半導體產業是台灣經濟不可或缺的命脈之一，也是讓台灣在全球的科技業屹立不搖的關鍵。半導體業的產品其中一個是動態隨機存取記憶體(Dynamic Random Access Memory，DRAM)，DRAM價格變化非常大，也因如此有起起伏伏的價格會造成半導體相關廠商成本花費非常多，那對公司財務狀況是一沉重的壓力。
本研究主要目的，希望建立量化基礎的預測模型，有別於之前業界的經驗法則。並將量化模型運用在公司採購DRAM的決策上。以降低購買成本，減少庫存壓力，並更進一步的優化公司財務狀況。
實證結果發現，DRAM現貨價格具有時間序列的非平穩特性，並對其價格變數建立ARIMA模型。除此之外，在相關性分析中，海力士DSI、美光DSI、歐洲PMI和美國PMI均與DRAM報價有高度相關。並將這4個變數加入原本的ARIMA模型，建構解釋力更強的ARIMAX模型來預測DRAM價格。發現DRAM價格用ARIMAX具有良好的解釋和預測力，加上搭配Rolling預測的方式，最後預測值跟實際值的方向幾乎一致。後續的情境分析上，發現運用此預測模型進行採購策略，發現有效的降低成本。

Semiconductor industry is the most important industry of Taiwan’s economics, and it is also the industry that lets Taiwan become a steady role in the whole technology industry in the worldwide. Dynamic Random Access Memory (DRAM) is one of product from semiconductor industry. The price of DRAM is fluctuation usually. Because the fluctuated price causes the heavy purchasing cost of semiconductor firms, the heavy cost is a huge pressure to firm’s financial status.
The aim of this research is to establish a forecasting model based on a quantitative analysis, and this model is different the rule of thumb that firms most use in the past. Using this forecasting model on the decision of purchasing DRAM can reduce the purchasing cost, decrease the inventory pressure, and improve firms’ financial status.
In the empirical results, the spot price of DRAM has non-stationary feature, and we establish the Autoregressive Integrated Moving Average (ARIMA) model of spot price of DRAM. In the correlation analysis, the Days sales of inventory (DSI) of SK Hynix and Micron Company, European Purchasing Managers' Index (PMI), and America PMI have strong correlation with the price of DRAM. We add these four explanatory variables to ARIMA model to form a ARIMAX model which has more power of explanation and prediction. Moreover, we also use rolling forecasting method to predict the price of DRAM. We find that the trend of estimated DRAM price is almost the same as the real trend of DRAM price. Finally, in the situations analysis, we use these forecasting results to construct the purchasing strategy, and we find that this forecasting model can reduce the purchasing cost of DRAM significantly.

中文文獻
巫木誠，2000。我國 DRAM 產業製造策略，行政院國家科學委員會補助專題研究計畫成果報告。
陳雨農、王淑芬、承立平，2015。台灣 DRAM 產業與景氣循環之關聯性分析，國立交通大學管理學院財務金融學程碩士論文。
彭茂榮，2022。2021年第四季台灣IC產業回顧與展望，IEK產業情報網，取自：https://ieknet.iek.org.tw/iekrpt/rpt_more.aspx?actiontype=rpt&indu_idno=1&domain=2&rpt_idno=790524196
外文文獻
Bollerslev, T., 1986. Generalized autoregressive conditional heteroskedasticity. Journal of econometrics, 31(3), 307-327.
Box, G. E., Jenkins, G. M., Reinsel, G. C., and Ljung, G. M., 2015. Time series analysis: forecasting and control. John Wiley and Sons.
Chen, T., 2012. A hybrid fuzzy and neural approach with virtual experts and partial consensus for DRAM price forecasting. International Journal of Innovative Computing, Information and Control, 8(1), 583-597.
Chen, T. C. T., and Wu, H. C., 2020. Forecasting the unit cost of a DRAM product using a layered partial-consensus fuzzy collaborative forecasting approach. Complex and intelligent systems, 6(3), 479-492.
Chen, T., and Chiu, M. C., 2021. An interval fuzzy number-based fuzzy collaborative forecasting approach for DRAM yield forecasting. Complex and intelligent systems, 7(1), 111-122.
Dinh, D. V., 2020. Forecasting domestic credit growth based on ARIMA model: Evidence from Vietnam and China. Management Science Letters, 10(5), 1001-1010.
EE Times, 2018. The Trouble with SRAM, retrieved from
https://www.eetimes.com/the-trouble-with-sram/
Gordy, M. B., and McNeil, A. J., 2020. Spectral backtests of forecast distributions with application to risk management. Journal of Banking and Finance, 116, 105817.
IC Insight, 2022a. Semiconductor Growth Still Seen at 11% Despite 2022 Headwinds, retrieved from
https://www.icinsights.com/news/bulletins/Semiconductor-Growth-Still-Seen-At-11-Despite-2022-Headwinds/
IC Insight, 2022b. Top Three Suppliers Held 94% of 2021 DRAM Marketshare, retrieved from
https://www.icinsights.com/news/bulletins/Top-Three-Suppliers-Held-94-Of-2021-DRAM-Marketshare/
Kawakita, S., and Takahashi, H., 2022. Time‐series analysis of population dynamics of the common cutworm, Spodoptera litura (Lepidoptera: Noctuidae), using an ARIMAX model. Pest Management Science.
Lepselter, M. P., and Sze, S. M., 1985. DRAM pricing trends—the π rule. IEEE Circuits and Devices Magazine, 1(1), 53-54.
Mahmudimanesh, M., Mirzaee, M., Dehghan, A., and Bahrampour, A., 2022. Forecasts of cardiac and respiratory mortality in Tehran, Iran, using ARIMAX and CNN-LSTM models. Environmental Science and Pollution Research, 1-11.
Namin, S. S., and Namin, A. S., 2018. Forecasting economic and financial time series: Arima vs. LSTM. Machine Learning, 1-19.
Simms, L. E., Engebretson, M. J., and Reeves, G. D., 2022. Removing diurnal signals and longer term trends from electron flux and ULF correlations: a comparison of spectral subtraction, simple differencing, and ARIMAX models. Journal of Geophysical Research: Space Physics, 127(2), e2021JA030021.
SK Hynix, 2022. SK hynix to Supply Industry’s First HBM3 DRAM to NVIDIA, retrieved from
https://news.skhynix.com/sk-hynix-to-supply-industrys-first-hbm3-dram-to-nvidia/
Tarui, Y., and Tarui, T., 1991. New DRAM pricing trends: The Bi rule. IEEE Circuits and Devices Magazine, 7(2), 44-45.
TechPowerUP, 2022. Micron Moving to EUV Lithography in Taiwan, retrieved from https://www.techpowerup.com/295419/micron-moving-to-euv-lithography-in-taiwan
TrendForce, 2022. World’s Top Ten IC Design Company Revenue Reached US$39.43 billion in 1Q22, Marvell Growth Rate Tops List, Says TrendForce, retrieved from https://www.trendforce.com/presscenter/news/20220609-11245.html
Yao, H., Huang, J., Li, Y., and Humphrey, J. E., 2021. A general approach to smooth and convex portfolio optimization using lower partial moments. Journal of Banking and Finance, 129, 106167.