| Summary: | 碩士 === 國立中央大學 === 工業管理研究所碩士在職專班 === 95 === Because the imbalance of supply and demand often leads to the sudden change of the Nand flash prices, members of the semiconductor supply chain hope that they can control the market volatility and the change of price trend. However, they merely perform some simple analyses according to the experiences of experts in the industry or with their own subjective judgment in order to find out the price trend. They often determine their purchase and inventory policy based on these simple analyses. However, such analyses usually lack of scientific basis and decisions based on which could result in error and expose a company asset to risks.
The purpose of this study is to develop a simple and effective pattern of price forecast, and to help the people in this industry make an accurate forecast of both the price trend of the Nand flash and its delivery amount to final customers under the situation of extremely irregular changing trend, and thus, make a policy to avoid risk, continuously maximize company profit, improve operational efficiency, and reduce the possibility of losses.
In the price forecast analysis, evidence shows that, in 2004 and 2005, when the original figures presented a stable trend change, among all time-series analysis methods, the naïve method, moving average method, exponential smoothing method and grey prediction of GM(1,1) method had good results in the forecast of price change from October to December of the year. The MABE value was lower than 10% which indicates a high forecasting accuracy,and maintain a relatively high level of profit. When the figures aroused irregular changes for the period of October to December 2006, various forecast methods indicated larger differences in the forecast results. Among them, the exponential smoothing method ( =0.3) had best forecast ability,and earning a profit of US$368,500, but the MAPE value still reached to 27.72%. By combining the forecast methods, this study averaged the forecast values of various kinds of forecast methods, and the MAPE value can be reduced to 23.34%.The results of the delivery quantity forecast analysis in this study indicates that, due to the delivery quantity is highly volatile, evidently higher than the change of price trend, the accuracy of various kinds of forecast in the forecast of delivery amount from October to December of the year was far from ideal. All MAPE values were higher than 20%.
According to the foregoing findings in this study, a hazard estimation of a product is formulated after performing hazard analysis and giving fully consideration of the management’s risk-bearing capacity. Through continuous monitoring on highly hazardous parts and mature policy making, it is expected that highest profits can be achieved under the lowest hazardous conditions. In actual applications and future practice, various kinds of predictive models should be established in the policy making data system, and the historic figures in the oracle data system of the company can be utilized for computation and analysis to derive the prediction value. In addition to comparing and analyzing the various prediction results to find out their differences, the daily and weekly trend of price change can be predicted and controlled through the review of the collective team members on the point of every trend change.
|
|---|
