用於預測價格波動之時間序列分析

Abstract

現今，價格波動點預測主要是依靠人為判斷，而容易錯失許多減少成本的機會。對一個企業來說，逢低買進物料和高價售出產品是最直接能夠提高利潤的方式。如果有個方法可以準確的預測出物料或產品價格波動的時間點，企業便能因在對的時間採取對的措施而獲利。因此，本研究針對價格波動時間點的預測提出一個新穎的方法。 本研究提出了價格波動點預測模型（PFPFA）：我們不只預測價格波動程度，還會預測價格波動的時間點。由於交易資料是非均勻採樣的時間序列資料，我們會採用數量來代表時間，來解決這個問題。 價格波動點預測模型共有四個步驟：資料格式轉換、預測價格波動時間點、根據第二步的結果，接著預測價格波動程度、最後是評估評估模型預測的結果，以讓使用者選擇。本研究中，在價格波動時間點預測中提出了四種模型，在價格波動程度預測中提出了三個模型。因此，針對單一產品或物料，將會有十二種的預測結果。 本研究將價格波動點預測模型應用至真實世界的資料庫，並和經常被使用的時間序列分析方法-指數平滑法作比較。在時間點預測方面，價格波動點預測模型的結果是令人接受的；在價格預測方面，價格波動點預測模型也得到比指數平滑法更好的表現。

Nowadays, price fluctuation point forecast is usually relying on the human judgments, and cause many opportunities of saving cost missed. For a company, buying material at a lower price and selling products at a higher price are the straightest way to obtain higher revenue. If there is a way to predict the price fluctuation of material or products accurately, a company can maximize its profit by taking a right action at a right time. This study introduces a novel forecast procedure for price fluctuation points forecast. This study proposes a price fluctuation forecast model: Price Fluctuation Point Forecast Approach (PFPFA). We not only forecast the price change degree, but also the price change time. Since the transaction data are non-uniform sampled time series, we will use quantity to present time to solve this problem. The main process of PFPFA has four phases: (1) transforming data based on the number of fluctuation points; (2) calculating times with different forecast models; (3) calculating prices based on the results of P2 with different forecast models; and (4) evaluating and selecting the best forecast model combination for groups. In this paper, we propose four models for time forecast and three models for price forecast. In consequence, for a single product, there would be twelve different forecast outcomes. we applied PFPFA in a real world case, and compare the result with the Exponential Smoothing (ES) which is commonly and currently used. The time forecast result is acceptable and the price forecast result shows that PFPFA has better performance than ES.