.387818754 0.163704258 0.124754286 0.204659777 0.155607462 0.277923617 0.257652256 0.350401833 0.108742253 0.049331851 0.484465905 0.393003949 0.600202753 0.272886413 0.446108781 0.424058707 1.211598489 0.60686348 2.185129906 0.172240744 0.080812005 0.040691623 0.018342048 0.3925893 0.417784093 0.012951687 0.285748679 0.577168727 0.911386214 0.095914965 0.73697635 0.449844145 0.014103256 0.459454243 0.036882335 0.074814488 0.603371497 0.039594664 0.135860337 0.333837464 2.38765366 0.108709249 0.204370803 0.487687975 0.074353128 0.038523996 84 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 859,112 794,946 600,394 814,820 1,073,763 949,066 1,173,758 1,212,861 1,649,596 2,204,547 2,528,855 2,136,125 1,750,787 1,452,891 975,078 1,302,898 1,364,959 1,544,982 1,863,259 1,806,980 1,806,074 2,229,541 2,002,932 2,147,831 2,164,777 2,057,634 2,224,213 1,979,464 2,145,665 2,345,687 1,609,402 2,087,459 1,927,879 2,019,837 1,770,749 1,914,976 2,225,317 1,866,076 2,365,402 2,217,748 1,671,398 20 21 21 21 21 22 22 22 22 23 23 23 23 24 24 24 24 25 25 25 25 26 26 26 26 27 27 27 27 28 28 28 28 29 29 29 29 30 30 30 30 887,480 1,003,125 907,830 979,130 859,112 794,946 600,394 814,820 1,073,763 949,066 1,173,758 1,212,861 1,649,596 2,204,547 2,528,855 2,136,125 1,750,787 1,452,891 975,078 1,302,898 1,364,959 1,544,982 1,863,259 1,806,980 1,806,074 2,229,541 2,002,932 2,147,831 2,164,777 2,057,634 2,224,213 1,979,464 2,145,665 2,345,687 1,609,402 2,087,459 1,927,879 2,019,837 1,770,749 1,914,976 2,225,317 -28,368 804743424 0.033020142 -208,179 43338496041 0.261878165 -307,436 94516894096 0.512057083 -164,310 26997776100 0.201651899 214,651 46075051801 0.199905379 154,120 23752974400 0.162391235 573,364 328746276496 0.488485701 398,041 158436637681 0.328183526 575,833 331583643889 0.349075167 1,255,481 1576232541361 0.569496137 1,355,097 1836287879409 0.535853973 923,264 852416413696 0.432214407 101,191 10239618481 0.057797436 -751,656 564986742336 0.517351956 -1,553,777 2414222965729 1.593489957 -833,227 694267233529 0.639518212 -385,828 148863245584 0.282666366 92,091 8480752281 0.05960652 888,181 788865488761 0.476681449 504,082 254098662724 0.278963796 441,115 194582443225 0.244239716 684,559 468621024481 0.307040328 139,673 19508546929 0.06973427 340,851 116179404201 0.158695447 358,703 128667842209 0.165699746 -171,907 29552016649 0.083545956 221,281 48965280961 0.099487324 -168,367 28347446689 0.085056864 -19,112 365268544 0.008907262 288,053 82974530809 0.122801124 -614,811 377992565721 0.382012077 107,995 11662920025 0.051735148 -217,786 47430741796 0.112966633 -325,850 106178222500 0.161324899 161,347 26032854409 0.091117939 -172,483 29750385289 0.090070581 297,438 88469363844 0.133660957 -153,761 23642445121 0.082398037 594,653 353612190409 0.251396169 302,772 91670883984 0.136522274 -553,919 306826258561 0.331410592 85 APPENDIX E Table 4.7 Calculation for Johor Tourism Data using 3-month Moving Average Period Actual Data t 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 33 34 35 At 1,213,874 1,107,735 1,129,539 981,022 1,205,673 1,240,579 1,334,745 1,514,357 1,213,478 1,474,035 1,514,357 1,393,287 1,088,378 1,080,564 1,109,080 1,049,116 1,013,572 1,205,808 1,109,254 2,146,602 953,087 952,753 930,598 1,193,784 1,080,693 944,095 1,305,077 1,217,528 1,190,167 1,345,568 1,120,333 1,393,845 981,729 969,556 962,730 3-month moving average Error 1150383 1072765 1105411 1142425 1260332 1363227 1354193 1400623 1400623 1460560 1332007 1187410 1092674 1079587 1057256 1089499 1109545 1487221 1402981 1350814 945479 1025712 1068358 1072857 1109955 1155567 1237591 1251088 1218689 1286582 1165302 1115043 -169,361 132,908 135,168 192,320 254,025 -149,749 119,842 113,734 -7,336 -372,182 -251,443 -78,330 -43,558 -66,015 148,552 19,755 1,037,057 -534,134 -450,228 -420,216 248,305 54,981 -124,263 232,220 107,573 34,600 107,977 -130,755 175,156 -304,853 -195,746 -152,313 Square Error Absolute Percentage Error 28683035414 17664447859 18270298112 36987110613 64528531275 22424763001 14362025069 12935346933 53821787 138519193003 63223749878 6135536680 1897299364 4357936215 22067696704 390273195 1075487912620 285299486045 202705251984 176581486656 61655207488 3022947015 15441376011 53925973587 11571950329 1197183067 11659104514 17096782855 30679507565 92935351609 38316627013 23199351511 0.172636971 0.110235252 0.108955308 0.144087697 0.167744242 0.123404792 0.081301778 0.075103603 0.005265486 0.341959932 0.232696382 0.070625804 0.041518764 0.065130713 0.12319706 0.017809567 0.483115796 0.560425578 0.472554796 0.451554807 0.207997985 0.050875997 0.131621641 0.177935606 0.088353615 0.029071831 0.080246657 0.116710538 0.125663662 0.310526632 0.201892756 0.158209813 et 86 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 74 75 76 77 78 79 80 81 1,239,244 1,234,351 1,148,229 1,333,280 985,363 914,062 1,035,614 1,270,599 1,911,344 1,505,878 2,590,348 1,747,456 3,450,757 1,057,455 1,171,256 1,087,495 1,083,396 1,277,491 1,274,229 1,133,624 1,103,639 2,103,175 2,188,585 1,148,499 1,545,169 1,333,513 1,145,025 1,270,344 889,574 919,763 1,129,101 870,424 923,640 994,139 2,846,747 906,309 1,068,855 1,492,337 840,330 1,016,850 887,480 1,003,125 907,830 979,130 859,112 794,946 971338 1057177 1145442 1207275 1238620 1155624 1077568 978346 1073425 1405852 1562607 2002523 1947894 2596187 2085223 1893156 1105402 1114049 1149461 1211705 1228448 1170497 1446813 1798466 1813420 1627418 1342394 1341236 1249627 1101648 1026560 979479 973096 974388 929401 1588175 1582398 1607304 1155834 1133841 1116506 914887 969152 932812 963362 915357 267,906 177,174 2,787 126,005 -253,257 -241,562 -41,954 292,253 837,919 100,026 1,027,741 -255,067 1,502,863 -1,538,732 -913,967 -805,661 -22,006 163,442 124,768 -78,081 -124,809 932,678 741,772 -649,967 -268,251 -293,905 -197,369 -70,892 -360,053 -181,885 102,541 -109,055 -49,456 19,751 1,917,346 -681,866 -513,543 -114,967 -315,504 -116,991 -229,026 88,238 -61,322 46,318 -104,250 -120,411 71773446232 31390744392 7769227 15877344028 64139108049 58352199844 1760166085 85411621174 702108250561 10005133992 1056251563081 65059344534 2258597196769 2367696167824 835335067778 649089646921 484264036 26713287364 15567137003 6096694615 15577286481 869887629899 550226194499 422457534400 71958420167 86379953088 38954390582 5025628403 129638402844 33082031968 10514588320 11893065728 2445895936 390088834 3676215683716 464941696533 263726755211 13217334444 99542563680 13686816087 52452755992 7786003469 3760346803 2145388003 10867993000 14498889195 0.21618476 0.143536428 0.002427506 0.09450778 0.257018987 0.264273102 0.040511555 0.230011724 0.438392566 0.066423486 0.396757887 0.145964953 0.435516903 1.455127641 0.780330403 0.740841107 0.020312056 0.127939845 0.097916727 0.068877629 0.11308861 0.44346175 0.338927816 0.565927644 0.173606037 0.220398801 0.172370618 0.055805094 0.40474804 0.197751667 0.090816204 0.125289897 0.053544671 0.019867108 0.673521743 0.752355249 0.480461179 0.077038006 0.375452104 0.11505204 0.258062905 0.087963448 0.067547522 0.047305601 0.121345839 0.151471085 87 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 600,394 814,820 1,073,763 949,066 1,173,758 1,212,861 1,649,596 2,204,547 2,528,855 2,136,125 1,750,787 1,452,891 975,078 1,302,898 1,364,959 1,544,982 1,863,259 1,806,980 1,806,074 2,229,541 2,002,932 2,147,831 2,164,777 2,057,634 2,224,213 1,979,464 2,145,665 2,345,687 1,609,402 2,087,459 1,927,879 2,019,837 1,770,749 1,914,976 2,225,317 1,866,076 2,365,402 2,217,748 1,671,398 877729 751484 736720 829659 945883 1065529 1111895 1345405 1689001 2127666 2289842 2138589 1779934 1392919 1243622 1214312 1404280 1591067 1738407 1825438 1947532 2012849 2126768 2105180 2123414 2148875 2087104 2116447 2156939 2033585 2014183 1874913 2011725 1906155 1901854 1970347 2002123 2152265 2149742 -277,335 63,336 337,043 119,407 227,875 147,332 537,701 859,142 839,854 8,459 -539,055 -685,698 -804,856 -90,021 121,337 330,670 458,979 215,913 67,667 404,103 55,400 134,982 38,009 -47,546 100,799 -169,411 58,561 229,240 -547,537 53,874 -86,304 144,924 -240,976 8,821 323,463 -104,271 363,279 65,483 -478,344 76914887115 4011448896 113597983849 14258031649 51927015625 21706718224 289122365401 738124976164 705354181413 71554681 290580652395 470181747204 647793717307 8103720427 14722586678 109342869347 210662028427 46618567511 4578822889 163299504011 3069196933 18220140324 1444684081 2260622116 10160438401 28699973980 3429429762 52550824773 299796401344 2902443792 7448322880 21002869160 58069432576 77810041 104628312369 10872510955 131971631841 4288023289 228812982336 0.461922227 0.077730051 0.313889564 0.125815275 0.194141382 0.121474761 0.325959205 0.389713624 0.332108273 0.003959974 0.307893155 0.471954193 0.825427641 0.069092643 0.088894001 0.214028599 0.246331473 0.119488502 0.03746635 0.181249564 0.027659618 0.062845727 0.017557929 0.023107122 0.045318951 0.085584111 0.027292859 0.097728157 0.34021125 0.025808571 0.044766122 0.071750179 0.136087046 0.004606324 0.14535592 0.055877324 0.153580237 0.029526799 0.286193953 88 APPENDIX F Table 4.8 Calculation for Johor Tourism Data using Exponential Smoothing ( 0.37 ) Period Actual Data Forecast Error Square t 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 33 At 1,213,874 1,107,735 1,129,539 981,022 1,205,673 1,240,579 1,334,745 1,514,357 1,213,478 1,474,035 1,514,357 1,393,287 1,088,378 1,080,564 1,109,080 1,049,116 1,013,572 1,205,808 1,109,254 2,146,602 953,087 952,753 930,598 1,193,784 1,080,693 944,095 1,305,077 1,217,528 1,190,167 1,345,568 1,120,333 1,393,845 981,729 Ft 1213874 1213874 1174603 1157929 1092473 1134357 1173659 1233261 1337267 1291465 1359016 1416492 1407906 1289681 1212308 1174113 1127864 1085576 1130062 1122363 1501331 1298481 1170562 1081775 1123218 1107484 1047030 1142507 1170265 1177629 1239766 1195576 1268936 et 0 -106,139 -45,064 -176,907 113,200 106,222 161,086 281,096 -123,789 182,570 155,341 -23,205 -319,528 -209,117 -103,228 -124,997 -114,292 120,232 -20,808 1,024,239 -548,244 -345,728 -239,964 112,009 -42,525 -163,389 258,047 75,021 19,902 167,939 -119,433 198,269 -287,207 Error 0 11265487321 2030725341 31296104021 12814140173 11283054270 25948597910 79014950841 15323600562 33331887084 24130901403 538472845 102098245755 43729811581 10655926491 15624338684 13062737489 14455692925 432970665 1049065515384 300571959658 119527845532 57582546575 12545995775 1808408549 26695963320 66588220073 5628085673 396087934 28203586172 14264309052 39310608973 82487579365 Absolute Percentage Error 0 0.095816238 0.039895541 0.180329339 0.093889105 0.085622699 0.120686487 0.185620684 0.102011352 0.123857456 0.10257901 0.016654873 0.293581973 0.193525549 0.093074933 0.119145409 0.112761929 0.099710592 0.018758505 0.477144339 0.575230209 0.362872637 0.257859608 0.093826781 0.039350109 0.173064145 0.197725448 0.06161712 0.016721988 0.124809176 0.106605164 0.142246112 0.292551722 89 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 74 75 76 77 78 79 969,556 962,730 1,239,244 1,234,351 1,148,229 1,333,280 985,363 914,062 1,035,614 1,270,599 1,911,344 1,505,878 2,590,348 1,747,456 3,450,757 1,057,455 1,171,256 1,087,495 1,083,396 1,277,491 1,274,229 1,133,624 1,103,639 2,103,175 2,188,585 1,148,499 1,545,169 1,333,513 1,145,025 1,270,344 889,574 919,763 1,129,101 870,424 923,640 994,139 2,846,747 906,309 1,068,855 1,492,337 840,330 1,016,850 887,480 1,003,125 907,830 979,130 1162669 1091217 1043677 1116037 1159813 1155527 1221296 1134001 1052623 1046330 1129309 1418662 1450932 1872516 1826244 2427314 1920466 1643258 1437626 1306561 1295805 1287822 1230769 1183731 1523925 1769849 1539950 1541881 1464785 1346474 1318306 1159675 1070908 1092439 1010294 978232 984117 1673290 1389507 1270866 1352810 1163193 1109046 1027066 1018208 977368 -193,113 -128,487 195,567 118,314 -11,584 177,753 -235,933 -219,939 -17,009 224,269 782,035 87,216 1,139,416 -125,060 1,624,513 -1,369,859 -749,210 -555,763 -354,230 -29,070 -21,576 -154,198 -127,130 919,444 664,660 -621,350 5,219 -208,368 -319,760 -76,130 -428,732 -239,912 58,193 -222,015 -86,654 15,907 1,862,630 -766,981 -320,652 221,471 -512,480 -146,343 -221,566 -23,941 -110,378 1,762 37292669885 16508974390 38246463263 13998256776 134189884 31596149442 55664179375 48372953234 289315216 50296656496 611578070736 7606592269 1298268687416 15639992831 2639043225101 1876512767313 561315564395 308872817725 125478795972 845054075 465523029 23776982835 16161949824 845377849773 441772807826 386076134514 27241529 43417146183 102246280601 5795719899 183810839845 57557742992 3386480241 49290710101 7508832661 253041618 3469388980634 588260406291 102817869713 49049438520 262635976584 21416138728 49091403546 573193207 12183328460 3103893 0.199176841 0.133461359 0.157811561 0.095851366 0.010088611 0.133320126 0.239437216 0.240616634 0.016424332 0.17650664 0.409154276 0.057916896 0.439869833 0.07156687 0.470770103 1.295429751 0.639663712 0.511049039 0.326962499 0.022755396 0.016932579 0.136022058 0.115191342 0.437169668 0.303693902 0.541010703 0.003377845 0.156254806 0.279260036 0.059928355 0.481951658 0.260841052 0.051539652 0.255065477 0.093817419 0.016001064 0.654301063 0.846269164 0.29999603 0.14840554 0.609855915 0.143917529 0.249657231 0.02386687 0.121584565 0.001799339 90 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 859,112 794,946 600,394 814,820 1,073,763 949,066 1,173,758 1,212,861 1,649,596 2,204,547 2,528,855 2,136,125 1,750,787 1,452,891 975,078 1,302,898 1,364,959 1,544,982 1,863,259 1,806,980 1,806,074 2,229,541 2,002,932 2,147,831 2,164,777 2,057,634 2,224,213 1,979,464 2,145,665 2,345,687 1,609,402 2,087,459 1,927,879 2,019,837 1,770,749 1,914,976 2,225,317 1,866,076 2,365,402 2,217,748 1,671,398 978020 934024 882565 778162 791725 896079 915684 1011172 1085797 1294402 1631156 1963305 2027248 1924958 1750293 1463463 1404054 1389589 1447084 1601069 1677256 1724919 1911629 1945411 2020306 2073761 2067794 2125669 2071573 2098987 2190266 1975346 2016828 1983917 1997207 1913418 1913994 2029184 1968834 2115564 2153372 -118,908 -139,078 -282,171 36,658 282,038 52,987 258,074 201,689 563,799 910,145 897,699 172,820 -276,461 -472,067 -775,215 -160,565 -39,095 155,393 416,175 205,911 128,818 504,622 91,303 202,420 144,471 -16,127 156,419 -146,205 74,092 246,700 -580,864 112,113 -88,949 35,920 -226,458 1,558 311,323 -163,108 396,568 102,184 -481,974 14139130106 19342714210 79620583086 1343819769 79545226341 2807591298 66601996683 40678607893 317869664605 828363132438 805863632974 29866897069 76430759890 222846795411 600958148937 25781244514 1528434334 24146992731 173201303336 42399340859 16594059459 254643662431 8336246389 40973823851 20871739629 260065741 24466988352 21375860867 5489615377 60860851201 337403044048 12569246019 7911930163 1290254324 51283375457 2428149 96921822677 26604121756 157266296867 10441556352 232299053920 0.138408117 0.174952873 0.469976706 0.044989258 0.262662834 0.055830373 0.219869536 0.16629225 0.341780237 0.412848792 0.354982424 0.080903701 0.157906779 0.324915301 0.795028608 0.123237115 0.028642029 0.100579183 0.223358431 0.113953116 0.071324836 0.226334612 0.045584696 0.094243877 0.06673692 0.007837426 0.070325671 0.073860833 0.034530991 0.105171714 0.36091918 0.053707713 0.046138285 0.017783668 0.127888442 0.000813719 0.139900382 0.087406783 0.167653595 0.046075539 0.28836586 91 APPENDIX G Table 4.9 Calculation for Johor Tourism Data using Holt’s Model ( 0.36, 0 and a0 1236616, b 6712.383 ) Period t 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 33 34 35 36 37 38 Actual Data At 1,213,874 1,107,735 1,129,539 981,022 1,205,673 1,240,579 1,334,745 1,514,357 1,213,478 1,474,035 1,514,357 1,393,287 1,088,378 1,080,564 1,109,080 1,049,116 1,013,572 1,205,808 1,109,254 2,146,602 953,087 952,753 930,598 1,193,784 1,080,693 944,095 1,305,077 1,217,528 1,190,167 1,345,568 1,120,333 1,393,845 981,729 969,556 962,730 1,239,244 1,234,351 1,148,229 at bt Ft 1232724.8 1192024.4 1173825.6 1108712.2 1147914 1185569.3 1243568.5 1345348.3 1302170.9 1368337.9 1425200.7 1418007.7 1303636.9 1227626.6 1189245.7 1143095 1100762.6 1142874.9 1135067.3 1503515.7 1309657.3 1185467.7 1098010.5 1136784.9 1120887.7 1061538.3 1153508.1 1180851.2 1188500.8 1249340.9 1207194 1278684.3 1176076.3 1106024.9 1058734.7 1128014 1170591.2 1166836.7 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 1239437 1198737 1180538 1115425 1154626 1192282 1250281 1352061 1308883 1375050 1431913 1424720 1310349 1234339 1195958 1149807 1107475 1149587 1141780 1510228 1316370 1192180 1104723 1143497 1127600 1068251 1160221 1187564 1195213 1256053 1213906 1285397 1182789 1112737 1065447 1134726 1177304 Error -131,702 -69,198 -199,516 90,248 85,953 142,463 264,076 -138,583 165,152 139,307 -38,626 -336,342 -229,785 -125,259 -146,842 -136,235 98,333 -40,333 1,004,822 -557,141 -363,617 -261,582 89,061 -62,804 -183,505 236,826 57,307 2,603 150,355 -135,720 179,939 -303,668 -213,233 -150,007 173,797 99,625 -29,075 Square Error 17345466356 4788333227 39806620040 8144773667 7387848221 20295785134 69736193943 19205156450 27275086702 19406359510 1491974710 113125994522 52801290510 15689813056 21562611941 18560070028 9669377772 1626772397 1009667907538 310406198773 132217092335 68425174259 7931879252 3944377588 33674129876 56086714744 3284146556 6777693 22606563826 18420004006 32377905225 92214053378 45468180661 22502191693 30205376876 9925072793 845332416 Absolute Percentage Error 0.118893226 0.061261969 0.203375627 0.074853132 0.069284256 0.106734452 0.174381677 0.11420287 0.112040561 0.091990667 0.027722995 0.309030575 0.212653127 0.112939539 0.13996749 0.13441112 0.081549463 0.036360713 0.468099036 0.584564782 0.381648426 0.281090288 0.074604031 0.058114821 0.194371459 0.181465415 0.047068711 0.002187424 0.111740762 0.121142835 0.12909514 0.309319241 0.219928185 0.155814512 0.140244328 0.080710154 0.025321256 92 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 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 1,333,280 985,363 914,062 1,035,614 1,270,599 1,911,344 1,505,878 2,590,348 1,747,456 3,450,757 1,057,455 1,171,256 1,087,495 1,083,396 1,277,491 1,274,229 1,133,624 1,103,639 2,103,175 2,188,585 1,148,499 1,545,169 1,333,513 1,145,025 1,270,344 889,574 919,763 1,129,101 870,424 923,640 994,139 2,846,747 906,309 1,068,855 1,492,337 840,330 1,016,850 887,480 1,003,125 907,830 979,130 859,112 794,946 600,394 814,820 1,073,763 949,066 1,173,758 1,212,861 1,649,596 1231052.2 1146900 1067374.3 1060236.5 1140262.9 1422148 1456586.7 1869036.7 1829563.6 2417489.1 1932172.8 1662538.7 1459818.9 1328602.6 1314498.3 1304297.3 1247150.8 1199782.5 1529299.7 1770938.3 1551156.1 1553296.7 1478470.5 1362726 1333764.4 1178151.8 1089427.8 1108006 1026772.4 993940.68 998308 1668042 1398114 1283876.7 1363218.3 1179274.5 1125097.6 1043851.2 1033485.7 992545.56 992011.88 948463.85 897493.35 794833.51 806324.57 906898.33 926374.62 1019728.6 1093552.2 1298023.9 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 1173549 1237765 1153612 1074087 1066949 1146975 1428860 1463299 1875749 1836276 2424202 1938885 1669251 1466531 1335315 1321211 1311010 1253863 1206495 1536012 1777651 1557868 1560009 1485183 1369438 1340477 1184864 1096140 1114718 1033485 1000653 1005020 1674754 1404826 1290589 1369931 1185987 1131810 1050564 1040198 999258 998724 955176 904206 801546 813037 913611 933087 1026441 1100265 159,731 -252,402 -239,550 -38,473 203,650 764,369 77,018 1,127,049 -128,293 1,614,481 -1,366,747 -767,629 -581,756 -383,135 -57,824 -46,982 -177,386 -150,224 896,680 652,573 -629,152 -12,699 -226,496 -340,158 -99,094 -450,903 -265,101 32,961 -244,294 -109,845 -6,514 1,841,727 -768,445 -335,971 201,748 -529,601 -169,137 -244,330 -47,439 -132,368 -20,128 -139,612 -160,230 -303,812 13,274 260,726 35,455 240,671 186,420 549,331 25513951609 63706580121 57384405049 1480145073 41473370852 584259498803 5931707707 1270239152039 16459121341 2606548991151 1867996063071 589254525557 338440095523 146792621015 3343608502 2207280809 31465677905 22567314886 804035236093 425851381019 395831901014 161277073 51300463957 115707369482 9819702723 203313341751 70278635509 1086418421 59679769188 12065884275 42433071 3391956920686 590508256006 112876786341 40702221813 280476918853 28607271160 59697130122 2250416824 17521303409 405134035 19491584720 25673727573 92301568580 176201955 67978071271 1257077355 57922530474 34752437768 301765046001 0.119802946 0.256150905 0.262072401 0.037149607 0.160278828 0.399911629 0.051144635 0.435095543 0.073417074 0.467862857 1.292486701 0.655389735 0.534950547 0.353642852 0.04526368 0.036870694 0.156476641 0.136117168 0.426345938 0.298171144 0.547803465 0.008218836 0.169849156 0.297074614 0.078005968 0.506874984 0.288227706 0.029192129 0.280661415 0.118926009 0.006552471 0.646958305 0.847884496 0.31432833 0.135189248 0.630229453 0.166334112 0.275307569 0.047290775 0.145807101 0.020556966 0.162507642 0.201561154 0.5060206 0.016290848 0.242815264 0.037358083 0.205043118 0.153702739 0.333009691 93 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 2,204,547 2,528,855 2,136,125 1,750,787 1,452,891 975,078 1,302,898 1,364,959 1,544,982 1,863,259 1,806,980 1,806,074 2,229,541 2,002,932 2,147,831 2,164,777 2,057,634 2,224,213 1,979,464 2,145,665 2,345,687 1,609,402 2,087,459 1,927,879 2,019,837 1,770,749 1,914,976 2,225,317 1,866,076 2,365,402 2,217,748 1,671,398 1628668.1 1957031.3 2025801 1931091.9 1763235.5 1483794.7 1422967.8 1406380.6 1460573 1609835.9 1685103.7 1732948.9 1916018 1951603 2026541 2080601.9 2076629.4 2134055.4 2082698.4 2109662.3 2198927.1 1990994 2030017.3 1997543.5 2009865.1 1928079.2 1927658 2039111.1 1981114.4 2123753.9 2161887.7 1989607.3 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 6712.383 1304736 1635381 1963744 2032513 1937804 1769948 1490507 1429680 1413093 1467285 1616548 1691816 1739661 1922730 1958315 2033253 2087314 2083342 2140768 2089411 2116375 2205640 1997706 2036730 2004256 2016577 1934792 1934370 2045824 1987827 2130466 2168600 1996320 2003032 2009744 2016457 2023169 2029882 2036594 2043306 2050019 2056731 2063444 2070156 899,811 893,474 172,381 -281,726 -484,913 -794,870 -187,609 -64,721 131,889 395,974 190,432 114,258 489,880 80,202 189,516 131,524 -29,680 140,871 -161,304 56,254 229,312 -596,238 89,753 -108,851 15,581 -245,828 -19,816 290,947 -179,748 377,575 87,282 -497,202 m=1 m=2 m=3 m=4 m=5 m=6 m=7 m=8 m=9 m=10 m=11 m=12 809659381054 798296673764 29715310520 79369739013 235140860122 631818099206 35197172662 4188834251 17394720566 156795097210 36264241998 13054872554 239982107304 6432299273 35916182162 17298466289 880917671 19844709738 26018909925 3164534662 52584132297 355499167250 8055531214 11848477469 242772591 60431621761 392657385 84649949063 32309175077 142563029273 7618102658 247209895918 0.40816129 0.353311872 0.08069813 0.160914123 0.333757488 0.815185927 0.143993694 0.047416226 0.085366073 0.212516675 0.105386737 0.063263145 0.219722215 0.040042106 0.08823583 0.060756204 0.014424459 0.063335325 0.081488616 0.026217605 0.097759123 0.37047146 0.042996107 0.056461382 0.007714069 0.138827377 0.010347693 0.130743909 0.0963238 0.159624113 0.039356024 0.297476763