預測資訊分享與半導體零件通路商績效 - 政大學術集成

全文

(1)國立政治大學資訊管理學系碩士學位論文. 預測資訊分享與半導體零件通路商績效 The Role of Forecast Sharing in治 Semiconductor Distributor. 立. 政. Performance. 大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. 指導教授: 張欣綠博士、周彥君博士. 研究生：楊惠晴撰. 中華民國 107 年 9 月. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(2) ABSTRACT We cooperate with a leading semiconductor distributor, and find out they make replenishment decisions based on past orders instead of forecasts from their customers. We therefore study the role of forecast sharing in semiconductor distributor, and examine if forecast information is relevant to inventory and sale management of semiconductor distributors. We also examine how external and internal complexity which are specified as forecast fluctuation and hubs’ order diversification moderate the relationships among forecasts, inventory and sales. We collect orders and forecasts of. 政治大 company from 2016/10 to 立 2017/10. The data shows a hierarchical relationship: orders. one main customer of W company, and the associated inventory records prepared by W. ‧ 國. 學. of a component belonged to a storage hub. We discover that forecast information is relevant to the semiconductor distributor’s inventories and sales. Our findings also. ‧. show that forecast signal decreases by forecast fluctuation and hubs’ order. n. al. er. io. sit. y. Nat. diversification which give semiconductor distributors some management implications.. i n U. v. Keyword: forecast sharing, supply chain management, semiconductor industry. Ch. engchi. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(3) Table of Content Table of Content ............................................................................................................ i List of Figures ...............................................................................................................ii List of Tables............................................................................................................... iii CHAPTER 1: INTRODUCTION............................................................................... 1 CHAPTER 2: LITERATURE REVIEW ................................................................... 5 2.1 Forecast sharing between manufacturers and retailers ............................................ 5. 政治大 CHAPTER 3: RESEARCH FRAMEWORK.......................................................... 11 立. 2.2 The credibility of forecast sharing ........................................................................... 7. 3.1 Forecast sharing, inventory and sales. ................................................................... 13. ‧ 國. 學. 3.2 The moderating effect of forecast fluctuation on forecast sharing. ....................... 14. ‧. 3.3 The moderating effect of order diversification on forecast sharing. ...................... 16. sit. y. Nat. CHAPTER 4: DATA AND MEASUREMENT ........................................................ 17. io. er. 4.1 Company Backgroud .............................................................................................17 4.2 Data ........................................................................................................................ 18. al. n. v i n CRESULTS........................................................... CHAPTER 5: ANALYSIS AND ..24 hengchi U. 5.1 Models.................................................................................................................... 24 5.2 Estimation Results ................................................................................................. 26 CHAPTER 6: CONCLUSION.................................................................................. 35 REFERENCE ............................................................................................................. 38. i. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(4) List of Figures Figure 3-1. Research Model ......................................................................................... 13 Figure 4-2.1 The integration of forecast and sales data set .......................................... 20 Figure 4-2.2 The integration of forecast, sales, and inventory at the month level ....... 20 Figure 5-1 Illustration of the three-level variance-components model. ....................... 25 Figure 5-2.1 The total effect of forecasts on inventory................................................ 29 Figure 5-2.2 The total effect of forecasts on sales ....................................................... 33. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. ii. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(5) List of Tables Table 4-2.1 Description of variables ............................................................................ 22 Table 4-2.2 Descriptive statistics of variables ............................................................. 23 Table 4-2.3 Correlation of variables ............................................................................ 23 Table 5-2.1 Estimated results for forecast in models of Inventory. ............................. 30 Table 5-2.2 Estimated results for forecast in models of sales. ..................................... 34. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. iii. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(6) CHAPTER 1: INTRODUCTION Sharing demand forecast information has been recognized as a key element in supply chain coordination (Cachon 2001). Forecasts are critical for decision making and planning processes in the supply chain, so companies have engaged in Collaborative Planning, Forecasting, and Replenishment (CPFR) to facilitate sharing of demand forecasts among the supply chain members. It is also commonly believed that forecasts sharing within a supply chain improves the forecast accuracy and results. 政治大 to better inventory management (Yue, & Liu, 2006). Much literature has suggested that 立. in higher profitability (Yue, & Liu, 2006). In addition, better forecasting can contribute. information sharing is beneficial in lowering supply chain costs (Aviv and Federgruen,. ‧ 國. 學. 1998; Chen ,1998; Aviv, 2001). Aviv (2001) has suggested that each member of supply. ‧. chain incorporates the forecast updates into the replenishment process, so updated. sit. y. Nat. information is relevant to future demand (Aviv, 2001). Also, it is found out that sharing. io. er. forecasts of items with nonstationary demand rates has significantly improved the outlet fill rate and effectively reduced costs resulted from stockout (Angulo, Nachtmann &. al. n. v i n Waller, 2004). Downstream dataC sharing between manufacturer and distributor is found hengchi U to benefit distributor in reduced inventory and stock-outs (Dong, Dresner & Yao, 2014). Demand forecasts from multiple retailers are somewhat informative for the central planner to make aggregate inventory decision (Spiliotopoulou, Donohue & Gürbüz, 2016). In much research and practice, forecast sharing has been considered beneficial for supply chain performance. In this study, we cooperate with W company, a semiconductor distributor which manages sourcing of electronic components from manufacturers such as Intel and TSMC, and offers a comprehensive range of components to fulfill diverse demand for customers which are manufacturers of consumer electronics, such as Quanta Computer 1. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(7) and Foxconn.. W company implements vendor management inventory (VMI), helping. their customers manage inventories of electronic components in customers’ plants. Interestingly, we found out that the company makes replenishment decisions not on forecast information provided by customers, but on the real call off information eight weeks ago. Forecast sharing seemed less valuable in this context. W company encounters two major problems. First, due to suppliers’ lead time so long as twelve weeks, W company needs to place orders to suppliers three months before customers need the components. Although customers do share forecast information three months. 政治大 information of market demands. In W company, forecast information shared five to 立. prior to the demand date, they adjust forecasts along the way as they receive new. thirteen weeks prior to the demand date is referred as soft orders that can reflect. ‧ 國. 學. customers’ intent of purchasing but are not legally binding. Therefore, it is difficult for. ‧. W company to make replenishment decisions based on the frequently-adjusted forecast. sit. y. Nat. information. The other problem is that customers often inflate the forecasts to assure. io. er. sufficient supply. It is common for W company to get inflated forecasts because W company might not be the only supplier of the customers. Customers place soft orders. al. n. v i n C h at the same time toUadverse the risk of insufficient to W company and other distributors engchi. supply of components. Therefore, the demand forecasts may not be as reliable as the actual purchased record for W company to make replenishment decisions. Some studies have examined the issues of forecast sharing and mentioned that forecast sharing might not result in potential benefits for supply chains. For example, Spiliotopoulou et al. (2016) examine the issue of unreliable forecasting in the retailing industry. The authors observes the existence of misreporting forecasts and mistrusting forecast information, and discuss the magnitude and causes between multiple retailers and a central planner in retailing supply chain. In sight of the inconsistency of the role of forecast sharing in the main stream of 2. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(8) the literature and in the case company, we therefore aim for the role of forecast sharing in the semiconductor industry. Our research questions are two folds. First, we aim to examine, under the uncertainty demand of semiconductor industry, whether forecast is trustful and meaningful to distributors’ inventory and sales management (RQ1).. In. addition, we would like to examine what contingency factors may make forecast information more credible or less valuable. We include two contextual factors: forecast fluctuations and order diversification in the study. Forecast fluctuation is one contextual factor reflecting external forces. Because of market uncertainty, it is difficult for. 政治大 demands. Therefore, W company’s customers not only provide forecasts, but also 立 manufacturers of consumer electronics, i.e. W company’s customers, to capture actual. frequently update forecasts. We then are interested in whether effects of forecast signals. ‧ 國. 學. vary by different levels of forecast fluctuations (RQ2). Order diversification is another. ‧. contextual factor related to internal structures of W company’s storage hubs. W. sit. y. Nat. company implements VMI, and helps each customer manage their inventory stored in. io. er. storage hubs which are set near customers’ production plants. Some hubs have orders distributed among multiple items with low volume of each one. Others have orders that. al. n. v i n Cwith are concentrated on certain items volume. According to previous literature, h ehigh ngchi U. manufacturing processes of high customized and lower volume production experience higher levels of complexity than processes that produce high volumes of standardized products (Bozarth, Warsing, Flynn & Flynn,2009). Following the logic, we argue that hubs with different order structure may have different level of complexity in managing inventory and sales. Thus, hubs with various order structures may react to forecast signals differently (RQ3). The following research questions are thus proposed: 1.. Is forecast information relevant and helpful to inventory and sale management?. 2.. How forecast fluctuation plays a role in the relationships among forecast,. inventory and sales? 3. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(9) 3.. What is the impact of hub’s order composition on the relationships among forecast,. inventory and sales? To answer the above research questions, we collect empirical data from W company, the leading semiconductor distributor in Asia. Specially, we obtain orders and forecasts of one main customer of W company, and the associated inventory records prepared by W company. Such data are span from 2016/10 to 2017/10 for one year, and shows a hierarchical relationship: orders of a component belonged to a storage hub. Using the dataset, we expect to empirically discover the role forecast sharing in the. 政治大. semiconductor industry, and identify how contextual factors affect achievable outcomes.. 立. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. 4. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(10) CHAPTER 2: LITERATURE REVIEW 2-1 Forecast sharing between manufacturers and retailers Information sharing is key element in automatic replenishment programs such as Continuous Replenishment Planning (CRP), Efficient Consumer Response (ECR), Quick Response (QR), and Vendor-Managed Inventory (VMI). Information is shared among supply chain members to match supply and demand as closely as possible. Research often suggests that information sharing leads to improved performance in. 政治大 Prajogo and Olhager, 2012). For example, Lee et al. (2000) analyze the benefits of 立. supply chains (Vereecke and Muylle, 2006; Carr and Kaynak, 2007; Hsu et al., 2008;. information sharing using a model of a two-stage supply chain that consists of a retailer. ‧ 國. 學. and a manufacturer. The analysis suggests that demand information sharing can reduce. ‧. inventory and improve cost savings to manufacturer. Other benefits of information. sit. y. Nat. sharing such as improved service levels, reduced lead times and increased inventory. io. er. turns are also examined. (Metters 1997; Lee and Whang 1998; Frohlich and Westbrook 2001; Stank, Keller, and Daugherty 2001; Angulo, Nachtmann & Waller, 2004; Småros,. al. n. v i n 2007). The information includesC inventory levels and position, h e n g c h i U sales data and forecasts,. order status, production etc. Sales and forecast information sharing in VMI environment was also examined (Angulo, Nachtmann & Waller, 2004). The research focused on examine the effect of information delay, and accuracy on measures, including inventory levels and fill rates demand respectively.. a variety of performance. for stationary and nonstationary. Most research on information sharing has focused solely on. inventory and replenishment related savings (i.e. Bourland et al., 1996; Gavirneni et al., 1999; Lee et al., 2000; Cachon and Fisher, 2000). Ali et al. (2011) have investigated the relationship between inventory savings and improvement in forecasting accuracy under 5. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(11) an ARIMA demand process. The analysis includes two information sharing strategies: Forecast Information Sharing (FIS) and No Information Sharing (NIS). Some literature examined forecast sharing between manufacturers and retailers (Mishra et al., 2009; Yue and Liu, 2006; Li & Zhang, 2015; Jiang, Tian, Xu, & Zhang, 2016). Yue and Liu (2006) assessed the benefits of sharing demand forecast information in a manufacturer–retailer supply chain consisting of a traditional retail channel and a direct channel. The results show that information sharing improves manufacturers’ profits, but decreases the retailer’s profits. They also found out that the direct channel. 政治大 channel is beneficial for the manufacturer and the whole supply chain. 立. had a negative impact on the retailer’s performance, and, under some conditions, direct Li and Zhang. (2015) study the retailer’s incentive of sharing demand information to make-to-stock. ‧ 國. 學. manufacturer and how the retailer’s information sharing strategy interacts with the. ‧. manufacturer’s marketing decision (wholesale price) and operational decision (stock. sit. y. Nat. level). Jiang, et al. (2016) investigate the firms’ preferences regarding three information. io. er. sharing formats: no information sharing, voluntary information sharing and mandatory information sharing in a distribution channel where manufacturers possess better. al. n. v i n C h They also investigate retailers. the engchi U. demand information than. impacts of forecast. accuracy on firms’ profits under three sharing formats, and risk preference is included as well. In the literature above, production strategies, make-to-order and make-to-stock scenarios, are taken into consideration when investigating demand information sharing. Li and Zhang (2015) highlight the distinction between a supply chain with a make-tostock manufacturer and that with a make-to-order manufacturer in terms of information sharing strategies and firms’ profitability. They find out that information sharing never benefits the supply chain when the manufacturer is make-to-order but may benefit the supply chain when the manufacturer is make-to-stock. Yue and Liu (2006) compare the value of information sharing in the make-to-order and make-to-stock scenarios, and 6. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(12) results show that both the manufacturer and the retailer can benefit from information sharing in the make-to-stock scenario. Other literature examined forecast sharing in different context. Spiliotopoulou et al. (2016) examined whether demand forecast sharing between retailers and a benevolent central planner is reliable both in theory and in practice where inventory decisions are made at a centralized level relying on demand forecast information passed from regional managers within a supply chain. Terwiesch et al. (2005) empirically studied the relationship of buyer’s forecasting behavior and semiconductor equipment suppliers’ delivery performance in the semiconductor equipment supply chain. Gümüş (2014) studied the impact of forecast sharing on the. 政治大 decisions, profits and costs of channel partners in a supply chain model with a buyer 立 facing a demand risk and two suppliers competing for order allocation from the buyer.. ‧ 國. 學. Most research before focused on the impact of forecast sharing on retailers and. ‧. manufacturers. Such as under what conditions could be beneficial for both retailer and. sit. y. Nat. manufacturers (Yue & Liu, 2006) to share demand forecasts. Unlike them, our research. io. er. focuses on examining forecast information sharing between an electronic device manufacturer and a semiconductor distributor. We empirically investigate the impacts. al. n. v i n an C electronic device manufacturer hengchi U. of forecast sharing from. on the distributor’s. performance which is specified as inventory and sales in a semiconductor supply chain.. 2-2 The credibility of forecast sharing There is a growing body of literature on the role of trust and trustworthiness in information sharing and supply chain performance. The credibility of forecast sharing is getting attention in supply chain management (Ozer, Zheng, Chen, 2011; Voigt and Inderfurth, 2012; Ebrahim‐Khanjari, Hopp, & Iravani, 2012; Hyndman, Kraiselburd, Watson, 2013; Inderfurth,Sadrieh, Voigt, 2013; Gümüş, 2014; Spiliotopoulou, Donohue & Gürbüz, 2016; Fu, Dong,. Liu, & Han, 2016). Our research focuses on what factors 7. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(13) can impact the credibility of forecast. According to previous research, supply chain complexity, which is categorized as internal and external complexity, can impact manufacturing plant performance (Bozarth, Warsing & Flynn, 2009). We therefore assume that supply chain complexity of semiconductor distributor can also impact the credibility of forecast in inventory and sales management. Internal complexity of manufacturing plant was defined as complexity resulting from within the plant such as number of products, number of parts and low volume batch productions etc. While external complexity of manufacturing was defined as complexity resulting from. 政治大 number of suppliers and long and unreliable supplier lead times etc. (Bozarth, Warsing 立 connections with downstream and upstream partners such as demand variability,. & Flynn, 2009). In our research, internal complexity is defined as order diversification,. ‧ 國. 學. the order structure in customer’s hubs managed by the distributor. While external. ‧. complexity is defined as forecast fluctuation resulting from customers’ forecast. sit. y. Nat. behavior. Previous literature also examined factors that affect the credibility of forecast.. io. complexity.. er. Some of them are related to external complexity, while the others are regarding internal. al. n. v i n C h complexity factors Literatures regarding to external that affected the credibility engchi U. and supply chain performance are as below. Özer and Wei (2006) are the first to analyze. the credibility of the forecast sharing between a supplier and a manufacturer. They suggested different degree of forecast information asymmetry, which was the external complexity resulting from the connections between the supply chain members, affected the contract type should be adopted to enable credible forecast sharing between the supplier and a manufacturer. The forecast information asymmetry degree was categorized as symmetry, low asymmetry, and high asymmetry. In addition, Gümüş (2014) studied when and how a buyer could credibly share his forecast information with his upstream suppliers, and how it impacted on the intensity of price competition. In 8. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(14) the study, the degree of demand information asymmetry, which was an external complexity, could affect when and how a credible forecast sharing could be sustainable. They found out that the buyer could use a procurement mechanism called request for quotation (RFQ) with quantity restrictions as a credible signal for forecast sharing as long as the degree of demand information asymmetry was not too high. They also found out that under asymmetric information, the equilibrium prices that emerged between competing suppliers may indeed increase if the buyer could not share forecast information credibly with its upstream partners. Oh and Özer (2013) investigated the. 政治大 dynamic environment, where supplier had problem of eliciting credible forecast 立. role of time in forecast information sharing and decision making under uncertainty in a. information from a manufacturer when both firms obtained forecast information over. ‧ 國. 學. time. Time in the study was an external factor that indicated how long the supplier and. ‧. manufacturer interacted with each other. The supplier relied on the demand forecast. sit. y. Nat. information to make capacity investment decisions. However, because the. io. er. manufacturer had superior relationship with the market, she could get forward-looking information. Therefore, firms had asymmetric information that changed over time. The. al. n. v i n C htime for the supplier study examined what was the right to elicit credible information engchi U. and make the capacity investment decision. Terwiesch et al. (2005) showed that supplier might not achieve potential performance improvements from forecast sharing because of buyer’s forecasting behavior: forecast inflation and forecast volatility. In the study, forecast inflation and forecast volatility were two external complexity resulted from buyer’s response to market demands. Also, the two factors affected the way the supplier acted to the forecast orders. The study suggested that if the customer changed the requested delivery more frequently, which was referred to as forecast volatility, the supplier was more likely to delay the order. Similarly, if the customer cancelled forecast orders more frequently in the past, which was referred as forecast inflation, the supplier 9. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(15) was more likely to delay production. Spiliotopoulou et al. (2016) examined the reliability of demand forecast sharing in the context of one central planner and multiple retailers within a company. They found that the forecasts reported by regional managers are somewhat informative, and the planner partially used the forecast information to make and inform the aggregate inventory decision. However, inventory competition and demand uncertainty were two external factors that decreased the planner’s reliance on forecasts. In addition to external complexity factors, research that examined internal. 政治大 (2016) proposed a quantitative method to study the trust relationship between a retailer 立 complexity factors or both internal and external complexity is as followed. Fu et al.. and an agent in the supply chain where retailer procured the optimal quantity of a. ‧ 國. 學. product from a suppler based on the demand forecast information shared by the agent.. ‧. They studied how retailer’s updated trust in the agents influence decisions of the retailer. sit. y. Nat. and the agent and their impacts on the supply chain performance. In the study, social. io. er. characteristics of the agents was found to affect the decisions and supply chain performance. Social characteristics of the agents can be the internal complexity as it’s. al. n. v i n C h Özer et al. (2011)Uexamined the issue of forecast the different kind of nature of agents. engchi. information sharing in a single supplier and manufacturer context. The suppliers somewhat depended on manufacturers’ forecast to make capacity decisions although manufacturers tended to inflate the forecasts. In the research, trust of forecast was affected by two factors, capacity costs and market uncertainty, which are internal complexity and external complexity. For example, when the capacity cost is low, the loss caused by trusting manufacturers’ forecasts is lower, so the suppliers tend to believe the reports from manufacturers. Therefore, manufacturers are less likely to inflate forecasts for sufficient supply, resulting in more effective forecast information sharing and cooperation in a supply chain. Lower market uncertainty can decrease forecast 10. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(16) inflation and increase overall channel efficiency when capacity cost is high. In our research, we also examine the credibility of forecast sharing, and examine external and internal complexity factors that might affect the credibility of forecasts, and, hence impact distributor’s performance. However, unlike the literature mentioned above, we focus on the forecast sharing between semiconductor distributor and its customers instead of between retailers and manufacturers. Our research examines both external and internal complexity factors that affect the credibility of forecasts, which are forecast fluctuation and order diversification.. 政治 FRAMEWORK CHAPTER 3: RESEARCH 大. 立. Information sharing such as inventory level, sales data, sales forecast, and. ‧ 國. 學. production schedule, is beneficial for supply chain (Lee & Whang, 2000). As we know,. ‧. forecasts are considered essential to the supply chains decision making and planning. sit. y. Nat. processes. Therefore, among different types of information shared between supply. io. performance.. er. chain members, we focus on examining the effect of forecasts on distributors. al. n. v i n C h not only salesUbut inventory management is For a semiconductor distributor, engchi. important to its profitability.. Better forecasting not only serves as advanced. notification for future orders and sales but contributes to better inventory management (Yue, & Liu, 2006; Lee & Whang, 2000). Therefore, our research examines how forecast impact sales and inventories in semiconductor distributor. We not only want to confirm forecast sharing is related and essential to distributor’s sales and inventory management, but also aim to find out what factors impact the forecast signal. In previous research (Bozarth, Warsing & Flynn, 2009), they categorized supply chain complexity into internal complexity arising from within the plant and external complexity resulting from connections with downstream and upstream partners, when 11. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(17) examining the impact of supply chain complexity on manufacturing plant performance. In our research, we focus on two moderators: forecast fluctuation and order diversification. The former is external complexity resulted from forecast sharing activities with downstream partners, which are electronic device manufacturers in our research. The latter is internal complexity arising from hubs’ order structures. When we cooperate with a semiconductor distributor, we find out that customers update their forecasts from 13 to 5 weeks prior to actual demand date. Previous literature has shown that customers’ bad forecasting behaviors, categorized as forecast. 政治大 & Cohen, 2005). Therefore, we want to examine how forecast fluctuation, the variance 立 inflation and forecast volatility, make suppliers discount the forecasts (Terwiesch, Ren. inventories and sales in a semiconductor distributor.. 學. ‧ 國. of forecasts provided by customers, moderates the relationships among forecast sharing,. ‧. We add order diversification as a moderator because we find out that the hubs of. sit. y. Nat. the semiconductor distributor have different order structures. Some hubs have orders. io. er. spread across various items with low volume of each one. Some hubs have orders concentrated on few items with high volume. An extreme case is that only one item. al. n. v i n accounts for the total orders of aC hub. Prior literature has h e n g c h i U suggested that manufacturing complexity increase as the number of supported parts or products increases and production volumes are spread across more distinct items (Bozarth, Warsing, Flynn & Flynn,2009). Following the logic, we argue that order diversification, how orders distributed among items, creates complexity to the distributor’s hub management, and hence moderates the relationship between forecast sharing and semiconductor distributors’ performance. The research framework is shown in Figure 3-1.. 12. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(18) Figure 3-1. Research Model. 治 sales. 政 and 3-1 Forecast sharing, inventory 大. 立. Forecast information sharing is essential to the supply chains decision making and. ‧ 國. 學. planning processes and has been recognized as a key element in supply chain coordination (Cachon, 2001). For example, Texas Instruments share their forecasts with. ‧. suppliers as part of their quantity-flexible contracts (Lee & Whang, 2000). And. y. Nat. sit. suppliers use the forecasts to develop their production plan. Forecast sharing is shown. n. al. er. io. beneficial for supply chain members. For example, a 5%-20% reduction in inventory. i n U. v. costs and an 2%-12% increase in off-the-shelf availability have been reported by. Ch. engchi. GlobalNetXchange, a consortium consisting of more than 30 trade partners including Sears, Kroger, Unilever, Procter & Gamble, and Kimberly Clark, after engaging in CPFR program (VICS CPFR Committee,2002; Terwiesch, Ren & Cohen, 2005). Because forecast sharing contributes to supply chain performance, we assume that forecast sharing is relevant to inventories and sales, the performance of a semiconductor distributor.. In addition, prior literature suggested that demand forecast updating. induced the supplier to produce more or less than the manufacturer’s uncommitted quantity request (Angulo, Nachtmann & Waller, 2004), so we assume that demand forecast also induces a semiconductor distributor to replenish more or less than the 13. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(19) customers’ uncommitted quantity request, which is related to inventory management. Also, forecasts are informative for the distributor to make replenishment decisions. We argue that forecast sharing is a signal for inventory replenishment and might somewhat reflect the volume of inventories. We argue that when the total forecast demands shared by customers are higher, the total inventories are intuitively higher since the distributor must assure sufficient supply. Also, because forecast is shared by customers and serves a notification of future orders, we argue that forecast sharing is relevant to sales for a semiconductor distributor. And as the demand forecast increases, the future sales. 政治大 H1a: Forecast sharing has positive relationship with inventories. 立. accordingly increase. Thus:. H1b: Forecast sharing has positive relationship with sales.. ‧ 國. 學 sit. y. Nat. sharing.. ‧. 3-2 The moderating effect of forecast fluctuation on forecast. n. al. er. io. Although the value of sharing demand forecast within a supply chain has been. i n U. v. investigated, much of the literature assumes truthful information is exchanged (Li 2002,. Ch. engchi. Zhang 2002; Özer and Wei 2006; Oh & Özer, 2013). However, forecasts are not always shared truthfully between supply chain members (Terwiesch, Ren & Cohen, 2005; Ebrahim‐Khanjari, Hopp & Iravani, 2012; Fu, Dong, Liu, & Han, 2016; Spiliotopoulou, Donohue & Gürbüz, 2016). The credibility and trust of forecast sharing is gaining attention in supply chain management (Ozer and Wei, 2006; Ozer, Zheng, Chen, 2011; Inderfurth, Sadrieh, Voigt, 2013; Spiliotopoulou, Donohue & Gürbüz, 2016). To earn more profit, the agent has an incentive to inflate her forecast to the retailer who seeks demand forecast information from the agent before purchasing products (Fu, Dong, Liu, & Han, 2016). Sometimes customers inflate their forecast to guarantee sufficient 14. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(20) products provided from suppliers, and manufacturers inflate her forecast to induce the supplier to build more capacity. Sometimes customers continually revise their forecasts as they receive new information about market demand. The bad forecasting behaviors, referred to as forecast inflation and volatility, prevent optimal supply chain performance (Terwiesch, Ren & Cohen, 2005), and make forecast less informative. It was shown that the suppliers may not trust the manufacturer's forecast when they are aware of forecast bias, in turn harming supply chain performance (Cachon & Lariviere, 2001). Also, Cattani and Hausman (2000) show that demand forecasts do not necessarily. 政治大 reacts to the wrong forecast update. Therefore, suppliers might delay acting on the 立. become more accurate as they are updated, in turn causing inefficiencies if the firm. forecast. Because of bad forecasting behaviors, suppliers may not take forecast. ‧ 國. 學. seriously and discount forecast, in turn decrease the impact of forecast signal. In our. ‧. research, we define forecast fluctuation as the variance of forecasts shared by customers. We argue that forecast. io. er. and update forecasts frequently, and inflates demands.. sit. y. Nat. due to market uncertainty they face. Because of market uncertainty, customers change. fluctuation decreases the forecast signal to inventories because a semiconductor. n. al. Ch. distributor might consider forecasts with higher. engchi. v i n variance U. as unreliable, in turn. discounting the forecasts. Therefore, when forecast fluctuation is higher, the forecasts are less informative for the distributor to make replenishment decisions. In addition, because of high market uncertainty, the actual demand and future sales are difficult to understand and forecast. We therefore argue that forecast fluctuation decreases the impact of forecast sharing on sales for a semiconductor distributor. Thus: H2a: Forecast fluctuation decrease the impact of forecast sharing on inventories. H2b: Forecast fluctuation decrease the impact of forecast sharing on sales.. 15. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(21) 3-3 The moderating effect of order diversification on forecast sharing. Previous literature categorized supply chain complexity into internal complexity from the manufacturing plant itself and external complexity resulted from interactions with supply chain members (Bozarth, Warsing & Flynn, 2009). While customers’ forecast fluctuations resemble external complexity, we explore order diversification of a hub as the internal complexity. Some hubs in our research have orders concentrated on few items with high volume, which is alike high volumes of standardized products. 政治大. in manufacturing literature (Hayes and Wheelwright, 1979; Hill, 1994; Safizadeh et al.,. 立While some hubs in our research has orders spreading. 1996; Duray et al., 2000).. ‧ 國. 學. across multiple distinct items with relatively low volume, which is similar to the concept of producing customized, or very low volume products in manufacturing. ‧. literature.. y. Nat. sit. At the manufacturing planning level, greater numbers of products and parts, and. n. al. er. io. higher level of customization will increase the size and scope of manufacturing. i n U. v. operations. An unstable master production schedule also makes it more difficult for. Ch. engchi. plants to effectively balance demands against capacity and identify feasible production schedules. Also, One-of-a-kind and low volume batch production requires more complex interactions between different areas of the plant and higher levels of decentralized decision making (Hill, 1994). Internal manufacturing complexity was shown to have negative impact on manufacturing performance such as schedule attainment and manufacturing costs (Bozarth, Warsing & Flynn, 2009). Although our research context is in a semiconductor distributor, we assume that internal hub complexity, referred to as order diversification in our research, also has negative impacts on distributor performance:. sales and inventories.. We argue that with. 16. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(22) higher order diversification, indicating that orders in the hub spread across many distinct items with lower volume, distributors’ planning and replenishment tasks are more complex. Such complexity makes it difficult for a distributor to interpret forecasts provided by customers. Therefore, we argue that higher order diversification decreases the impact of forecast signal on inventories. In addition, because orders of customized and very low volume products are with high erratic and discontinuous demand, their actual demands of market are accordingly more difficult to forecast than orders of standardized and large volume products. Therefore, we argue that higher order. 政治大 related to the moderating effect of order diversification are as below: 立. diversification decreases the impact of forecast signal on sales. Thus, the hypotheses. H3a: Order diversification decreases the impact of forecast sharing on inventories.. ‧ 國. 學. H3b: Order diversification decreases the impact of forecast sharing on sales.. ‧. Nat. er. io. sit. y. CHAPTER 4: DATA AND MEASUREMENT 4-1 Company Background a. n. iv l C n hcompany, We have been working with W the largest electronic component e n g cone h iof U. distributors in Asia, to improve their inventory and sales management. W company was founded in 1980 with its headquarters located in Taipei, Taiwan. It has more than 30 branches offices located across Hong Kong, China, Singapore, India, Malaysia, Philippines etc. W company is an intermediate trader between electronic components manufacturers and electronic devices manufacturers. Their electronic component trading partners are more than 60 worldwide including Intel, Texas Instruments, Philips, Hynix, Vishay, and Omni Vision. W company has eight major customers with about 110 plants across different cities in China. W company aggregates orders from small17. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(23) medium manufacturers and gets discounts from electronic component suppliers to help their customers decrease their costs. In addition, W company provides Original Equipment Manufacturers (OEM) and Original Designer Manufacturers (ODM) assistance and considerable number of electronic components to produce their end products. Also, W company helps small companies to control their inventory pooling. Furthermore, W company implements vendor management inventory (VMI), buying electronic components from electronic components suppliers and managing inventory in the plants for their customers. As a distributor, W company receives soft orders from. 政治大 needs. Therefore, we collect inventory, forecast, and sales data from W company to 立. customers, and replenishes items to plants to assure sufficient items for customers’. examine if forecast sharing is essential and related to a distributor’s sales and inventory. ‧ 國. ‧ sit. y. Nat. 4-2 Data. 學. management.. n. al. i n U. company provides the VMI service for the 13 plants.. Ch. engchi. er. io. Our data come from one major customer of W company. For the customer, W. v. We obtain operational data of. the 13 plants from 2016/10 to 2017/10. We have three sets of operational data: sales, inventory, and forecast. The sales data set includes the purchased items and quantities, the plant stored the item, and the transaction date. The inventory data set includes stored items and quantities, the plant stored the item, and the upload date. The forecast data set includes demanded items and quantities, the plant to which the items is distributed, and the demand date and MRP date of the item, where MRP date is the time the customer provides the forecast. To examine our research hypotheses, we integrate the three sets of operational data. Because of the long lead time of semiconductor components, a customer has to indicate 18. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(24) the demand 13 weeks ahead. Such a signal is represented as the forecast information 13 weeks prior to the actual demand date. As market conditions evolve, the customer would continuous to provide forecast 12 weeks prior, 11 weeks prior, and till the frozen window: 5 weeks prior. In other words, an actual order is associated with a series of forecasts from 13 weeks ahead to 5 weeks ahead, the frozen window to which a customer no longer can modify quantities. Thus, we follow the operational logic to match the sales data set with the forecast data set (see Figure 4-2.1). Because inventory is a periodic measure, we cannot locate specific inventories for each order. Instead, we. 政治大 month, and then match with the aggregated monthly demand of the item, which in turn 立. aggregate data at the month level. We calculate the average inventory of an item in a. is associated with aggregated monthly forecasts from 13 weeks to 5 weeks prior. The. ‧ 國. 學. aggregated monthly forecasts are actually average aggregated monthly forecasts. ‧. because the number of forecasts from 5 weeks to 13weeks ahead are different. For. sit. y. Nat. example, in a month the number of forecasts 5 weeks ahead might be four but the. io. er. number of forecasts 6 weeks ahead might be two. Following the process (see Figure 42.2), we combine the three operational data sets, and remove missing value. To reflect. al. n. v i n C observations forecast variations, we eliminated less than four forecast signals. In h e n g c with hi U. addition, we eliminated any item associated with only one observation as these items have no variations in a panel setting. In total, the sample contains 7821 observations for 583 components.. 19. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(25) Figure 4-2.1 The integration of forecast and sales data set. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 4-2.2 The integration of forecast, sales, and inventory at the month level To examine the relationship among forecast sharing, inventory and sales, we operationalize the variables in Table 4-2.1. We measure forecast sharing by calculating the average forecast of aggregated monthly forecasts from 13 weeks to 5 weeks prior. Inventory and sales are measured by the average inventory quantity and sales quantity in the monthly integrated data set mentioned above. To examine forecast fluctuation, 20. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(26) we calculate standard deviation of aggregated monthly forecasts from 13 weeks to 5 weeks prior. In addition to standard deviation, we also measure median absolute of deviation (MAD): median(|𝑋𝑋𝑖𝑖 − 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚(𝑋𝑋)|) and quartile coefficient of dispersion (QCOD):. 𝑄𝑄3−𝑄𝑄1. . To examine order diversification, the degree of orders spread among. 𝑄𝑄3+𝑄𝑄1. items, we use entropy measure: ∑𝑖𝑖[𝑃𝑃𝑃𝑃 × ln(1/𝑃𝑃𝑃𝑃)]. Pi is the proportion of each item accounted for total orders at plant p in month t. The entropy measure was used to. measure product diversification (Jacquemin & Berry, 1979; Palepu, 1985). Higher value indicates higher diversity of orders, which is distributed among more items with. 政治大 the hub in the month is concentrate 立 on one item. When order diversification is 3.697341, less volume of each. For example, when order diversification is 0, the total orders of. ‧ 國. 學. the total orders of the hub in a month is spread across 102 items, and 𝑃𝑃𝑃𝑃 is in the range of 0.0000675 and 0.1102. Table 4-2.2 and Table 4-2.3 provide descriptive statistics and. ‧. correlation for the variables used in the analysis.. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. 21. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(27) Table 4-2.1 Description of variables Research variable. Measurements items. Description. Forecast sharing. Average forecast. The average forecast of each item i from 5 to 13 weeks before demand date in plant p in month t.. Forecast. 1. Standard deviation (STD). The variance of forecast of. fluctuation. 2. Median absolute of. each item i from 5 to 13 week. deviation (MAD):. before demand date in plant p. median(|𝑋𝑋𝑖𝑖 − 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚(𝑋𝑋)|). in month t.. 政治大 . dispersion (QCOD): 立 3. Quartile coefficient of. 𝑄𝑄3−𝑄𝑄1. QCOD sometimes are null. 學. because most of the aggregated monthly forecasts from 13 weeks to 5 weeks prior are. ‧. ‧ 國. 𝑄𝑄3+𝑄𝑄1. zero, in turn resulting in zero. Nat. sit. ∑𝑖𝑖[𝑃𝑃𝑃𝑃 × ln(1/𝑃𝑃𝑃𝑃)], where Pi is. Ch. the proportion of each item. engchi. accounted for total orders at. er. al. How the total orders are. n. diversification. Entropy measure:. io. Order. y. Q1 and Q3.. v. distributed among various. i n U. items.. Larger value means less concentrated order. In other. plant p in month t.. words, total orders are distributed among various items with low volume of each item i in customer plant p in month t. Inventory. Inventory quantity. The inventory of item i managed by plant p in month t.. Sales. Sales quantity. The quantity of item i purchased by the customer from plant p in month t. 22. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(28) Table 4-2.2 Descriptive statistics of variables Variable. Mean. Std. Dev.. Min. Max. ForecastSharing. 32043.87. 112567.4. 0. 2235456. ForecastFluctuation (STD). 10352.74. 32617.71. 0. 635570.3. ForecastFluctuation (MAD). 6654.792. 25446.18. 0. 610175.5. ForecastFluctuation (QCOD). 0.385727. 0.319175. 0. 1. OrderDiversification. 3.100769. .4988452. 0. 3.934536. Inventory. 199792. 630514.8. 50. 1.71E+07. Sales. 73681.71. 0. 5694000. 268122.3 政治大. 立. N= 7821, but only N for ForecastFluctuation (QCOD) is 6950 because of null values resulted from zero Q1 and Q3.. ‧ 國. 學. Table 4-2.3 Correlation of variables (4). (5). (6) (7). n. Ch. y. 1 0.8252 1 0.7695 0.9273 1 -0.1531 -0.0278 -0.0054 1 -0.0342 -0.0251 -0.0298 -0.0696 1 0.6421 0.5125 0.455 -0.1469 -0.0396 1 0.7553 0.5844 0.5236 -0.1367 -0.0353 0.6085. sit. io. al. (3). er. Nat. ForecastSharing (1) ForecastFluctuation (STD) (2) ForecastFluctuation (MAD) (3) ForecastFluctuation (QCOD) (4) OrderDiversification (5) Inventory (6) Sales (7). (2). ‧. (1). engchi. i n U. v. 1. 23. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(29) CHAPTER 5: ANALYSIS AND RESULTS 5-1 Models Unlike conventional two-level data, our panel data is in three-level structure such that the clusters/groups themselves are nested in superclusters forming a hierarchical structure. Specifically, we have repeated observations (level 1) for items (level 2) which are clustered in plants (level 3). It seems reasonable to expect that observations on items within the same plant are correlated, and that observations on the same item are even. 政治大 model with the error structure as 𝜍𝜍 + 𝜍𝜍 + 𝜀𝜀 . 𝜍𝜍 is the random intercept for plant 立. more correlated. To reflect the hierarchical structure, we specify a three-level mixed 𝑝𝑝. 𝑝𝑝𝑝𝑝. 𝑝𝑝𝑝𝑝𝑝𝑝. 𝑝𝑝. ‧ 國. 學. p, 𝜍𝜍𝑝𝑝𝑝𝑝 is the random intercept for item i in plant p, and 𝜀𝜀𝑝𝑝𝑝𝑝𝑝𝑝 is the error term for item. i in plant p in month t. The three error components are assumed to have zero means and. ‧. each has its own variance: 𝜃𝜃,𝜓𝜓 (2) ,𝜓𝜓 (3) . These three error terms are also assumed to be. sit. y. Nat. mutually uncorrelated and thus their variances add up to the total variance. This three-. io. er. level design is displayed in figure 5-1. The figure illustrates error terms for the threelevel variance-components model for plant p. In the first stage, the random intercept 𝜍𝜍𝑝𝑝. al. n. v i n C h with mean 0Uand variance for plant p is drawn from a distribution engchi. 𝜓𝜓 (3) . Therefore,. observations at the plant p have mean β + 𝜍𝜍𝑝𝑝 . In the second stage, the random intercept. for item i = 1, 𝜍𝜍1𝑝𝑝 , is drawn from a distribution with mean 0 and variance 𝜓𝜓 (2) , so the. mean β + 𝜍𝜍𝑝𝑝 + 𝜍𝜍1𝑝𝑝 . The random intercept for item i = 2, 𝜍𝜍2𝑝𝑝 , is also drawn from a. distribution with mean 0 and variance 𝜓𝜓 (2) , so the mean for item 2 in plant p is β + 𝜍𝜍𝑝𝑝 + 𝜍𝜍2𝑝𝑝 . Finally, in the third stage, residuals for observations, 𝜀𝜀11𝑝𝑝 and. 𝜀𝜀21𝑝𝑝 , are. drawn from a distribution with mean and variance 𝜃𝜃, resulting in two observations’. variance: β + 𝜍𝜍𝑝𝑝 + 𝜍𝜍2𝑝𝑝 + 𝜀𝜀11𝑝𝑝 and β + 𝜍𝜍𝑝𝑝 + 𝜍𝜍2𝑝𝑝 + 𝜀𝜀21𝑝𝑝 , which are represented by filled dots. 24. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(30) 政治大 The two equations of our models are as below: 立. Fig 5-1 Illustration of the three-level variance-components model.. ‧ 國. 學. 𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝐼𝑡𝑡𝑡𝑡𝑡𝑡 = 𝛽𝛽0 + 𝛽𝛽1 𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹ℎ𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑡𝑡𝑡𝑡𝑡𝑡 + 𝛽𝛽2 𝐹𝐹𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑡𝑡𝑡𝑡𝑡𝑡 +. 𝛽𝛽3 𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹ℎ𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑡𝑡𝑡𝑡𝑡𝑡 × 𝐹𝐹𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑡𝑡𝑡𝑡𝑡𝑡 + 𝛽𝛽4 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑡𝑡𝑡𝑡 +. (1). sit. y. Nat. 𝜀𝜀𝑡𝑡𝑡𝑡𝑡𝑡. ‧. 𝛽𝛽5 𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹ℎ𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑡𝑡𝑡𝑡𝑡𝑡 × 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑡𝑡𝑡𝑡 + ∑15 𝑡𝑡=2 𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑡𝑡𝑡𝑡𝑡𝑡 + 𝜍𝜍𝑝𝑝 + 𝜍𝜍𝑖𝑖𝑖𝑖 +. n. al. er. io. 𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑆𝑡𝑡𝑡𝑡𝑡𝑡 = 𝛾𝛾0 + 𝛾𝛾1 𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹ℎ𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑡𝑡𝑡𝑡𝑡𝑡 + 𝛾𝛾2 𝐹𝐹𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑡𝑡𝑡𝑡𝑡𝑡 +. i n U. v. 𝛾𝛾3 𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹ℎ𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑡𝑡𝑡𝑡𝑡𝑡 × 𝐹𝐹𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑡𝑡𝑡𝑡𝑡𝑡 + 𝛾𝛾4 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑟𝑟𝐷𝐷𝐷𝐷𝐷𝐷𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑡𝑡𝑡𝑡 +. Ch. engchi. 𝛾𝛾5 𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹ℎ𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑡𝑡𝑡𝑡𝑡𝑡 × 𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑂𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑒𝑡𝑡𝑡𝑡 + ∑15 𝑡𝑡=2 𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑇𝑡𝑡𝑡𝑡𝑡𝑡 + 𝜍𝜍𝑝𝑝 + 𝜍𝜍𝑖𝑖𝑖𝑖 +. 𝜀𝜀𝑡𝑡𝑡𝑡𝑡𝑡. (2). The two equations test how forecast sharing impact inventory and sales. Also, they. test how forecast fluctuation and plant’s order diversification moderate the impact of forecast signal on inventory and sales. In equation (1) and (2), the index t stands for each observation in month t, and index i and p stand for item and plant respectively. The effect of forecast sharing on inventory and sales are denoted by the parameters𝛽𝛽1. 𝛽𝛽2 is coefficient for the effect of forecast fluctuation on inventory and sales.. reflects the moderation effect of forecast fluctuation on inventory and sales. 25. 𝛽𝛽3 𝛽𝛽3. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(31) represents the impact of order diversification and 𝛽𝛽4 represent the moderation effect of it on inventory and sales. We have fifteen time dummies representing months of each. observation, so we have 14 dummies in our models. 𝜍𝜍𝑝𝑝 indicates plant-level random intercept; 𝜍𝜍𝑖𝑖𝑖𝑖 is the item-level random intercept, and. 𝜀𝜀𝑡𝑡𝑡𝑡𝑡𝑡 is the level-1 error term.. 5-2 Estimation Results Table 5-2.1 shows the analysis of forecast sharing, forecast fluctuation (STD) and order diversification on inventory. Model(1) includes forecast sharing, forecast. 政治大 diversification and the interaction 立 term. Model(4) is the full model with time dummies fluctuations and the interaction term. Model(2) and Model(3) additionally include order. ‧ 國. 學. included. We measure forecast fluctuation not only by standard deviation, but also by median absolute of deviation (MAD) in Model(5) and quartile coefficient of dispersion. ‧. (QCOD) in Model(6). The sample size of models is 7821 except for the sample size of. sit. y. Nat. model(6), which is 6950. It is resulted from null values of quartile coefficient of. n. al. er. io. dispersion (QCOD) discussed in data section. We mainly look at our full model of. i n U. v. Model (4). The results of likelihood ratio test show that our full model not only fits. Ch. engchi. significantly better than null model but better than models form Model(1) to Model(3). Also, the pseudo R-squared is 0.726, indicating the high correlation between the model’s predicted values and the actual values. We also compare the fitted mixed model to standard regression with no group-level random effects, and the results show that the variances of random intercepts: plant, item and month are significantly different from zero (chi2(3) = 741.87, Prob > chi2 = 0.0000). Therefore, our three-level full model fits the data better than standard regression with no group-level random effects. The results show that forecast fluctuation, assessed by standard deviation, has a significant and negative moderating effect on the relationship between forecast and 26. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(32) inventory in Model(4) (β3 = -0.0000078, p < 0.01). H2a is therefore supported. To better asses forecast fluctuation, we not only measure it with standard deviation (STD), but with median absolute of deviation (MAD) and quartile coefficient of dispersion (QCOD). Although forecast fluctuation (QCOD) moderator does not have significant moderating effect in Model(6), forecast fluctuation (MAD) moderator has significant and negative coefficient in Model(5) ( β3 = -0.0000079, p < 0.01).. Forecast. fluctuation moderates the impact of forecast on inventory negatively. With greater forecast fluctuation, the impact of forecast on inventory decreases. The distributor. 政治大 inventory than in the condition with lower forecast fluctuation. 立. discounts the forecasts when the variance of the forecast is higher, and prepares less. In addition to forecast fluctuation, we examine the other factor, order. ‧ 國. 學. diversification, that might also impact forecast signal on inventories. In Model(4), the. ‧. results show that order diversification ,how the orders are distributed among items in. sit. y. Nat. the plant, negatively moderates the relationship between forecast and inventory (β5 =. io. er. -1.0005771, p < 0.01). H3a is supported. Because we use entropy measure to assess order diversification, higher value means less concentrated total orders. Higher value. al. n. v i n C h means that the total of order diversification in our models orders are distributed among engchi U. various items with low volume. The estimation shows that plants with high mix low volume orders tend to discount forecast signal than plants with low mix high volume orders. Because of orders with higher product variety and demand fluctuation, it is more difficult for the distributor to analyze and understand the actual demands. Therefore, to avoid overstock, the distributor discounts forecast signal and prepare less inventory when a plant shows higher order diversification (high mix low volume). To better realize the effect of moderators forecast fluctuation (STD) and order diversification on the forecast signal to inventories, we have Figure 5-2.1 shown. Figure 5-2.1 shows the integral effect of forecasts on inventory including the 27. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(33) moderation effects of forecast fluctuation and order diversification. Y-axis indicates the integral effect: β1 + β3 × forecast fluctuations + β5 × order diversification . X-. axis shows the low, medium, and high levels of forecast fluctuations. Order. diversification is indicated by the three lines: blue line for high level, red for medium level, and black line for low level of order diversification. We note that low and high level of forecast fluctuation and order diversification are their mean minus and add their standard deviation. From the figure, it shows that forecast fluctuation has considerably decreasing effect on forecast signal to inventories form low to high level of forecast. 政治大 the two lines overlaps, which indicates that there is barely no different of moderation 立 fluctuation. The graph also shows that from low to median level of order diversification. effects from low to median order diversification. However, from median to high level. ‧ 國. 學. of order diversification, the moderation effect is significant. Therefore, it is shown that. sit. y. Nat. is at high level.. ‧. the decreasing moderation effect of order diversification is mainly significant when it. io. er. After knowing that external and internal complexity, specified as forecast fluctuation and order diversification, decrease the forecast signal to distributor’s. al. n. v i n C hoverall forecast sharing inventories, we discuss how the impacts inventories. The engchi U. results from Table 5-2.1 show that forecast sharing significantly impacts inventory (β1=6.98, p<0.01). However, the integral effect of forecast sharing on inventories should include the moderation effect of forecast fluctuation and order diversification 𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹ℎ𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎(𝛽𝛽1 + 𝛽𝛽3 × 𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓 𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓𝑓 + 𝛽𝛽5 × 𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜𝑜 𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑𝑑).. Figure 5-2.1 shows that both forecast fluctuation and order diversification decrease the effect of forecast sharing on inventories. When both the levels of forecast fluctuation and order diversification are high, the negative moderation effect is the highest. The total effect of forecast sharing on inventories is still positive. Also, the 95% confidence intervals are significantly different from zero at all ranges, indicating significant effect 28. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(34) of forecast sharing. Therefore, H1a is supported. It indicates that greater forecast of an item contributes to higher inventory of the item. The distributor does prepare stock according to the forecast, and the forecast signal is essential to its replenishment.. 立. 政治大. ‧ 國. 學 ‧. Fig 5-2.1 The total effect of forecasts on inventory. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. 29. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(35) Table 5-2.1 Estimated results for forecast in models of Inventory.. β1: ForecastSharing β2: ForecastFluctuation. Model(1). Model(2). Model(3). Model(4). Model(5). Model(6). Inventory. Inventory. Inventory. Inventory. Inventory. Inventory. 4.1808589*** 4.1772251*** 7.0230974*** 6.9806645*** 6.7579616*** 6.2910716*** (0.1078030). (0.1078620). (0.5113509). (0.4882609). (0.4933998). (0.5467589). 3.5634231*** 3.5738484*** 3.9909525*** 4.5436123***. (STD). (0.3786156). (0.3788618). (0.3868703). (0.3746922). β3: ForecastSharing X. -0.0000076*** -0.0000076*** -0.0000077*** -0.0000078***. ForecastFluctuation. (0.0000003). (0.0000003). (STD). (1.847e+04). (1.872e+04). (1.872e+04). (1.916e+04). (2.466e+04). (0.1727671). (0.1650278). (0.1654236). (0.1818606). 4.5804128***. y. sit. (0.4495172). n. al. er. ForecastFluctuation. 2.785e+04. -1.0005771*** -1.0174439*** -1.0167941*** -1.2343211***. io. β3: ForecastSharing X. 1.182e+04. ‧. (MAD). 9.476e+03. Nat. β2: ForecastFluctuation. 5.190e+03. ‧ 國. OrderDiversification. -1.558e+04. 學. β5: ForecastSharing X. (0.0000003). 政治大. 立. β4: OrderDiversification. (0.0000003). (MAD) β2: ForecastFluctuation. Ch. (QCOD). engchi U. v ni. -0.0000079*** (0.0000004) -4.587e+04** (2.149e+04). β3: ForecastSharing X. -0.4497213. ForecastFluctuation. (0.3402744). (QCOD) TimeDum. Not included in models. Included in models but not shown. N. 7821. 7821. 7821. 7821. 7821. 6950. Pseudo 𝑅𝑅2. 0.6924. 0.6925. 0.6918. 0.726. 0.7184. 0.67. Wald chi2. 2462.74. 2458.45. 2375.77. 3338.09. 3028.9. 1839.13. Log likelihood. -112675.04. -112674.69. -112658.5. -112212.6. -112272.54. -100309.77. Standard errors are reported in parentheses. Chi-Square test statistic indicates the rejection of the null hypothesis that all model coefficients are zero. + Indicates statistical significance at 15% level.* Indicates statistical significance at 10% level.** Indicates statistical significance at 5% level.***Indicates statistical significance at 1% level.. 30. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(36) Table 5-2.2 shows the analysis of forecast sharing, forecast fluctuation (STD) and order diversification on sales. Model(1) includes forecast sharing, forecast fluctuations and the interaction term. Model(2) and Model(3) additionally include order diversification and the interaction term. Model(4) is the full model with time dummies included. Like models for inventory, we also measure forecast fluctuation by median absolute of deviation (MAD) in Model(5) and by quartile coefficient of dispersion (QCOD) in Model(6). The sample size of models is 7821 except for the sample size of model(6), which is 6950. It is resulted from null values of quartile coefficient of. 政治大 that our full model not only fits significantly better than null model but better than 立. dispersion (QCOD) discussed in data section. The results of likelihood ratio test show. models form Model(1) to Model(3). Also, the pseudo R-squared is 0.7599, indicating. ‧ 國. 學. the high correlation between the model’s predicted values and the actual values. We. ‧. also compare the fitted mixed model to standard regression with no group-level random. sit. y. Nat. effects, and the results show that the variances of random intercepts: plant, item and. io. er. month are significantly different from zero (chi2(3) = 1973.54, Prob > chi2 = 0.0000). Therefore, our three-level full model fits the data better than standard regression with. n. al. no group-level random effects.. Ch. engchi. i n U. v. The results show that forecast fluctuation negatively moderates the impact of forecast on sales in Model(4) ( γ3 = -0.0000014, p< 0.01).. We assess forecast. fluctuation not only by standard deviation but also by median absolute of deviation (MAD) and quartile coefficient of dispersion (QCOD). In the Model (5), forecast fluctuation (MAD) also has significant and negative coefficient (γ3 = -0.0000015, p<. 0.01). In addition, in the Model (6) forecast fluctuation (QCOD) has significant and negative coefficient (γ3 = -0.3496, p< 0.01). With greater forecast fluctuation, the effect. of forecast signal on sales reduces. H2b is supported. The higher forecast fluctuation might result from customers’ forecasting behavior: forecast volatility and inflation. And 31. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(37) it’s because when customers face substantial uncertainty about the market demand, they revise forecast frequently as they receive updated information. The forecast with higher forecast variance does not represent sales as well as forecast with lower forecast variance. Therefore, the effect of forecast signal on sales is discounted when forecast fluctuation is higher. In addition to forecast fluctuation, we examine how the other factor order diversification moderates the relationship between forecast sharing and sales. In Model (4), the results show that order diversification, how the orders are distributed among. 政治大 = -0.9719, p< 0.01). H3b is supported. Plants with higher value of order 立. items in the plant, negatively moderates the relationship between forecast and sales ( γ5. diversification we measured indicates that the orders of the plants distributed among. ‧ 國. 學. high variety of items with low volume, which is alike high mix low volume in. ‧. manufacturing literature. The estimation shows that the impact of forecast signal on. sit. y. Nat. sales decreases in the plants with high mix low volume orders (higher value of order. io. er. diversification) compared to in the plants with low mix high volume orders (lower value of order diversification). Because the demand of high mix low volume orders is more. n. al. C h on sales reduces.U n i variable, the effect of forecast signal engchi. v. Furthermore, Figure 5-2.2 shows the effect of moderators forecast fluctuation. (STD) and order diversification on the forecast signal to sales. Y-axis indicates the integral effect: γ1 + γ3 × forecast fluctuations + γ5 × order diversification. X-axis shows the low, medium, and high levels of forecast fluctuations. Order diversification is indicated by the three lines: blue line for high level, red for medium level, and black line for low level of order diversification. We note that low and high level of forecast fluctuation and order diversification are their mean minus and add their standard deviation. Figure 5-2.2 shows that forecast fluctuation has marginally decreasing effect on forecast signal. Compared to forecast fluctuation, order diversification has more 32. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(38) moderation effects on forecast signal. From low to median and to high level of order diversification, the decreasing moderation effect is significant. After knowing that external and internal complexity, specified as forecast fluctuation and order diversification, decrease the forecast signal to distributor’s sales, we discuss how forecast sharing impacts sales integrally. The results from Table 5-2.2 show that forecast sharing has significant and positive coefficient in Model (4) (γ1 = 4.311, p< 0.01). However, the integral effect of forecast sharing on sales should include the. moderation. effect. of. forecast. fluctuation. and. order. diversification. 政治大 Figure 5-2.2 shows that both forecast fluctuation and order diversification decrease the 立 𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹𝐹ℎ𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎(𝛾𝛾1 + 𝛾𝛾3 × forecast fluctuations + 𝛾𝛾5 × order diversification).. effect of forecast sharing on sales. When both the levels of forecast fluctuation and. ‧ 國. 學. order diversification are high, the negative moderation effect is the highest. The total. ‧. effect of forecast sharing on sales is still positive. Also, the 95% confidence intervals. sit. y. Nat. are significantly different from zero at all ranges, indicating significant effect of. io. er. forecast sharing on sales. Therefore, H1b is supported. The estimation confirms a positive relationship between forecast and sales. Higher forecast of an item is related to. al. n. v i n higher sales of it. The forecast C signal reflects customer h e n g c h i Udemands, and the sales of the distributor.. Fig 5-2.2 The total effect of forecasts on sales 33. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(39) Table 5-2.2 Estimated results for forecast in models of sales. Model(1). Model(2). Model(3). Model(4). Model(5). Model(6). Sales. Sales. Sales. Sales. Sales. Sales. 1.5691724*** 1.5703122*** 4.2669813*** 4.3114246*** 4.3292612*** 4.1609627***. γ1 : ForecastSharing γ2 : ForecastFluctuation (STD). γ3 : ForecastSharing X. (0.0406915). (0.0407132). (0.1805426). (0.1797232). 0.0658560. 0.0572305. 0.3859605*** 0.6109280***. (0.1344858). (0.1345097). (0.1349883). (0.1371598). -. -. -. -. (0.1799634). (0.1898171). 1.283e+04*. 1.066e+04. 0.0000013*** 0.0000013*** 0.0000012*** 0.0000014***. ForecastFluctuation (STD) (0.0000001). γ4 : OrderDiversification. (6.670e+03). (6.744e+03). (8.336e+03). -. -. -. -. 0.9686029*** 0.9719816*** 1.0099449*** 0.9964973*** (0.0613810). (0.0611389). (0.1618976) -. 0.0000015***. y. (0.0000001). al. n. (QCOD). (0.0639153). 0.9153057***. io. γ2 : ForecastFluctuation. (0.0607848). sit. ForecastFluctuation (MAD). (6.512e+03). Nat. γ3 : ForecastSharing X. 政治大. γ3 :ForecastSharing X. -3.262e+03. er. (MAD). 3.047e+04*** 1.214e+04*. (6.456e+03). 立. (0.0000001). ‧. γ2 : ForecastFluctuation. 9.028e+03. ‧ 國. OrderDiversification. (0.0000001). 學. γ5 : ForecastSharing X. (0.0000001). Ch. engchi. i n U. v. (7.549e+03) 0.3496858***. ForecastFluctuation (QCOD) TimeDum. (0.1188788) Not included in models. N. Included in models but not shown. 7821. 7821. 7821. 7821. 7821. 6950. 0.7654. 0.7648. 0.7523. 0.7599. 0.7559. 0.745. Wald chi2. 2125.05. 2124.78. 2099.72. 2316.69. 2232.71. 1886.1. Log likelihood. -104501.83. -104500.88. -104384.2. -104323.34. -104334.5. -93066.747. Pseudo. 𝑅𝑅. 2. Standard errors are reported in parentheses. Chi-Square test statistic indicates the rejection of the null hypothesis that all model coefficients are zero. + Indicates statistical significance at 15% level.* Indicates statistical significance at 10% level. ** Indicates statistical significance at 5% level.***Indicates statistical significance at 1% level.. 34. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.

(40) CHAPTER6: CONCLUSION Forecast sharing has been considered essential for supply chain improvements. However, when we cooperated with a semiconductor distributor, we surprisingly found out that they do not make replenishment decisions by forecast information shared by their customers. Instead, they use past purchased orders to decide the volume of components to be replenished. We therefore, aim to investigate whether forecast sharing is a signal to semiconductor distributor’s inventories and sales management. Also, we examine if the signal of forecast sharing becomes more or less valuable to the. 政治大. semiconductor distributor’s inventories and sales with external factor referred to as. 立. forecast fluctuation and internal factor referred to as order diversification considered in. ‧ 國. 學. the study. Our results show that forecast sharing is a signal to the semiconductor distributor’s sales and inventories. As we show with respect to Hypothesis 1a and. ‧. Hypothesis 1b, forecast sharing positively related to the distributor’s inventories and. y. Nat. sit. sales. When the forecasts of the electronic item are higher, the sales and the inventories. n. al. er. io. of the item are correspondingly higher. Our results also show that forecast fluctuation. i n U. v. and order diversification both negatively moderate the relationships among forecast. Ch. engchi. sharing, inventories and sales. They both decrease the signal of forecast sharing on the semiconductor distributor’s inventories and sales. Hypothesis 2a and Hypothesis 2b demonstrate that forecast fluctuation, resulted from external complexity such as market uncertainty, discounts the signal of forecast sharing on the distributor’s sales and inventories. When the forecast fluctuation is higher, the positive relationships among forecast sharing, inventories and sales decrease. In addition, Hypothesis 3a and Hypothesis 3b show that order diversification, arising from internal complexity of hubs’ order structure, decreases the signal of forecast sharing on the distributor’s sales and inventories. We further examine moderation effects at different levels of forecast 35. DOI:10.6814/THE.NCCU.MIS.028.2018.A05.