再探臺灣肉類與漁產品需求之模型選擇：NEP模型之應用

Model Selection of Demand for Meats and Fish in Taiwan Revisited: A Nested PIGLOG Model Approach

模型選擇是探討需求分析時最重要的課題之一。本文利用臺灣1962–2002年肉類及漁產品之年資料來估計NEP (nested PIGLOG)需求模型，並探討臺灣豬肉、牛肉、雞肉和漁產品之需求行為，特別是模型選擇的問題。在考量結構性轉變、時間趨勢及CUUS (closure under unit scaling)等因素，本研究之實證結果顯示：從樣本內模型選擇得知，NEP與GAITL (generalized almost ideal translog)模型為優良的模型。根據樣本外預測能力而言，若使用trace來進行排序，則以NEP模型的預測能力最佳；若使用行列式值來評斷，則以GFLES (globally flexible linear expenditure system)模型最好，NEP模型次之。綜合樣本內外之模型評斷條件，在分析臺灣肉類與漁產品需求時，NEP模型是一個相當好的選擇。另外，從NEP模型的彈性值顯示：自身價格彈性全為負值，符合需求法則。交叉價格彈性顯示多數產品間為替代關係，而正的支出彈性顯示四種產品無一劣等財。甚者，雞肉與牛肉之支出較具彈性，顯示臺灣民眾在增加肉類與漁產品的支出時，雞肉與牛肉消費量所增加之比例相對於豬肉與漁產品而言較多，對於畜牧業之生產者與進出口業者應可為其決策之參考。

Choice of functional forms is one of the difficult issues in demand analysis. In this study, we estimate the nested PIGLOG (NEP) model using the annual disappearance data of meat and aquatic products in Taiwan from 1962 to 2002. Four food items are considered in this study including fish, chicken, pork and beef. Incorporating factors such as structural changes, time trend and the property of“closure under unit scaling,”our empirical results show that both the NEP and GAITL (generalized almost ideal translog) are preferred models according to the likelihood ratio test. Moreover, from the results of out-of-sample forecasting, the NEP model is the best since it is ranked number one and two in terms of trace and determinant, respectively. Therefore, based on in-sample nested tests and out-of-sample forecasting, the NEP is the best choice of functional forms to explain demand for meats and fish in Taiwan. Empirical results also show that the own-price elasticities of meats and fish are all negative, satisfying the law of demand; whereas the negative cross-price elasticities between chicken and beef imply they are complements, but others are substitutes. Expenditure elasticities of beef and chicken are elastic, indicating that as the expenditure on meats and fish increases by one percent in Taiwan, the percentage increase of beef and chicken demands will be more than unity. This may be valuable to chicken farmers and beef importers.