中國智慧型手機電源管理晶片需求量之預測

Abstract

智慧型手機自2007年iphone 問市以來，開始了每年二位數字的成長率，直到2016成長力道趨緩，全球智慧型手機從成長期走入了成熟期。然而，從市場專業機構分析中指出，中國大陸近年來智慧型手機出貨量佔全球總出貨量的30% ~ 32%，且仍在繼成長中，是一個相當具規模的經濟體，也是各公司各產業重點經營的市場。 準確的預測對於企業分配資源和計劃，庫存管理，產品開發藍圖以及跨部門的高效率溝通是至關重要的。要能正確地預測變化快速的中國廣大智慧型手機市場是複雜且困難的，然而，對於手機相關產業和企業而言，中國智慧型手機出貨量預測是很重要的。 本研究是以中國大陸市場智慧型手機為例來建構電源管理晶片需求量之預測模型，以2011 ~ 2016大陸智慧手機出貨量為樣本資料，先運用4個時間序列模型方法來做預測並參照預測誤差指標找出最佳績效模型。接下來再參照智慧型手機BOM表(Board of Material)資訊來進行電源管理晶片的需求轉換。 這4個時間序列法分別為，指數平滑分析法(Exponential Smoothing Method)、模型分解法(Decomposition Method)、多元廻歸分析法(Multi-Regression Method)、和自我廻歸差分移動平均法(ARIMA)、季節性自我廻歸差分移動平均法(SARIMA)。研究結果顯示，SARIMA(p,d,q)(P,D,Q)s，SARIMA(2,1,0)(2,1,0)4 模型MAPE 5.65% 預測準度最佳。參照手機BOM做需求轉換後，電源管理晶片2016年度需求量預測值和實際值，誤差值約為4.1%，說明本研究建構之模型是具有準度的模型。

The worldwide smartphone demand enjoyed double digit growth from 2007 until 2016, but is now leveling off and entering the mature phase of its lifecycle. According to leading market intelligence analysis, China smartphone shipments account for a growing portion of the entire market and are responsible for 30% – 32% of total global shipments today. Accurate forecasts are critical for companies to enable correct resource & allocation plans, inventory management, product roadmaps, and efficient communications across divisions. Correctly forecasting the China smartphone market is complicated by the dynamism of the China segment within the wider smartphone market. It is critical for many industries and companies related to smartphone business, however, that the China smartphone shipment volume be accurately forecast. This thesis explores Power Management IC (PMIC) forecast models for the China smartphone industry. We select the 2011 to 2016 China smartphone shipment volumes for our study, utilizing the 2011 – 2015 data to construct four time series forecast models for 2016. The models are then compared to the actual 2016 data to evaluate model efficacy. Smartphone bill of materials (BOM) information is then used to convert the shipment volume information into PMIC demand. The four time series models explored are the exponential smoothing, decomposition, multi-regression, and SARIMA (seasonal autoregressive integrated moving average) methods. The SARIMA(2,1,0)(2,1,0)4 resulted in significantly lower forecast errors in PMIC value. It achieved this by accounting for the seasonal variability in volume that was inadequately captured by the other models, but which is a significant effect in mobile markets.