國立臺灣科技大學 管理研究所 EMBA 碩士在職專班 碩士學位論文 全球銅箔基層板市場預測分析

Transcription

1 國立臺灣科技大學 管理研究所 EMBA 碩士在職專班 碩士學位論文 學號 :M 全球銅箔基層板市場預測分析 Forecasting Analysis for Global Copper Clad Laminate Market 研究生 : 蕭裕耀 指導教授 : 王福琨博士 中華民國九十九年六月

2

3

4 摘要需求預測為高階主管在未來策略投資之重要參考之一 銅箔基層板為印刷電路板之基礎材料, 廣泛運用在民生家電 電腦 通訊 手機 汽車 醫療 軍事用途 而全球印刷電路板在 2008 年之產值為美金四佰捌拾貳億 在本研究中, 我們運用灰色模式, 滾動灰色模式, 貝氏擴散模式, 分析 2001 年至 2008 年全球銅箔基層板六大產品市場, 分別為紙基板 複合板 FR 4 板 高玻璃轉化温度之 FR-4 板, 無鹵素之 FR-4 板與特殊板. 而運用平均絕對值百分比誤差 (Mean Absolute Percentage Error, 簡稱 MAPE) 進行預測效益評估 本研究結果分析, 貝氏擴散模式之平均絕對值百分比誤差低於灰色模式與滾動灰色模式, 可運用在全球銅箔基層板之預測分析 關鍵字 : 銅箔基層板, 印刷電路板, 灰色模式, 滾動灰色模式, 貝氏擴散模式 I

5 Abstract Demand forecasting is one of critical reference by top managers to make the strategy decision for future investment. The copper clad laminate (CCL) is the key material for print circuit board (PCB) and it can apply for consumer, computer, LCD, communication, automotive, aero space, medicine and defense application. The total global sale for PCB in 2008 is US$ 48.2 billion. In this research, we use grey model GM(1,1), rolling grey model (RGM) and Bass diffusion model to analysis global CCL market by six market segments paper, composite, FR-4, FR-4 High Tg, FR-4 halogen free, Specialty between The forecasting accuracy of global CCL market by six market segment was evaluated along with mean absolute percentage error (MAPE). In this study, Bass diffusion model MAPE outperforms the others two models GM(1,1) and RGM for this global CCL market forecasting analysis and is recommend for global CCL market forecasting analysis. Keywords: Copper clad laminate, Print circuit board, GM(1,1), Rolling GM, Bass diffusion model II

6 誌謝在東海研究所畢業二十四年後 重回校園, 當起學生, 與眾多社會精英, 經歴每週上課, 小組討論, 哈佛個案分析, 上台報告, 交學期作業, 在課內課外相互切磋, 擴展領域, 追求成長 不齗學習與自我超越 在這二年 EMBA 碩士生涯中, 最重要的是完成人生中的第二篇碩士論文 最需感謝的就是恩師王福琨教授, 在碩一的研究方法課程中, 將我引領進入預測領域, 一窺此領域的奧妙, 與其所帶來的研究樂趣! 而在口試期間, 清大林則孟教授及本院歐陽超副院長給予諸多的建議, 使本論文更臻完備 同時感謝 PRISMARK 姜旭高博士, 提供全球铜箔基層板數據, 作為本研究之預測分析母體 本公司的林宗輝總經理與太平洋地區商業總監 -Carols Lansangan 的支持, 讓我能在這二年當中, 能工作與學業兼顧, 更感謝美國同事 - Julian Yvonne 在內容的校稿與修正 也感謝 MA529 研究室的世賢與道明, 在軟體分析的協助 最感謝的是子娟這二年的陪伴與支持, 老大文博與老二文祺的陪讀, 享受一家人在一起研讀, 準備報告與完成論文 更感謝爸媽這二年的支持與鼓勵, 讓我們無後顧之憂 最後謹以此文, 獻給所有關心與支持我的每一個人! 蕭裕耀 謹誌於台灣科技大學 民國 99 年 4 日 2 日 III

7 Contents Chapter 1 Introduction Motivation Research Objectives Research Procedure 4 Chapter 2 Literature Review Global CCL Industry Forecasting Models.14 Chapter 3 Methodology Forecasting Models Performance Measurement...22 Chapter 4 Forecasting Analysis for Global CCL Comparison by Different Market Segment Hierarchical Forecasting For Global CCL Market MAPE Comparison 33 Chapter 5 Conclusions and Future Reasrch Conclusions Future Research.37 References 38 IV

8 LIST OF FIGURES Figure 1.1. The total global sales for PCB from Figure 1.2. The geographic migration of PCB manufacturing...2 Figure 1.3. The flowchart of research 5 Figure 2.1. Global electronic industry value chain analysis...6 Figure 2.2. Electrical laminate processing Figure 2.3. Standard multilayer s procedure Figure 2.4. Sequential lamination multilayer Figure 4.1. The plots of the original and predicted values (Paper) Figure 4.2. The plots of the original and predicted values (Composite) Figure 4.3. The plots of the original and predicted values (FR-4). 28 Figure 4.4. The plots of the original and predicted values (FR-4 high Tg)..29 Figure 4.5. The plots of the original and predicted values (FR-4 Halogen Free)...30 Figure 4.6. The plots of the original and predicted values (Specialty).31 Figure 4.7. The plots of the original and predicted values (Global total CCL)...32 Figure 4.8. MAPE by different forecasting models for historic periods...34 Figure 4.9. MAPE by different forecasting models for holdout periods...34 V

9 LIST OF TABLE Table 1.1. Global CCL market demand by products segment.. 3 Table 2.1. IPC definition of CCL products..8 Table 3.1. Criteria for forecasting accuracy..23 Table 4.1. The parameter estimation of GM(1,1) by different markets...25 Table 4.2. The parameter estimation of Bass by different markets...25 Table 4.3. Paper CCL market analysis by different forecasting models...26 Table 4.4. Composite CCL market analysis by different forecasting models...27 Table 4.5. FR-4 CCL market analysis by different forecasting models...28 Table 4.6. FR-4 high Tg CCL market analysis by different forecasting models...29 Table 4.7. FR-4 Halogen Free CCL market analysis by different forecasting model...30 Table 4.8. Specialty CCL market analysis by different forecasting models...31 Table 4.9. Global CCL market analysis by different forecasting models..32 Table MAPE ranking by three forecast model.33 Table MAPE comparison by HF & Bass Diffusion Model..33 Table Comparison of all models for historic period (MAPE)...33 Table Comparison of all models for holdout period (MAPE) VI

10 Chapter 1 Introduction 1.1. Motivation Copper clad laminates (CCLs) are composite materials manufactured by pressing layers of filler material impregnated with resin together with layers of copper foil under heat and pressure. Laminates provide mechanical support for electronic components interconnect them electronically. The most common fillers materials used are paper and glass fiber. Paper has been used as reinforcement in a vast majority of printed circuit boards (PCBs). Paper Laminates are low priced and easily manufactured. Composite Epoxy Materials (CEM) is the group of composite materials typically made of woven glass fabric surfaces and non-woven glass core combined with epoxy resin. They are two major products CEM-1 which is low cost, flame retardant, cellulose paper based laminate with only one layer of woven glass fabric & CEM-3 which is very similar to the most commonly used PCB material- FR4 but modified with white color and flame retardant. The most widely used and industry standard material for PCB process is FR-4 with Tg 135 degree C. The Tg of high performance FR-4 is 170 degree C and BT is 195 degree C, where Tg is glass transition temperature, a measure of temperature resistance. The CCL is the key materials for print circuit board.the total global sale for PCB in 2008 were US$48.2 billion and the market share by geography are as follows. China/ 31.2%, Asia/21.1%, Japan / 14%, USA/9.3%, Europe/6.7% (see figure 1.1.). In addition, Figure 1.2 depicts the geographic migration of PCB manufacturing. PCB is widely use for Consumers, Communication, Computer, LCD, Automotive, Aero Space, Medical and others electronic products. To align with different PCB applications, manufacturers need to use different raw materials paper, composite, FR-4 high Tg, FR-4, high Tg, FR-4 Halogen Free and Special. All raw materials differences are based on different resin structure and end use application. 1

11 Figure 1.1. The total global sales for PCB from Figure 1.2. The geographic migration of PCB manufacturing CCL forecasting is difficulty due to complex demand from different market. Prismark is a well known global consultant company that collects demand by six market segments and 2

12 issues an annual report. They collect these data one by one (bottom up) then consolidate into completed data (see Table 1.1 for global CCL market demand by products segment from ). Table 1.1. Global CCL market demand by products segment ( M m*2) Paper Composite FR FR-4 High Tg FR-4 Halogen Free Specialty(High Speed, Low Loss),BT and BT-Equivalent, FR-5 Total Source: Prismark But Prismark can only provide limited forecasting data based on their existing estimation by different markets. This is why they intend to develop a new forecasting model for this rapidly changing market and compare the results with their bottom up statistic data. 1.2 Research Objectives The purposes of this study are: 1. To develop an efficient forecasting model that provides insight into global CCL market demand and apply MAPE to find out the best forecasting model. 2. To assist CCL/PCB manufactures in making the strategy decision for future expansion and investment. 3

13 1.3 Research Procedure To determine global trend of CCL market, this thesis attempts to compare different models and then select the most accurate for this market. Generally, a reliable and accurate forecasting method can improve the ability of CCL related industry (Epoxy resin, Copper foil, Glass cloth, CCL and PCB) to make correct investment decision. Data on the levels of demand in the global CCL industry were applied for this forecasting model analysis! The procedure for this research is shown in Figure 1 and is given as follows: Chapter 1 Introduce the motivation, objective and procedure of this research. Chapter 2 Present a literature review including the global CCL market introduction, CCL & PCB process introduction, forecasting model introduction and literature review of GM(1,1), RGM and Bass model. Chapter 3 Present Methodology of GM(1,1), RGM, Bass and Genetic algorithm(ga). Chapter 4 Apply GM(1,1), RGM and Bass diffusion model to analysis global CCL market from 2001 to 2007 and use 2008 data as holdout period by MAPE comparison. Chapter 5 Conclude the research results with summary and provide several future extensions on this research. 4

14 Research Topics Global CCL Industry Background Literature Review Problems definition &Research Target Research Methodology & tools/gm(1,1) RGM BASS Model Global CCL demand Paper Composite FR 4 FR 4 High Tg FR 4 Halogen Speciality (High speed, Low Loss ) Data analysis & verification Conclusions & future study Figure 1.3. The flowchart of research 5

15 Chapter 2 Literature Review This chapter is divided into two sections as follows: 2.1 Global CCL Industry CCL is key raw material for PCB and is a composite of copper foil, glass cloth and epoxy resin. The total values for base raw materials used in CCL manufacture was US$ 25 billion in Epoxy is one of key materials and total value is US$ 0.5 billion (2% share). PCB manufactures use the CCL & FCCL as the base to apply circuit patterns using photo mask/dry film then etching. The value of CCL used is about US$ 5 billion, which transfers to PCB of US $ 45 billion. These PCBs can be incorporated into electronics for consumer, computer, communication, automotive, medical and defense applications with a total value of US $ 950 Billion (see Figure2.1) Value Chain Analysis Hardeners & other additives - dicy, imidazoles, phenolics, fillers, solvents Reinforcement - Glass Conductor - Copper Electrical Components Integrated Circuit Chips Other Components Basic RM Resin Suppliers Laminators Board Shops OEMs Consumer Epi Bis A Dow TBBA DOP Solvents Function: Manufactures resin BER PER LER LEN Dow Huntsman Hexion et al Function: Manufactures laminate and prepreg Isola Nelco TUC Polyclad et al Fu nction : Manufactures PCB Photocircuits Sanmina Tyco Multek Merix et al Resins: $1 Bn Epoxy: $0.5 Bn $5 Bn $45 Bn EMS Function: Make boards and assemble final product Flextronics FoxConn Benchmark et al Function: Manufactures End Product Motorola Sony Nokia et al $950 Bn Electronic Manufacturing Services (New segment) OEM Contract Manufacturer Overlaps with board shops and OEMs (via distributors & stores) Figure 2.1. Global electronic industry value chain analysis 6

16 All CCL products definition is based on IPC classifications. IPC is a global trade association dedicated to the competitive excellence and financial success of its more than 2,700 member companies, IPC represents all facets of the industry, including design, printed circuit board manufacturing and electronics assembly (see Table 2.1). Prismark s data analysis by market segment is based on IPC definitions which are as follows: 1) Paper- Single and double-sides paper phenolic or paper epoxy XXXP, FR-1, FR-2, FR-3. 2) Composite: All single-and double sides CEM1, CEM3. 3) FR-4 : All rigid PCBs with more than 2 layers made with FR-4 epoxy and Tg is degree C. 4) FR-4 high Tg : All rigid PCBs with more than 2 layers made with FR-4 epoxy and Tg is above170 degree C. 5) FR-4 Halogen Free: All rigid PCBs with more than 2 layers made with non- Bromine ( Fluorine, Chlorine, Iodine, Astatine) epoxy and Tg is above 140 degree C. 6) Specialty(high speed & low loss),bt, BT equivalent and FR-5: All non-epoxy (non FR-4) rigid boards, including PTFE, rigid polyimide, APPE, Cyan ate Ester, Bismaleimide Triazine (BT) and GETEK; includes 2-side high performance and boards made with exotic materials such as Teflon and blended resin with unusual characteristics. 7

17 Table 2.1. IPC definition of CCL products Class IPC-4101 FR Appr Tg [C] MOT App Dk App Df Rel. Cost Segmen t Comment FR-1: 02 + Paper phenolic resin/cellulose paper, general purpose FR-2: Paper phenolic resin/cellulose paper, higher quality, tested for electrical properties FR-3: Paper epoxy resin/ cellulose paper FR Rigid & ML FR-5: Rigid & ML epoxy resin/ glass cloth, higher peel strength epoxy resin/glass cloth, higher heat resistance FR-6: + polyester resin/glass cloth CEM-1: Paper/Ri epoxy resin/glass cloth/ gid cellulose paper core CEM-3: Rigid epoxy resin/ glass cloth/glass fiber core CRM-5: + 80 Woven E-Glass face sheets, non woven E-Glass core/polyster/kaolin G Composi Woven E-glass fabric / te Epoxy G Composi Woven E-glass fabric / te Epoxy Source: IPC-2008 version 8

18 The type of CCL products, their applications and the raw materials used to make them have been described. The next section explains in detail the manufacturing process and where each raw material is used in the ten process steps. The details of electrical laminate processing as shown in Figure 2.2 and are given as follows: 1) Copper foil cutting: Put the roll of copper foil on the feed frame, and calibrate the beginning allocation. It can set the cutting length by operation panel and executive automatic. 2) Paper cutting: Paper is used as cushion during lamination.they put the roll of paper on the feed frame and calibrates the beginning allocation. Set the cutting length by operation panel and executive automatic. The most of the manufacturers select this way for production efficiency and cost down. 3) Prepreg manufacture: Prepreg is the main method to adhere fiber-glass and copper foil. The manufacturing methods use the conveyor system to deliver the fiber-glass, pass by tank of gel then go into the oven for drying and cutting. The quality of prepreg is related to the life of printed circuit board. 4) Stack: Put the layers between cover and carriage. The material layers are sandwiched between mirror face steel plate (to give a very smooth surface) and paper. 5) Lamination: The main purpose of this step is to melt the prepreg, then to increase the temperature and pressure and to adhere the copper foil and glass fiber closely. Presently, in order to raise the well production ratio, most of manufacturers select vacuum hydraulic press and controllers that have multi-temperature and pressure control device. 6) Un-stack: After lamination finishing, copper clad laminate will be transferred by conveyor into un-stack area, the mirror face steel plate is moved to temporary stock area by suction system, then laminate board is delivered to edge shearing process. 7) Steel plate brushing: There are many mirror face steel plates used in the laminating process. During the laminating process, high pressure and high temperature make the surface of steel plate rough and scratched during moving process. So the surface brushing and maintenance 9

19 are very important to prolong the life of steel plates and cost down as well. 8) Edge shearing: The purpose of this step is to make packaging process easily. On the other hand, it enhances the efficiency of sizing process. 9) Packaging: The packaging process can divide into full size packaging and custom-made size packaging 10) Stock management: In order to match the customer request and to make the production stable, proper stock is necessary. Three Steps A-Stage B-Stage C-Stage Compounding / A-staging Treating / B-staging Lamination / C-staging Resin Prepreg Laminate Reinforcement (Woven glass) Treater (Heated Space Where B-Staging Occurs) Prepreg Heated Press Copper Clad Laminate Varnish Copper Figure 2.2.Electrcal laminate processing 2. PCB Process The PCB industry offers a wide array of materials to meet different performance and cost requirement. Copper foil is the primary conductive layer which can provide good electrical conductivity and process stability. The copper foil is typically laminated together with epoxy resin and woven glass to serve as the substrate of PCB. The board is double sided, with through-hole plating, green solder resist, and white silkscreen printing. Both surface mount and through-hole components have been used for the production of print circuit board. Figures 2.3 & 2.4 (9) depict a standard multilayer s procedure and a 10

20 sequential lamination multilayer procedure, respectively. The key processes are as follows: 1) Pattering (Etching) The vast majority of printed circuit boards are made by bonding a layer of copper over the entire substrate, sometimes on both sides, (creating a "blank PCB") then removing unwanted copper after applying a temporary mask (eg. by etching), leaving only the desired copper traces. 2) Lamination: Some PCBs have trace layers inside the PCB and are called multi-layer PCBs. These are formed by bonding together separately etched thin boards. 3) Drilling: The drilling is performed by automated drilling machines with placement controlled by a drill tape or drill file. When very small vials are required, drilling with mechanical bits is costly because of high rates of wear and breakage. In this case, the vials may be evaporated by lasers. Laser-drilled vials typically have an inferior surface finish inside the hole. These holes are called micro vials. 4) Exposed conductor plating and coating: Electroplating is the coating of an electrically conductive object with a layer of metal using electrical current. The result is a thin, smooth, even coat of metal on the object. A new genre of high-density interconnect (HDI) boards is making the transition from leading edge to mainstream. These boards are characterized by combining a series of complexity features that include buried and blind vias, high-aspect ratio plating, small-hole plating, and fine lines and spaces side by side with ground plane areas of different sizes. 5) Solder resist: Areas that should not be soldered to may be covered with a polymer solder resist (solder mask) coating. The solder resist prevents solder from bridging between conductors and thereby creating short circuits. Solder resist also provides some protection from the environment. 11

21 6) Screen Printing: Line art and text may be printed onto the outer surfaces of a PCB by screen printing. When space permits, the screen print text can indicate component designators, switch setting requirements, test points, and other features helpful in assembling, testing, and servicing the circuit board. 7) Test: A computer will instruct the electrical test unit to apply a small voltage to each contact point on the bed-of-nails as required, and verify that such voltage appears at other appropriate contact points. 8) Printed circuit assembly: There are a variety of soldering techniques used to attach components to a PCB. High volume production is usually done with machine placement and bulk wave soldering or reflow ovens, but skilled technicians are able to solder very tiny parts (for instance 0201 packages which are 0.02" by 0.01") by hand under a microscope, using tweezers and a fine tip soldering iron for small volume prototypes. Some parts are impossible to solder by hand, such as ball grid array (BGA) package. 9) Protection and Packaging: PCBs intended for extreme environments protection often have a conformal coating, which is applied by dipping or spraying after the components have been soldered. The coat prevents corrosion and leakage currents or sorting due to condensation. Many assembled PCBs are static sensitive, and must be placed in antistatic bags during the transporting to avoid the damage. 12

22 Figure 2.3. Standard multilayer s procedure Figure 2.4. Sequential lamination multilayer s procedure 13

23 2.2 Forecasting Models The forecast process begins with a need to make the decisions that depend on the future value or on the future occurrence of some event. It s important for managers to make the decision by using right forecasting process. We refer Wilson and Keating (2009) to divide the entire forecasting process into nine steps as follows: 1) Specify objectives. 2) Determine what to forecast. 3) Identify time dimensions. 4) Data considerations. 5) Model selection. 6) Model evaluation. 7) Forecast preparation. 8) Forecast presentation. 9) Tracking results. These begin and end with communication, cooperation, and collaboration between managers who use the forecasts and technicians who prepare them. This communication is critical if the forecasting is to have the desired positive effect on decision. The objectives related to the decisions for which a forecast is important should be start clearly. Then you must decide exactly what to forecast. There are two types of time dimensions to consider and one is the length of the forecast horizon and the time interval. The data can be from different sources and need to be well managed. For model selection, it needs to be related to the type and quantity of data, the length of the forecast horizon. If we have sufficient historical data, we can use holdout period for evaluation. Once you are satisfied with a model based on historical and holdout period evaluation, you should specify the model using all the available data of historical and holdout then use it for the actual forecast. 14

24 Developing a forecast for new products by new tools and models is an especially difficult task because little or no historical data are available. It forces to use of the methods based on judgments and various marketing research methods. We try to apply the different forecast models for data mining to the global CCL market and to find the best forecasting model for this industry. Industry practitioners use both quantitative and qualitative method (Wilson and Keating, 2009) to do forecasting. Qualitative forecasting methods include the expert system, the Delphi method, etc. Quantitative forecasting including regression analysis, time series analysis, exponential smoothing, neural networks and Grey forecasting model (GM). Several studies have proposed time series models for industrial production and revealed the applicability of time series models to industrial production forecasting. These methods typically require large amounts of data to construct the forecasting. Hsu(2003) showed that the GM requires minimal data and is the best among all existing model for short-term prediction. The Grey model (GM) is applied to predict future trend in the global CCL industry, The GM has the following advantages: 1) It can be used in situation with relatively limited data down to as little as four observations. 2) Just a few discrete data are sufficient to characterize an unknown system 3) It is suitable for forecasting in completive environment where decision-makers only have access limited historic data. Moreover, three residual modification models were applied to enhance to enhance the GM model. Chang et al. (2005) ever applied a variable P value Rolling Grey Forecasting Model (RGM) for Taiwan semiconductor industry production. The univariate GM model makes forecast of a time series of data without considering possible correlation with any leading indicators. But for RGM, we can hypothesize to make variable P value to generate more accurate forecast. They applied real GDP by U.S. manufacturing industry from and find the P equation Then average residual error can reduce from 21.27% to 12.55%. Under the same period, we applied the yearly survey of anticipated industrial production growth rate and average residual error is reduced to 10.52%. 15

25 Hsu and Wang (2007) used grey model improved by Bayesian analysis for forecasting the output of integrated circuit industry. In this case, Bayesian method can be considered as an alternative approach to the classical approach to statistical analysis. The motivations behind adopting a Bayesian method are described as follows. First, prior knowledge or pilot information can easily be incorporated into methods. Such information is specified in a prior distribution on which inference is based. Second, the parameters in the model may be simulated directly via methods for the exploration of posterior distributions. Through BGM (1,1) assistance, we can improved MAPE from 19.63% by GM(1,1) to 8.38% by BGM(1,1) for Taiwan IC annual value for year The most famous first-purchase diffusion model in market research was the Bass innovation diffusion model which combined the modified exponential function and the logistic function diffusion model. The Bass model (Bass, 2004) is well known and widely applied in developing product life-cycle curve, while also being used to forecast the sales volume of initial purchases of new products. The basic assumption of the model is that the timing of a consumer s initial purchase is related to the number of previous buyer. Bass presumes that the initial purchase of the product is made by both innovators and imitators. Innovators and imitators are distinguished by how they are influenced by buying patterns. The number of people who have already bought the product does not influence the timing of an invocator s initial purchase, but does influence imitators. Imitators learn in some sense from those who have already bought the product. Tseng and Hu (2009) combine fuzzy regression with bass model to develop a quadratic-interval bass diffusion model. The empirical analysis show the quadratic interval bass model can be applied to new products, and can reveal the best and worst- case sales volume outcome. If the data are not sufficient, quadratic interval bass diffusion model are potentially useful tools. However, when there is high variability in the data, the quadratic interval bass model should not be used. Tsaur 16

26 (2008) applies fuzzy grey regression model, Watada s fuzzy regression model, linear regression model and GM(1,1) to forecasting the LCD TV. Based on the empirical data analysis, fuzzy grey regression can make the good forecast and also provide the decision makers with the best and worst-possible scenarios 17

27 Chapter 3 Methodology We will use Prismark historic data by different market segments then apply GM(1,1), RGM and Bass Innovation diffusion Model to analyze the data. Year 2008 data is reserved for post verification. 3.1 Forecasting Models (1) GM(1,1) The Grey forecasting model GM(1,1) is a time series prediction model encompassing a group of differential equations adapted for parameter variance as well as a first order differential equation. GM(1,1) can be denoted by the following function (Hsu,2003). xˆ u k (0) ( ) a a( k 1) x (1) (1 e e a k 2,3,..., n. (0) 0 ) (1) where a or u is estimated using OLS. To obtain the residual modification of GM(1,1), we (0) need to define the value of residual q ( k), (0) (0) (0) (0) (0) q q (2), q (3),..., q ( n), q ( k) (0)( ) ˆ (0) x k x0 ( k) k 2,3,..., n (2) According to q ( 0), Eq.(2) denote the residual GM(1,1). The value of u q or a q is estimated using ordinary least- squares (OLS). q (0) uq k) q (1) (1 ) ( 1 e a q e a q k, k 2,3,..., n (3) aq (0) ( ) 18

28 (0) (0) (0) where qˆ (1) q (1). Combining q ˆ ( k) and GM (1,1) yields the residual modification of GM (1,1): (0) (0) uq ( k) ˆ0 ( ) (1) (1 ) ( 1 x k q e a q e a q k, k 2,3,..., n (4) aq (0) ˆ ) xr (2) Rolling GM (1,1) Methodology Rolling GM (1,1) is to construct the model by creating a sequence of one-order liner moving (Chang et al., 2005). The first-order differential equation for the model is (1) dx a (1) X u dt (5) where t denotes the independent variables in the system, a represents the developed coefficient, and u is the Gray controlled variable. The parameters to be determined in the model are a and u. Now, we define the following: ( 0) X ( i; k) (0) X ( i), (0) X ( i 1), (0) X ( i 2),..., (0) X ( k), where r k i 1 (6) When 1, i ( 0) ( 1; r) (0)(1), (0)(2), (0)(3),..., (0) X X X X ( r) X r is the length of the rolling interval. In constructing the model, the Grey system must apply a one-order accumulated generating operation (AGO) to the primitive sequence to provide the middle message for building a model to weaken the tendency toward variation. The AGO of X 0) ( 1) ( k ( (i;k) is defined as X i; ). This is 19

29 1) ( ; ) (1) ( ), (1) ( 1), (1) ( 2),... (1) X ( i k X i X i X i X ( k i i 1 i 2 = ( 0) x ( j), (0) x ( j), (0) x ( j),..., (0) x ( j) (7) j i j i j i k ) j i From Eqs. (5) and (7) and the ordinary least-square method, coefficient â becomes: a B T 1 aˆ T ( B) B Y N u (8) Furthermore, the accumulated matrix B is: B P (1) (1) P x ( i) (1 P) x (1) x ( i 1) (1 P) x ( i 1) (1) ( i 2) 1 1 (1) ( 1) (1 ) (1) P k P x ( k) 1 x where P is equal 0.5 in the original model. The constant vector Y N is: Y N X ( 0)( i 1), X (0)( i 2),..., X (0)( k) T The approximate relationship can be obtained by substituting â obtained in the differential equation and solving Eq. (1) as follow: u u xˆ ( 1) ( t 1) ( (0) x (1) ) e at (9) a a (1) (0) When xˆ (1) xˆ (1), the sequence one-order inverse-accumulated generating operation (IAGO) is acquired. The sequence that must be reduced as Eq. (6) can be obtained as follow: 20

30 xˆ ( 0) (1) (1) t ( t 1) xˆ ( t 1) xˆ ( ) (10) Given t 1,2,..., k, obtain the sequence of reduction as follow: ˆ (0) ( ; ) ( ˆ (0)(1), ˆ (0)(2),..., ˆ (0) X i k x x x ( k 1)) (11) ( 0) ( 0) where xˆ ( k 1)) is the Grey elementary forecasting value for x ˆ ( k 1)). (3) Bass Model Methodology Bass diffusion model (Bass, 2004) is probably the most notable model for new product forecasting. It has been adapted for the use in forecasting a wide variety of products with short products life cycle and the new products with limited historical data. The model is: S t pm ( q p)* Y ( q / m)* 2 Y (12) t t where: S t =Sales at time periodt. p = Probability of initial purchase at time t 0. This reflects the importance of innovators and is called the coefficient of innovation. m = Number of initial purchases of product over the life cycle (excludes replacement purchases). q = Coefficient of imitation representing the propensity to purchase based on the number of people who have already purchased the product. Y t = Number of previous buyers at time t. 21

31 The generic algorithm (GA), developed by Holland in 1975, is based on the Darwinian theory of biological evolution. The main process imitated the nature genetic process, crossover, to exchange some of these individuals generic data randomly to generate the offspring. In additional, GA also simulates another process, mutation, to change some of these individuals genetic data randomly to generate the new population. By repeating these processes until the best genes are found with the most fit. When GA evaluates the fitness value, the process would remove to selection until achieving convergent condition. We apply GA software EVOLVER (2000) to find the p, q, m of Bass Model and MAPE for global CCL market data Performance Metrics According to Wilson and Keating (2009), we can apply seven metrics to evaluate the accuracy of forecasting models over a number of periods. The details are given as follows: 1) ME (mean error)= n ( At F t) 1. n 2) MAE (mean absolute error)= 3) MPE (mean percentage error)= t n At F t t 1. n n ( At F t) / At. n t 1 4) MAPE (mean absolute percentage error)= 5) MSE (mean-squared error)= n ( At Ft) t 1 n 2 n ( A F ) / A t 1 n 2 ( At Ft) 6) RMSE (root-mean-squared error)= t 1. n. t n t t. 22

32 7) Theil s U can be calculated in several ways, two of which are shown in n U= ( A F ) ( A A t 1 t t 2 n t 1 t t 1) 2 U=RMSE(model)/RMSE(no-change model) where A t is actual value in period t, F t is forecast value in period t and n is number of periods used in the calculation. According to above seven metrics, MAPE is selected to evaluate the accuracy of each forecasting model. The lower MAPE value shows the excellent forecasting ability by referring to Table 3.1. Table 3.1. Criteria for forecasting accuracy MAPE(%) Forecasting Power < 10 Excellent Good Reasonable >50 Incorrect Source: Lewis (1982) 23

33 Chapter 4 Forecasting Analysis for Global CCL We use Prismark historic data of global CCL demand from by six market segment and 2008 demand as the actual data to compare to access the accuracy of different forecasting models.. For Paper, composite, FR-4 and FR-4 high Tg market are mature market. So we apply data as the histrionic period data and 2008 data as the holdout period data. FR-4 halogen free, Specialty (high speed, low loss), BT and BT equivalent and FR-5 are new market and application. So we apply data as the histrionic period data and 2008 data as the verification data period. The global CCL market can be divided into six market segments. Paper and composite products are used in products for consumers market such as radio, washing machines, refrigerators etc. FR-4 is used mainly for Cellular phones, Computer.FR-4 high Tg is for servers which are computers used in telecommunications and FR-4 Halogen free products are requested to meet new environment regulation. Specialty products are for high speed servers & communication and IC substrate application. We utilize GM (1,1) theory for limit data base and find all MAPE by 6 market segment are over 10% and even high than 20% for FR-4 Halogen Free (27.42%).We also get the a and u value by OLS (See table 4.1) It shows this forecasting model is not suitable for CCL market analysis and the main reason are as follows. (1) 2008 global economic crisis dramatically reduced all demand in 3Q and this continued into 4Q. (2) GM (1,1) can t sense this change by limit data base. 24

34 Table 4.1 The parameter estimation of GM(1,1) by different markets Paper Composite FR-4 FR-4 FR-4 Specialty Total High Tg Halogen Free (High Speed, Low Loss) a u Rolling GM is by a sequence of one order liner moving and shows better MPAE than GM (1,1). It has big improvement for paper, composite, FR-4 high Tg and total market analysis. The most famous first purpose diffusion model in marketing research was bass diffusion model which combine the imitation and innovation rate. We apply GA model by Software EVOLVER (2000) to find the p, q, m (see Table 4.2) then calculate MAPE for six market segments. Table 4.2 The parameter estimation of Bass by different markets Paper Composite FR-4 FR-4 High Tg FR-4 Halogen Free Specialty(High speed,low Loss) Total m p q Comparison by Different Market Segment Paper CCL market is a traditional application and the majority is for consumers products like radios, calculator, toys etc. This market is mature and will be replaced by composite and FR-4. Table 4.3 shows the results of GM (1,1), RGM and Bass in paper CCL sales forecasting. RGM is a good forecasting model since the MAPE is 1.8% for historical period and 4.2% for holdout period, Bass is the second and GM (1,1) is the worse for this market segment (See Table 4.3 & Figure 4.1). 25

35 Table 4.3. Paper CCL market analysis by different forecasting models Year Actual value GM(1,1) RGM Bass historical period MAPE 3.1% 1.8% 2.2% holdout period MAPE 10.6% 4.2% 5.4% Volume (MM m 2 ) Paper Actual value GM(1,1) RGM Bass Year Figure 4.1. The plots of the original and predicted values (Paper) Composite CCL market is similar to paper CCL market and application. The main difference is substrate and higher performance for electronic property. Bass model show good performance for composite market. The MAPE for holdout period is 3.2% for 2008 and is better than RGM MAPE 4.0% and GM (1,1) MAPE 11.7% (See Table 4.4 & Figure 4.2). 26

36 Table 4.4. Composite CCL market analysis by different forecasting models Year Actual value GM(1,1) RGM Bass historical period MAPE 4.3% 3.6% 5.1% holdout period MAPE 11.7% 4.0% 3.2% Volume (MM m 2 ) Composite Year Actual value GM(1,1) RGM Bass Figure 4.2. The plots of the original and predicted values (Composite) The main applications for FR-4 are cellular phones, notebook computers and liquid crystal displays (LCD). This product family represents 44-48% share of global CCL demand. It declines by 2% in 2008 due to global financial crisis. Bass model show the good performance for FR-4 market. The MAPE for holdout period is 8.1% and is better than RGM at 13.2% and GM (1,1) at 14.2% for 2008 (See Table 4.5 & Figure 4.3). 27

37 Table 4.5. FR-4 CCL market analysis by different forecasting models Year Actual value GM(1,1) RGM Bass historical period 3.4% 1.3% 9.4% MAPE holdout period MAPE 14.2% 13.2% 8.1% Volume (MM m 2 ) FR Actual value GM(1,1) RGM Bass Year Figure 4.3. The plots of the original and predicted values (FR-4) The main application for FR-4 high Tg is server for communication. RGM shows good performance with MAPE 7.9%, Bass by 10.0% and GM (1,1) is the worse with MAPE 21.1% (See Table 4.6 & Figure 4.4). 28

38 Table 4.6. FR-4 high Tg CCL market analysis by different forecasting models Year Actual value GM(1,1) RGM Bass historical period MAPE 5.7% 1.3% 5.1% holdout period MAPE 21.1% 7.9% 10.0% Volume (MM m 2 ) FR-4 High Tg Actual value GM(1,1) RGM Bass Year Figure 4.4.The plots of the original and predicted values (FR-4 high Tg) Due to changes in environmental regulations, FR-4 halogen free products started being sold in the marketing place in year All global cellular phone producers such as Nokia, Motorola, Son-Ericson, Apple, LG, and Samsung have already shifted to 100% halogen free. This is why demand for halogen free CCL increased by 40% in 2007! Bass model show excellent performance for holdout period by 0.04% in 2008 and GM (1,1) is 7.32% then 12.92% for RGM (See Table 4.7 & Figure 4.5). 29

39 Table 4.7. FR-4 Halogen Free CCL market analysis by different forecasting model Year Actual value GM(1,1) RGM Bass historical period MAPE 9.10% 8.11% 14.46% holdout period MAPE 7.32% 12.92% 0.04% Volume (MM m 2 ) 40 FR-4 Halogen Free Actual value GM(1,1) RGM Bass Year Figure 4.5. The plots of the original and predicted values (FR-4 Halogen Free) The main applications for specialty CCL products are electronics used in defense, medical system, high speed communication boards and IC substrate. This market size is limited but the value is higher. Bass model show good performance at 0.4% for holdout period in 2008 and 22.0% for GM (1,1) and 22.1% for RGM (See Table 4.8. & Figure 4.6). 30

40 Table 4.8. Specialty CCL market analysis by different forecasting models Specialty (High Speed & Low loss, BT,BT-Equivalent and FR-5) Actual value GM(1,1) RGM Bass historical period MAPE 0.4% 0.4% 0.4% holdout period MAPE 22.0% 22.3% 0.4% Volume (MM m 2 ) Speciality Actual value GM(1,1) RGM Bass Year Figure 4.6.The plots of the original and predicted values (Specialty) The global CCL market continues to grow from 2001 to 2007 then maintained in The applications of CCL cover all electronic application for consumer, computer, communication, car and are the basic material for PCB. The Bass model shows good performance for holdout period MAPE of 2.8%, 7.9% for RGM and 11.4% for GM (1,1) (See Table 4.9. & Figure 4.7). 31

41 Table 4.9.Global CCL market analysis by different forecasting models Year Actual value GM(1,1) RGM Bass Historical period MAPE 1.6% 0.8% 5.8% Holdout period MAPE 11.4% 7.9% 2.8% Total Volume (MM m 2 ) Year Actual value GM(1,1) RGM Bass Figure 4.7. The plots of the original and predicted values (Global total CCL) 4.2. Hierarchical Forecasting For Global CCL Market Hierarchical forecasting (HF) is a family-based forecast methodology. The two level of the global CCL market are determined by the analysis of six market segments. We apply the best data (refer Table 4.10) by ranking one from holdout period for each market segment to compare with the total data by bass diffusion model (See Table 4.11.) It shows the bass diffusion model is better than HF for global CCL market forecast analysis. 32

42 Table MAPE ranking by three forecasting models Specialty Holdout FR-4 FR-4 High (High, period Paper Composite FR-4 Halogen Tg Speed, Low MAPE Free Loss) GM(1,1) RGM Bass Table MAPE comparison by HF & Bass Diffusion Model 2008 HF BASS Holdout period MAPE 4.5% 2.8% 4.3. MAPE Comparison Through our detail analysis of the global CCL market by three forecasting model- GM(1,1), RGM and Bass, we found RGM to show better performance than GM(1,1) and Bass based on Table 4.12 & Figure 4.8. Table Comparison of all models for historic period (MAPE) Historic period MAPE Paper Composite FR-4 FR-4 High Tg FR-4 Halogen Free Specialty( High Speed, Low Loss) Total GM(1,1) 3.10% 4.30% 3.40% 5.70% 9.10% 0.40% 1.60% RGM 1.80% 3.60% 1.30% 1.30% 8.10% 0.40% 0.80% Bass 2.20% 5.10% 9.40% 5.10% 14.4% 0.40% 5.80% 33

43 MAPE(%) 16.00% 14.00% 12.00% 10.00% 8.00% 6.00% 4.00% 2.00% 0.00% Paper FR-4 FR-4 Halogen Free Total Products GM(1,1) RGM Bass Figure 4.8. MAPE by different forecasting models for histrionic periods For holdout period MAPE, we found Bass to show the better performance than GM(1,1) and RGM based on Figure 4.9 and Table MAPE(%) 30.00% 20.00% 10.00% 0.00% Paper FR-4 FR-4 Halogen Total Market GM(1,1) RGM Bass Figure 4.9. MAPE by different forecasting models for holdout periods 34

44 Table Comparison of all models for holdout period (MAPE) Models Paper Composite FR-4 FR-4 High Tg FR-4 Halogen Free Specialty (High Speed, Low Loss) Total GM(1,1) 10.6 % 11.7 % 14.2 % 21.0 % 7.32 % 22.0 % % RGM 4.20 % 4.00 % 13.2 % 7.90 % 12.9 % 22.3 % 7.90 % Bass model 5.40 % 3.20 % 8.10 % % 0.04 % 0.40 % 2.80 % 35

45 Chapter 5 Conclusions and Future Research 5.1 Conclusions Global CCL market forecasting is the important for CCL/PCB manufactures to make the strategic decision for future expansion and investment. It also brings valuable information and indications for future investment by products of CCL raw materials such as copper foil, glass cloth, Epoxy producers for their future investment. We used Prismark-global consultant company data demand for six market segments and applied three forecasting models - GM (1,1), RGM and Bass. This is a pioneer to study this global market by different forecasting model and the key findings are as follows. 1) We investigate the demand forecasting model for global CCL market by GM(1,1), RGM and Bass model. GM(1,1) show good forecast accuracy of mean absolute percent error(mape) for historical period from 0.4% to 8.11% but the worse MAPE accuracy for holdout period from 7.32% to 22%. 2) RGM also show the better performance than GM(1,1) but still can t improve the MAPE accuracy for six market segments below 10%. 3) We also use HF to apply the best data from each market segment to compare with the holdout period data from bass diffusion model. The MAPE of HF is 4.5% and is worse than bass diffusion model (MAPE=2.8%). 4) The Bass model significantly improves MAPE for all market segment forecast for holdout period in 2008 MAPE below 10% and the lowest is 0.04% for FR-4 halogen free. In conclusions, this study shows the Bass model to have excellent MAPE improvement for holdout period in 2008 for global CCL market. It can also provide the forecasting solution for CCL/PCB manufactures for their future strategy investment. 36

46 5.2 Future Research Future research issues could include the price index of CCL and the values chain of upstream and downstream volume as the factors to improve BASS forecasting accuracy. Furthermore, it is worthy of investigating the turning point of the global CCL market. In addition, other models such as Fuzzy grey regression, Neural plus Fuzzy and Fuzzy plus Grey could be used for future global CCL market analysis. 37

47 References 1. Bass, F.M., (2004). Comment on: A new product growth model for consumer durables. Management Science, 50, Chang, P.C., Liu, C.H. and Lai, R. K., (2008). A fuzzy case-based reasoning model for sales forecasting in print circuit board industry. Expert Systems with Applications, 34, Chang, S.C., Lai, H.C. and Yu, H.C., (2005). A variable p value rolling grey forecasting model for Taiwan semiconductor industry production. Technological Forecasting & Social Change, 72, Deng, J.L., (1989). Introduction to grey system. Journal of Grey System, 1, Evolver, (2000). Software. Newfield, NY: Palisade Corporation. 6. Hsu, L.C., (2003). Applying the grey prediction model to the global integrated circuit industry. Technological Forecasting & Social Change, 70, Hsu, L.C. and Wang, C.H., (2007). Forecasting the output of integrated circuit industry using a grey model improved by the bayesian analysis. Technological Forecasting & Social Change, 74, Hsu, L.C. and Wang, C.H., (2009). Forecasting integrated circuit output using multivariate grey model and grey relational analysis. Expert Systems with Applications, 36, Lai, C.J., (2002). A modified rolling grey model for nonlinear time series forecasting. Journal of Grey System, 14, Lewis, C.D., (1982). Industrial and business forecasting methods. London: Butterworth Scientific. 12. Prismark, (2009). Annual report. 38

48 13. Tsaur, R.C., (2008). Forecasting analysis by using fuzzy grey regression model for solving limited time series data. Soft Computer, 12, Tseng, F.M and Hu, Y.C., (2009). Quadratic-Interval Bass model for new products sales diffusion. Expert Systems with Applications, 36, Wilson, J.H. and Keating, B., (2009). Business forecasting with forecast X. Sixth Edition. Boston, MA: Mc-Graw Hill. 39

49

50

51

52 摘要需求預測為高階主管在未來策略投資之重要參考之一 銅箔基層板為印刷電路板之基礎材料, 廣泛運用在民生家電 電腦 通訊 手機 汽車 醫療 軍事用途 而全球印刷電路板在 2008 年之產值為美金四佰捌拾貳億 在本研究中, 我們運用灰色模式, 滾動灰色模式, 貝氏擴散模式, 分析 2001 年至 2008 年全球銅箔基層板六大產品市場, 分別為紙基板 複合板 FR-4 板 高玻璃轉化温度之 FR-4 板, 無鹵素之 FR-4 板與特殊板. 而運用平均絕對值百分比誤差 (Mean Absolute Percentage Error, 簡稱 MAPE) 進行預測效益評估 本研究結果分析, 貝氏擴散模式之平均絕對值百分比誤差低於灰色模式與滾動灰色模式, 可運用在全球銅箔基層板之預測分析 關鍵字 : 銅箔基層板, 印刷電路板, 灰色模式, 滾動灰色模式, 貝氏擴散模式 I

53 Abstract Demand forecasting is one of critical reference by top managers to make the strategy decision for future investment. The copper clad laminate (CCL) is the key material for print circuit board (PCB) and it can apply for consumer, computer, LCD, communication, automotive, aero space, medicine and defense application. The total global sale for PCB in 2008 is US$ 48.2 billion. In this research, we use grey model GM(1,1), rolling grey model (RGM) and Bass diffusion model to analysis global CCL market by six market segments paper, composite, FR-4, FR-4 High Tg, FR-4 halogen free, Specialty between The forecasting accuracy of global CCL market by six market segment was evaluated along with mean absolute percentage error (MAPE). In this study, Bass diffusion model MAPE outperforms the others two models GM(1,1) and RGM for this global CCL market forecasting analysis and is recommend for global CCL market forecasting analysis. Keywords: Copper clad laminate, Print circuit board, GM(1,1), Rolling GM, Bass diffusion model II

54 誌謝在東海研究所畢業二十四年後 重回校園, 當起學生, 與眾多社會精英, 經歴每週上課, 小組討論, 哈佛個案分析, 上台報告, 交學期作業, 在課內課外相互切磋, 擴展領域, 追求成長 不齗學習與自我超越 在這二年 EMBA 碩士生涯中, 最重要的是完成人生中的第二篇碩士論文 最需感謝的就是恩師王福琨教授, 在碩一的研究方法課程中, 將我引領進入預測領域, 一窺此領域的奧妙, 與其所帶來的研究樂趣! 而在口試期間, 清大林則孟教授及本院歐陽超副院長給予諸多的建議, 使本論文更臻完備 同時感謝 PRISMARK 姜旭高博士, 提供全球铜箔基層板數據, 作為本研究之預測分析母體 本公司的林宗輝總經理與太平洋地區商業總監 -Carols Lansangan 的支持, 讓我能在這二年當中, 能工作與學業兼顧, 更感謝美國同事 - Julian Yvonne 在內容的校稿與修正 也感謝 MA529 研究室的世賢與道明, 在軟體分析的協助 最感謝的是子娟這二年的陪伴與支持, 老大文博與老二文祺的陪讀, 享受一家人在一起研讀, 準備報告與完成論文 更感謝爸媽這二年的支持與鼓勵, 讓我們無後顧之憂 最後謹以此文, 獻給所有關心與支持我的每一個人! 蕭裕耀 謹誌於台灣科技大學 民國 99 年 4 日 12 日 III

55 Contents Chapter 1 Introduction Motivation Research Objectives Research Procedure 4 Chapter 2 Literature Review Global CCL Industry Forecasting Models.14 Chapter 3 Methodology Forecasting Models Performance Measurement...22 Chapter 4 Forecasting Analysis for Global CCL Comparison by Different Market Segment Hierarchical Forecasting For Global CCL Market MAPE Comparison 33 Chapter 5 Conclusions and Future Reasrch Conclusions Future Research.37 References 38 IV

56 LIST OF FIGURES Figure 1.1. The total global sales for PCB from Figure 1.2. The geographic migration of PCB manufacturing...2 Figure 1.3. The flowchart of research 5 Figure 2.1. Global electronic industry value chain analysis...6 Figure 2.2. Electrical laminate processing Figure 2.3. Standard multilayer s procedure Figure 2.4. Sequential lamination multilayer Figure 4.1. The plots of the original and predicted values (Paper) Figure 4.2. The plots of the original and predicted values (Composite) Figure 4.3. The plots of the original and predicted values (FR-4). 28 Figure 4.4. The plots of the original and predicted values (FR-4 high Tg)..29 Figure 4.5. The plots of the original and predicted values (FR-4 Halogen Free)...30 Figure 4.6. The plots of the original and predicted values (Specialty).31 Figure 4.7. The plots of the original and predicted values (Global total CCL)...32 Figure 4.8. MAPE by different forecasting models for historic periods...34 Figure 4.9. MAPE by different forecasting models for holdout periods...34 V

57 LIST OF TABLE Table 1.1. Global CCL market demand by products segment.. 3 Table 2.1. IPC definition of CCL products..8 Table 3.1. Criteria for forecasting accuracy..23 Table 4.1. The parameter estimation of GM(1,1) by different markets...25 Table 4.2. The parameter estimation of Bass by different markets...25 Table 4.3. Paper CCL market analysis by different forecasting models...26 Table 4.4. Composite CCL market analysis by different forecasting models...27 Table 4.5. FR-4 CCL market analysis by different forecasting models...28 Table 4.6. FR-4 high Tg CCL market analysis by different forecasting models...29 Table 4.7. FR-4 Halogen Free CCL market analysis by different forecasting model...30 Table 4.8. Specialty CCL market analysis by different forecasting models...31 Table 4.9. Global CCL market analysis by different forecasting models..32 Table MAPE ranking by three forecast model.33 Table MAPE comparison by HF & Bass Diffusion Model..33 Table Comparison of all models for historic period (MAPE)...33 Table Comparison of all models for holdout period (MAPE) VI

58 Chapter 1 Introduction 1.1. Motivation Copper clad laminates (CCLs) are composite materials manufactured by pressing layers of filler material impregnated with resin together with layers of copper foil under heat and pressure. Laminates provide mechanical support for electronic components interconnect them electronically. The most common fillers materials used are paper and glass fiber. Paper has been used as reinforcement in a vast majority of printed circuit boards (PCBs). Paper Laminates are low priced and easily manufactured. Composite Epoxy Materials (CEM) is the group of composite materials typically made of woven glass fabric surfaces and non-woven glass core combined with epoxy resin. They are two major products CEM-1 which is low cost, flame retardant, cellulose paper based laminate with only one layer of woven glass fabric & CEM-3 which is very similar to the most commonly used PCB material- FR4 but modified with white color and flame retardant. The most widely used and industry standard material for PCB process is FR-4 with Tg 135 degree C. The Tg of high performance FR-4 is 170 degree C and BT is 195 degree C, where Tg is glass transition temperature, a measure of temperature resistance. The CCL is the key materials for print circuit board.the total global sale for PCB in 2008 were US$48.2 billion and the market share by geography are as follows. China/ 31.2%, Asia/21.1%, Japan / 14%, USA/9.3%, Europe/6.7% (see figure 1.1.). In addition, Figure 1.2 depicts the geographic migration of PCB manufacturing. PCB is widely use for Consumers, Communication, Computer, LCD, Automotive, Aero Space, Medical and others electronic products. To align with different PCB applications, manufacturers need to use different raw materials paper, composite, FR-4 high Tg, FR-4, high Tg, FR-4 Halogen Free and Special. All raw materials differences are based on different resin structure and end use application. 1

59 Figure 1.1. The total global sales for PCB from Figure 1.2. The geographic migration of PCB manufacturing CCL forecasting is difficulty due to complex demand from different market. Prismark is a well known global consultant company that collects demand by six market segments and 2

60 issues an annual report. They collect these data one by one (bottom up) then consolidate into completed data (see Table 1.1 for global CCL market demand by products segment from ). Table 1.1. Global CCL market demand by products segment ( M m*2) Paper Composite FR FR-4 High Tg FR-4 Halogen Free Specialty(High Speed, Low Loss),BT and BT-Equivalent, FR-5 Total Source: Prismark But Prismark can only provide limited forecasting data based on their existing estimation by different markets. This is why they intend to develop a new forecasting model for this rapidly changing market and compare the results with their bottom up statistic data. 1.2 Research Objectives The purposes of this study are: 1. To develop an efficient forecasting model that provides insight into global CCL market demand and apply MAPE to find out the best forecasting model. 2. To assist CCL/PCB manufactures in making the strategy decision for future expansion and investment. 3

61 1.3 Research Procedure To determine global trend of CCL market, this thesis attempts to compare different models and then select the most accurate for this market. Generally, a reliable and accurate forecasting method can improve the ability of CCL related industry (Epoxy resin, Copper foil, Glass cloth, CCL and PCB) to make correct investment decision. Data on the levels of demand in the global CCL industry were applied for this forecasting model analysis! The procedure for this research is shown in Figure 1 and is given as follows: Chapter 1 Introduce the motivation, objective and procedure of this research. Chapter 2 Present a literature review including the global CCL market introduction, CCL & PCB process introduction, forecasting model introduction and literature review of GM(1,1), RGM and Bass model. Chapter 3 Present Methodology of GM(1,1), RGM, Bass and Genetic algorithm(ga). Chapter 4 Apply GM(1,1), RGM and Bass diffusion model to analysis global CCL market from 2001 to 2007 and use 2008 data as holdout period by MAPE comparison. Chapter 5 Conclude the research results with summary and provide several future extensions on this research. 4

62 Research Topics Global CCL Industry Background Literature Review Problems definition &Research Target Research Methodology & tools/gm(1,1) RGM BASS Model Global CCL demand Paper Composite FR-4 FR-4 High Tg FR-4 Halogen Speciality (High speed, Low Loss ) Data analysis & verification Conclusions & future study Figure 1.3. The flowchart of research 5

63 Chapter 2 Literature Review This chapter is divided into two sections as follows: 2.1 Global CCL Industry CCL is key raw material for PCB and is a composite of copper foil, glass cloth and epoxy resin. The total values for base raw materials used in CCL manufacture was US$ 25 billion in Epoxy is one of key materials and total value is US$ 0.5 billion (2% share). PCB manufactures use the CCL & FCCL as the base to apply circuit patterns using photo mask/dry film then etching. The value of CCL used is about US$ 5 billion, which transfers to PCB of US $ 45 billion. These PCBs can be incorporated into electronics for consumer, computer, communication, automotive, medical and defense applications with a total value of US $ 950 Billion (see Figure2.1) Figure 2.1. Global electronic industry value chain analysis 6

64 All CCL products definition is based on IPC classifications. IPC is a global trade association dedicated to the competitive excellence and financial success of its more than 2,700 member companies, IPC represents all facets of the industry, including design, printed circuit board manufacturing and electronics assembly (see Table 2.1). Prismark s data analysis by market segment is based on IPC definitions which are as follows: 1) Paper- Single and double-sides paper phenolic or paper epoxy XXXP, FR-1, FR-2, FR-3. 2) Composite: All single-and double sides CEM1, CEM3. 3) FR-4 : All rigid PCBs with more than 2 layers made with FR-4 epoxy and Tg is degree C. 4) FR-4 high Tg : All rigid PCBs with more than 2 layers made with FR-4 epoxy and Tg is above170 degree C. 5) FR-4 Halogen Free: All rigid PCBs with more than 2 layers made with non- Bromine ( Fluorine, Chlorine, Iodine, Astatine) epoxy and Tg is above 140 degree C. 6) Specialty(high speed & low loss),bt, BT equivalent and FR-5: All non-epoxy (non FR-4) rigid boards, including PTFE, rigid polyimide, APPE, Cyan ate Ester, Bismaleimide Triazine (BT) and GETEK; includes 2-side high performance and boards made with exotic materials such as Teflon and blended resin with unusual characteristics. 7

65 Table 2.1. IPC definition of CCL products Class IPC-4101 FR Appr Tg [C] MOT App Dk App Df Rel. Cost Segmen t Comment FR-1: 02 + Paper phenolic resin/cellulose paper, general purpose FR-2: Paper phenolic resin/cellulose paper, higher quality, tested for electrical properties FR-3: Paper epoxy resin/ cellulose paper FR Rigid & ML FR-5: Rigid & ML epoxy resin/ glass cloth, higher peel strength epoxy resin/glass cloth, higher heat resistance FR-6: + polyester resin/glass cloth CEM-1: Paper/Ri epoxy resin/glass cloth/ gid cellulose paper core CEM-3: Rigid epoxy resin/ glass cloth/glass fiber core CRM-5: + 80 Woven E-Glass face sheets, non woven E-Glass core/polyster/kaolin G Composi Woven E-glass fabric / te Epoxy G Composi Woven E-glass fabric / te Epoxy Source: IPC-2008 version 8

66 The type of CCL products, their applications and the raw materials used to make them have been described. The next section explains in detail the manufacturing process and where each raw material is used in the ten process steps. The details of electrical laminate processing as shown in Figure 2.2 and are given as follows: 1) Copper foil cutting: Put the roll of copper foil on the feed frame, and calibrate the beginning allocation. It can set the cutting length by operation panel and executive automatic. 2) Paper cutting: Paper is used as cushion during lamination.they put the roll of paper on the feed frame and calibrates the beginning allocation. Set the cutting length by operation panel and executive automatic. The most of the manufacturers select this way for production efficiency and cost down. 3) Prepreg manufacture: Prepreg is the main method to adhere fiber-glass and copper foil. The manufacturing methods use the conveyor system to deliver the fiber-glass, pass by tank of gel then go into the oven for drying and cutting. The quality of prepreg is related to the life of printed circuit board. 4) Stack: Put the layers between cover and carriage. The material layers are sandwiched between mirror face steel plate (to give a very smooth surface) and paper. 5) Lamination: The main purpose of this step is to melt the prepreg, then to increase the temperature and pressure and to adhere the copper foil and glass fiber closely. Presently, in order to raise the well production ratio, most of manufacturers select vacuum hydraulic press and controllers that have multi-temperature and pressure control device. 6) Un-stack: After lamination finishing, copper clad laminate will be transferred by conveyor into un-stack area, the mirror face steel plate is moved to temporary stock area by suction system, then laminate board is delivered to edge shearing process. 7) Steel plate brushing: There are many mirror face steel plates used in the laminating process. During the laminating process, high pressure and high temperature make the surface of steel plate rough and scratched during moving process. So the surface brushing and maintenance 9

67 are very important to prolong the life of steel plates and cost down as well. 8) Edge shearing: The purpose of this step is to make packaging process easily. On the other hand, it enhances the efficiency of sizing process. 9) Packaging: The packaging process can divide into full size packaging and custom-made size packaging 10) Stock management: In order to match the customer request and to make the production stable, proper stock is necessary. Three Steps A-Stage B-Stage C-Stage Compounding / A- staging Treating / B-staging Lamination / C-staging Resin Prepreg Laminate Reinforcement (Woven glass) Treater (Heated Space Where B-Staging Occurs) Prepreg Heated Press Copper Clad Laminate Varnish Copper Figure 2.2.Electrcal laminate processing 2. PCB Process The PCB industry offers a wide array of materials to meet different performance and cost requirement. Copper foil is the primary conductive layer which can provide good electrical conductivity and process stability. The copper foil is typically laminated together with epoxy resin and woven glass to serve as the substrate of PCB. The board is double sided, with through-hole plating, green solder resist, and white silkscreen printing. Both surface mount and through-hole components have been used for the production of print circuit board. Figures 2.3 & 2.4 (9) depict a standard multilayer s procedure and a 10

68 sequential lamination multilayer procedure, respectively. The key processes are as follows: 1) Pattering (Etching) The vast majority of printed circuit boards are made by bonding a layer of copper over the entire substrate, sometimes on both sides, (creating a "blank PCB") then removing unwanted copper after applying a temporary mask (eg. by etching), leaving only the desired copper traces. 2) Lamination: Some PCBs have trace layers inside the PCB and are called multi-layer PCBs. These are formed by bonding together separately etched thin boards. 3) Drilling: The drilling is performed by automated drilling machines with placement controlled by a drill tape or drill file. When very small vials are required, drilling with mechanical bits is costly because of high rates of wear and breakage. In this case, the vials may be evaporated by lasers. Laser-drilled vials typically have an inferior surface finish inside the hole. These holes are called micro vials. 4) Exposed conductor plating and coating: Electroplating is the coating of an electrically conductive object with a layer of metal using electrical current. The result is a thin, smooth, even coat of metal on the object. A new genre of high-density interconnect (HDI) boards is making the transition from leading edge to mainstream. These boards are characterized by combining a series of complexity features that include buried and blind vias, high-aspect ratio plating, small-hole plating, and fine lines and spaces side by side with ground plane areas of different sizes. 5) Solder resist: Areas that should not be soldered to may be covered with a polymer solder resist (solder mask) coating. The solder resist prevents solder from bridging between conductors and thereby creating short circuits. Solder resist also provides some protection from the environment. 11

69 6) Screen Printing: Line art and text may be printed onto the outer surfaces of a PCB by screen printing. When space permits, the screen print text can indicate component designators, switch setting requirements, test points, and other features helpful in assembling, testing, and servicing the circuit board. 7) Test: A computer will instruct the electrical test unit to apply a small voltage to each contact point on the bed-of-nails as required, and verify that such voltage appears at other appropriate contact points. 8) Printed circuit assembly: There are a variety of soldering techniques used to attach components to a PCB. High volume production is usually done with machine placement and bulk wave soldering or reflow ovens, but skilled technicians are able to solder very tiny parts (for instance 0201 packages which are 0.02" by 0.01") by hand under a microscope, using tweezers and a fine tip soldering iron for small volume prototypes. Some parts are impossible to solder by hand, such as ball grid array (BGA) package. 9) Protection and Packaging: PCBs intended for extreme environments protection often have a conformal coating, which is applied by dipping or spraying after the components have been soldered. The coat prevents corrosion and leakage currents or sorting due to condensation. Many assembled PCBs are static sensitive, and must be placed in antistatic bags during the transporting to avoid the damage. 12

70 Figure 2.3. Standard multilayer s procedure Figure 2.4. Sequential lamination multilayer s procedure 13

71 2.2 Forecasting Models The forecast process begins with a need to make the decisions that depend on the future value or on the future occurrence of some event. It s important for managers to make the decision by using right forecasting process. We refer Wilson and Keating (2009) to divide the entire forecasting process into nine steps as follows: 1) Specify objectives. 2) Determine what to forecast. 3) Identify time dimensions. 4) Data considerations. 5) Model selection. 6) Model evaluation. 7) Forecast preparation. 8) Forecast presentation. 9) Tracking results. These begin and end with communication, cooperation, and collaboration between managers who use the forecasts and technicians who prepare them. This communication is critical if the forecasting is to have the desired positive effect on decision. The objectives related to the decisions for which a forecast is important should be start clearly. Then you must decide exactly what to forecast. There are two types of time dimensions to consider and one is the length of the forecast horizon and the time interval. The data can be from different sources and need to be well managed. For model selection, it needs to be related to the type and quantity of data, the length of the forecast horizon. If we have sufficient historical data, we can use holdout period for evaluation. Once you are satisfied with a model based on historical and holdout period evaluation, you should specify the model using all the available data of historical and holdout then use it for the actual forecast. 14

72 Developing a forecast for new products by new tools and models is an especially difficult task because little or no historical data are available. It forces to use of the methods based on judgments and various marketing research methods. We try to apply the different forecast models for data mining to the global CCL market and to find the best forecasting model for this industry. Industry practitioners use both quantitative and qualitative method (Wilson and Keating, 2009) to do forecasting. Qualitative forecasting methods include the expert system, the Delphi method, etc. Quantitative forecasting including regression analysis, time series analysis, exponential smoothing, neural networks and Grey forecasting model (GM). Several studies have proposed time series models for industrial production and revealed the applicability of time series models to industrial production forecasting. These methods typically require large amounts of data to construct the forecasting. Hsu(2003) showed that the GM requires minimal data and is the best among all existing model for short-term prediction. The Grey model (GM) is applied to predict future trend in the global CCL industry, The GM has the following advantages: 1) It can be used in situation with relatively limited data down to as little as four observations. 2) Just a few discrete data are sufficient to characterize an unknown system 3) It is suitable for forecasting in completive environment where decision-makers only have access limited historic data. Moreover, three residual modification models were applied to enhance to enhance the GM model. Chang et al. (2005) ever applied a variable P value Rolling Grey Forecasting Model (RGM) for Taiwan semiconductor industry production. The univariate GM model makes forecast of a time series of data without considering possible correlation with any leading indicators. But for RGM, we can hypothesize to make variable P value to generate more accurate forecast. They applied real GDP by U.S. manufacturing industry from and find the P equation Then average residual error can reduce from 21.27% to 12.55%. Under the same period, we applied the yearly survey of anticipated industrial production growth rate and average residual error is reduced to 10.52%. 15

73 Hsu and Wang (2007) used grey model improved by Bayesian analysis for forecasting the output of integrated circuit industry. In this case, Bayesian method can be considered as an alternative approach to the classical approach to statistical analysis. The motivations behind adopting a Bayesian method are described as follows. First, prior knowledge or pilot information can easily be incorporated into methods. Such information is specified in a prior distribution on which inference is based. Second, the parameters in the model may be simulated directly via methods for the exploration of posterior distributions. Through BGM (1,1) assistance, we can improved MAPE from 19.63% by GM(1,1) to 8.38% by BGM(1,1) for Taiwan IC annual value for year The most famous first-purchase diffusion model in market research was the Bass innovation diffusion model which combined the modified exponential function and the logistic function diffusion model. The Bass model (Bass, 2004) is well known and widely applied in developing product life-cycle curve, while also being used to forecast the sales volume of initial purchases of new products. The basic assumption of the model is that the timing of a consumer s initial purchase is related to the number of previous buyer. Bass presumes that the initial purchase of the product is made by both innovators and imitators. Innovators and imitators are distinguished by how they are influenced by buying patterns. The number of people who have already bought the product does not influence the timing of an invocator s initial purchase, but does influence imitators. Imitators learn in some sense from those who have already bought the product. Tseng and Hu (2009) combine fuzzy regression with bass model to develop a quadratic-interval bass diffusion model. The empirical analysis show the quadratic interval bass model can be applied to new products, and can reveal the best and worst- case sales volume outcome. If the data are not sufficient, quadratic interval bass diffusion model are potentially useful tools. However, when there is high variability in the data, the quadratic interval bass model should not be used. Tsaur 16

74 (2008) applies fuzzy grey regression model, Watada s fuzzy regression model, linear regression model and GM(1,1) to forecasting the LCD TV. Based on the empirical data analysis, fuzzy grey regression can make the good forecast and also provide the decision makers with the best and worst-possible scenarios 17

75 Chapter 3 Methodology We will use Prismark historic data by different market segments then apply GM(1,1), RGM and Bass Innovation diffusion Model to analyze the data. Year 2008 data is reserved for post verification. 3.1 Forecasting Models (1) GM(1,1) The Grey forecasting model GM(1,1) is a time series prediction model encompassing a group of differential equations adapted for parameter variance as well as a first order differential equation. GM(1,1) can be denoted by the following function (Hsu,2003). xˆ u k (0) ( ) = a a( k 1) x (1) (1 e e a k = 2,3,..., n. (0) 0 ) (1) where a or u is estimated using OLS. To obtain the residual modification of GM(1,1), we (0) need to define the value of residual q ( k), (0) (0) (0) (0) (0) q = [ q (2), q (3),..., q ( n) ], q ( k) = (0)( ) ˆ (0) x k x0 ( k) k = 2,3,..., n (2) According to q ( 0), Eq.(2) denote the residual GM(1,1). The value of u q or a q is estimated using ordinary least- squares (OLS). (0) (0) uq q ( k) = q (1) (1 ) ( 1) e a q e a q k, k = 2,3,..., n (3) aq 18

76 (0) (0) (0) where q ˆ (1) = q (1). Combining q ˆ ( k) and GM (1,1) yields the residual modification of GM (1,1): (0) (0) (0) uq ˆ ( ) ˆ0 ( ) (1) (1 ) ( 1) xr k = x k q e a q e a q k, k = 2,3,..., n (4) aq (2) Rolling GM (1,1) Methodology Rolling GM (1,1) is to construct the model by creating a sequence of one-order liner moving (Chang et al., 2005). The first-order differential equation for the model is dx dt (1) + (1) a X = u (5) where t denotes the independent variables in the system, a represents the developed coefficient, and u is the Gray controlled variable. The parameters to be determined in the model are a and u. Now, we define the following: [ (0) X ( i), (0) X ( i + 1), (0) X ( i 2),..., (0 X ( k) ] where ( 0) ( + ) X i; k) =, r = k i + 1 (6) When =1, i ( 0) ( 1; r) [ (0) (1), (0) (2), (0) (3),..., (0) X X X X ( r) ] X = r is the length of the rolling interval. In constructing the model, the Grey system must apply a one-order accumulated generating operation (AGO) to the primitive sequence to provide the middle message for building a model to weaken the tendency toward variation. The AGO of X ( 0) (i;k) is defined as X ( 1) ( i; k ). This is 19

77 [ (1) ( ), (1) ( 1), (1) ( 2),... (1 X i X i + X i X ( )] ( 1) ( ; ) ) X i k = + k i i+ 1 i+ 2 = ( 0) x ( j), (0) x ( j), (0) x ( j),..., (0) x ( j) (7) j = i j = i j = i k j = i From Eqs. (5) and (7) and the ordinary least-square method, coefficient â becomes: a B T 1 aˆ = T = ( B) B Y N u (8) Furthermore, the accumulated matrix B is: B = [ (1) ( ) + (1 ) (1) P x i P x ( i + 1) ] [ (1) ( i + 1) + (1 P) (1) P x ( i + 2) ] x 1 1 [ (1) ( 1) + (1 ) (1) P k P x ( k) ] 1 x where P is equal 0.5 in the original model. The constant vector Y N is: [ X ( 0)( i+ 1), X (0)( i 2),..., X (0)( k) ] Y N = + T The approximate relationship can be obtained by substituting â obtained in the differential equation and solving Eq. (1) as follow: u u xˆ ( 1) ( t + 1) = ( (0) x (1) ) e at + (9) a a When ˆ (1) (1) ˆ (0) x = x (1), the sequence one-order inverse-accumulated generating operation (IAGO) is acquired. The sequence that must be reduced as Eq. (6) can be obtained as follow: 20

78 xˆ ( 0) (1) (1) t ( t + 1) = xˆ ( t + 1) xˆ ( ) (10) Given t = 1,2,..., k, obtain the sequence of reduction as follow: ˆ (0) ( ; ) ( ˆ (0)(1), ˆ (0)(2),..., ˆ (0) X i k = x x x ( k + 1)) (11) where ˆ ( 0) x ( k + 1)) is the Grey elementary forecasting value for ˆ ( 0) x ( k + 1)). (3) Bass Model Methodology Bass diffusion model (Bass, 2004) is probably the most notable model for new product forecasting. It has been adapted for the use in forecasting a wide variety of products with short products life cycle and the new products with limited historical data. The model is: St = pm + ( q p)* Y t ( q / m)* 2 Y t (12) where: S t =Sales at time periodt. p = Probability of initial purchase at time t = 0. This reflects the importance of innovators and is called the coefficient of innovation. m = Number of initial purchases of product over the life cycle (excludes replacement purchases). q = Coefficient of imitation representing the propensity to purchase based on the number of people who have already purchased the product. Y t = Number of previous buyers at time t. 21

79 The generic algorithm (GA), developed by Holland in 1975, is based on the Darwinian theory of biological evolution. The main process imitated the nature genetic process, crossover, to exchange some of these individuals generic data randomly to generate the offspring. In additional, GA also simulates another process, mutation, to change some of these individuals genetic data randomly to generate the new population. By repeating these processes until the best genes are found with the most fit. When GA evaluates the fitness value, the process would remove to selection until achieving convergent condition. We apply GA software EVOLVER (2000) to find the p, q, m of Bass Model and MAPE for global CCL market data Performance Metrics According to Wilson and Keating (2009), we can apply seven metrics to evaluate the accuracy of forecasting models over a number of periods. The details are given as follows: 1) ME (mean error)= n ( At F t) =1. n 2) MAE (mean absolute error)= 3) MPE (mean percentage error)= t n At F t t =1. n n t =1 4) MAPE (mean absolute percentage error)= 5) MSE (mean-squared error)= n [( ) / ] n A t n ( At Ft) t =1 F 2 t n A t. ( A F ) / A t =1 n 2 ( At Ft) 6) RMSE (root-mean-squared error)= t =1. n. t n t t. 22

80 7) Theil s U can be calculated in several ways, two of which are shown in n n 2 U= ( At F t) ( At At 1) 2 t= 1 t= 1 U=RMSE(model)/RMSE(no-change model) where A t is actual value in period t, F t is forecast value in period t and n is number of periods used in the calculation. According to above seven metrics, MAPE is selected to evaluate the accuracy of each forecasting model. The lower MAPE value shows the excellent forecasting ability by referring to Table 3.1. Table 3.1. Criteria for forecasting accuracy MAPE(%) Forecasting Power < 10 Excellent Good Reasonable >50 Incorrect Source: Lewis (1982) 23

81 Chapter 4 Forecasting Analysis for Global CCL We use Prismark historic data of global CCL demand from by six market segment and 2008 demand as the actual data to compare to access the accuracy of different forecasting models.. For Paper, composite, FR-4 and FR-4 high Tg market are mature market. So we apply data as the histrionic period data and 2008 data as the holdout period data. FR-4 halogen free, Specialty (high speed, low loss), BT and BT equivalent and FR-5 are new market and application. So we apply data as the histrionic period data and 2008 data as the verification data period. The global CCL market can be divided into six market segments. Paper and composite products are used in products for consumers market such as radio, washing machines, refrigerators etc. FR-4 is used mainly for Cellular phones, Computer.FR-4 high Tg is for servers which are computers used in telecommunications and FR-4 Halogen free products are requested to meet new environment regulation. Specialty products are for high speed servers & communication and IC substrate application. We utilize GM (1,1) theory for limit data base and find all MAPE by 6 market segment are over 10% and even high than 20% for FR-4 Halogen Free (27.42%).We also get the a and u value by OLS (See table 4.1) It shows this forecasting model is not suitable for CCL market analysis and the main reason are as follows. (1) 2008 global economic crisis dramatically reduced all demand in 3Q and this continued into 4Q. (2) GM (1,1) can t sense this change by limit data base. 24

82 Table 4.1 The parameter estimation of GM(1,1) by different markets Paper Composite FR-4 FR-4 FR-4 Specialty Total High Tg Halogen Free (High Speed, Low Loss) a u Rolling GM is by a sequence of one order liner moving and shows better MPAE than GM (1,1). It has big improvement for paper, composite, FR-4 high Tg and total market analysis. The most famous first purpose diffusion model in marketing research was bass diffusion model which combine the imitation and innovation rate. We apply GA model by Software EVOLVER (2000) to find the p, q, m (see Table 4.2) then calculate MAPE for six market segments. Table 4.2 The parameter estimation of Bass by different markets Paper Composite FR-4 FR-4 High Tg FR-4 Halogen Free Specialty(High speed,low Loss) Total m p q Comparison by Different Market Segment Paper CCL market is a traditional application and the majority is for consumers products like radios, calculator, toys etc. This market is mature and will be replaced by composite and FR-4. Table 4.3 shows the results of GM (1,1), RGM and Bass in paper CCL sales forecasting. RGM is a good forecasting model since the MAPE is 1.8% for historical period and 4.2% for holdout period, Bass is the second and GM (1,1) is the worse for this market segment (See Table 4.3 & Figure 4.1). 25

83 Table 4.3. Paper CCL market analysis by different forecasting models Year Actual value GM(1,1) RGM Bass historical period MAPE 3.1% 1.8% 2.2% holdout period MAPE 10.6% 4.2% 5.4% Volume (MM m 2 ) Paper Actual value GM(1,1) RGM Bass Year Figure 4.1. The plots of the original and predicted values (Paper) Composite CCL market is similar to paper CCL market and application. The main difference is substrate and higher performance for electronic property. Bass model show good performance for composite market. The MAPE for holdout period is 3.2% for 2008 and is better than RGM MAPE 4.0% and GM (1,1) MAPE 11.7% (See Table 4.4 & Figure 4.2). 26

84 Table 4.4. Composite CCL market analysis by different forecasting models Year Actual value GM(1,1) RGM Bass historical period MAPE 4.3% 3.6% 5.1% holdout period MAPE 11.7% 4.0% 3.2% Composite Volume (MM m ) Year Actual value GM(1,1) RGM Bass Figure 4.2. The plots of the original and predicted values (Composite) The main applications for FR-4 are cellular phones, notebook computers and liquid crystal displays (LCD). This product family represents 44-48% share of global CCL demand. It declines by 2% in 2008 due to global financial crisis. Bass model show the good performance for FR-4 market. The MAPE for holdout period is 8.1% and is better than RGM at 13.2% and GM (1,1) at 14.2% for 2008 (See Table 4.5 & Figure 4.3). 27

85 Table 4.5. FR-4 CCL market analysis by different forecasting models Year Actual value GM(1,1) RGM Bass historical period 3.4% 1.3% 9.4% MAPE holdout period MAPE 14.2% 13.2% 8.1% Volume (MM m 2 ) FR Actual value GM(1,1) RGM Bass Year Figure 4.3. The plots of the original and predicted values (FR-4) The main application for FR-4 high Tg is server for communication. RGM shows good performance with MAPE 7.9%, Bass by 10.0% and GM (1,1) is the worse with MAPE 21.1% (See Table 4.6 & Figure 4.4). 28

86 Table 4.6. FR-4 high Tg CCL market analysis by different forecasting models Year Actual value GM(1,1) RGM Bass historical period MAPE 5.7% 1.3% 5.1% holdout period MAPE 21.1% 7.9% 10.0% Volume (MM m 2 ) FR-4 High Tg Actual value GM(1,1) RGM Bass Year Figure 4.4.The plots of the original and predicted values (FR-4 high Tg) Due to changes in environmental regulations, FR-4 halogen free products started being sold in the marketing place in year All global cellular phone producers such as Nokia, Motorola, Son-Ericson, Apple, LG, and Samsung have already shifted to 100% halogen free. This is why demand for halogen free CCL increased by 40% in 2007! Bass model show excellent performance for holdout period by 0.04% in 2008 and GM (1,1) is 7.32% then 12.92% for RGM (See Table 4.7 & Figure 4.5). 29

87 Table 4.7. FR-4 Halogen Free CCL market analysis by different forecasting model Year Actual value GM(1,1) RGM Bass historical period MAPE 9.10% 8.11% 14.46% holdout period MAPE 7.32% 12.92% 0.04% Volume (MM m 2 ) 40 FR-4 Halogen Free Actual value GM(1,1) RGM Bass Year Figure 4.5. The plots of the original and predicted values (FR-4 Halogen Free) The main applications for specialty CCL products are electronics used in defense, medical system, high speed communication boards and IC substrate. This market size is limited but the value is higher. Bass model show good performance at 0.4% for holdout period in 2008 and 22.0% for GM (1,1) and 22.1% for RGM (See Table 4.8. & Figure 4.6). 30

88 Table 4.8. Specialty CCL market analysis by different forecasting models Specialty (High Speed & Low loss, BT,BT-Equivalent and FR-5) Actual value GM(1,1) RGM Bass historical period MAPE 0.44% 0.22% 4.2% holdout period MAPE 1.62% 7.82% 0.1% Figure 4.6.The plots of the original and predicted values (Specialty) The global CCL market continues to grow from 2001 to 2007 then maintained in The applications of CCL cover all electronic application for consumer, computer, communication, car and are the basic material for PCB. The Bass model shows good performance for holdout period MAPE of 0.1%, 7.82% for RGM and 1.62% for GM (1,1) (See Table 4.9. & Figure 4.7). 31