以類神經網路建構白銀價格預測模式

Constructing Silver Price Forecasting Model with Neural Networks

陳美惠(Mei-Hui Chen) ； 許雅惠(Ya-Hui Shi) ； 蔡坤穆(Kune-Muh Tsai)

《嘉南學報（科技類）》 35期 (2009/12) Pp. 238-252

https://doi.org/10.29540/CNABST.200912.0023

摘要

貴金屬材料中的白銀由於具備絕佳且多功能的材料特性，使其在製造產業上的需求大增。而此也導致白銀在投機性高的投資市場中，成爲熱門的操作產物，使得白銀的價格大幅上升，造成產業需求者沈重的採購成本壓力。因此，如果能在考量投資需求的因素下，嶺展出能夠準確預測白銀價格的模型，將可提供白銀採購者決策時的參考。本研究應用倒傳遞類神經網路田(Back-Propagation Neural Networks, BPNN)，來預測每日白銀的價格，以協助產業需求者，能夠將預測所得的價格實際應用於每日的採購工作中。模型的輸入變數包括直接性、間接性及投資需求因素，在經由不同網路架構及學習次數與學習率的調整與測試後，本研究提出一個較佳的BPNN模型來進行後續的相關實驗。結果顯示，本研究所建構的倒傳遞類神經網路模型，在連續30日滾動的驗證期間，其平均預測的準確率達到98.2%以上，證明該網路模型可有效幫助採購者來預測白銀的價格。

關鍵字

倒傳遞類神經網路 ； 預測模型 ； 白銀價格 ； 投資性需求因素

並列摘要

Among the precious materials, silver has some unique and superior characteristics that cause the demand for it in industry to boost in recent years. Because silver is in a highly speculative market of investment, its price goes up and variates substantially at the same time. As a result, the industry is now confronting with high cost pressure for purchasing silver; therefore, how to accurately predict the prices of silver becomes a critical challenge for companies in demand for silver. This study utilizes back-propagation artificial neural networks (BPNN) to predict daily silver price so that practitioners can apply the predicted prices to their purchases. We employ direct-related, indirect related and investment factors as the inputs to the BPNN. After performing different combinations of network structure and learning rate, a reasonably good BPNN model is proposed for the experiment. To test the accuracy of predicted silver prices, the model was executed for a period of consecutive 30 days and an average prediction accuracy of 98.2% was obtained. The results demonstrate the effectiveness of our BPNN model in predicting silver prices.

並列關鍵字

Back-Propagation Network ； Forecasting Model ； Silver Price ； Investment Factor