以層級貝氏模型預測新產品銷售模式：以美國電影為例

Abstract

現今市場環境變化迅速，企業使用以往的經營方式已無法因應瞬息萬變的動態市場，需隨時調整公司資源以強化其應變能力。開發新產品前，企業首要注重的是新產品的需求規模（市場潛力）以及其擴散速度是否快速，故需制定良好的銷售預測模式，為每項產品找到最適切的模型，將會有效降低公司的投資風險以及協助行銷策略的擬定。 科技進步促使人們改變生活習慣，在生活步調匆忙與高強度的工作環境下，能夠令人放鬆的休閒活動變得不可或缺。觀賞電影對於現代人而言是一種快速、也最有效的都市休閒活動。本研究將使用美國電影作為實證分析的對象，透過創新產品銷售預測模式預測出每部電影的銷售量，期望能提供業界預測新產品銷量之參考。研究目的為提出影響美國電影銷售擴散型態之因素，以此建構出不同的電影銷售預測模型，進而比較不同電影銷售預測模型之預測效度。 近年來Bass擴散模式不斷地發展，配合各種高效度的統計分析方法，若再加入足以影響銷售的各種重要變數，預測模式將更趨於精準穩定。因此本研究以過去文獻所得之最優良模式為基準，加入層級貝氏模式進行分析，在此兩套模式預測體系下，利用創新產品之屬性（製作預算、美國MPAA電影分級、電影類型、前三大主演是否為票房明星、票房明星分數、專家評分、上映前留言數、有無假日效應）來預測銷售量，比較優劣後，找出最佳之預測模式。由最終研究結論可知新產品銷售量之最佳預測模式為使用層級貝氏所建構的模型。

Nowadays, the market environment is changing rapidly. Enterprises using traditional operation methods could no longer adapt to the dynamic market, so they need to adjust the resources and strengthen their ability of problem solving. Before launching new products, companies should focus on the scale of demand (market potential) and whether it is spreading rapidly. Therefore, it is necessary to develop an accurate sales forecasting model to find the most consistent product cycle for each product, which will effectively reduce the risk of the company investment and assist in the formulation of marketing strategies. As the advance in technology makes people change their lifestyles, and with the fast-past life and stressful work environment, leisure activities become indispensable. Among different kinds of leisure activities, seeing movies is the fastest and the most effective way for modern people. This study will use American films as the object of empirical analysis, predicting the sales volume of each film through the innovative product sales forecasting model, and expect to provide a reference for the movie industry to predict the sales of new products. The purpose of this study is to propose factors that influence the diffusion pattern of American film sales, so as to construct different sales forecasting models, and then compare the predictive validity of different models. In recent years, the Bass diffusion model has been continuously developed by using different kinds of effective statistical methods. If we add various important variables that affect the prediction model, the model will become more precise and stable. Therefore, this study is based on the best model which is derived from the previous literature, and adds the hierarchical Bayesian model to do the analysis. Under these two model prediction systems, the attributes of the innovative products (production budget, MPAA rating, film type, main cast, star score, expert rating, number of comments before release, holiday effect) are considered in the models. After comparing the advantages and disadvantages, we can find the best prediction model. According to the research conclusion, the best prediction model for the sales of innovative products is the hierarchical Bayesian model.