Abstract: | 本次專題研究計畫案,配合國科會鼓勵申請多年期計畫的政策,打算申請三年期研究計畫,俾對研究主題有較為深入的探討,期獲得更有價值的研究成果。實證分析部分,將利用我國經濟部收集的「工廠校正資料」,期限涵蓋民國91 年至94 年,隨執行期間拉長,考慮逐步納入民國96 和97 兩年資料。第一年的研究計畫,引入平滑係數迴歸模型(smooth coefficient regression model) 又稱係數函數模型(functional-coefficient model),利用縱橫資料(panel data)分析生產效率,應屬生產力與效率領域首度嘗試。平滑係數模型是無母數迴歸的一般化模型,避免參數迴歸模型可能存在的模型設定錯誤問題,以及迴歸係數為未知常數的限制,得以更適切描述廠商生產特性。此外,在隨機干擾項為組合誤差的設定下,可以估計廠商技術效率,進一步評估各生產單位的經營績效。為證明本研究估計方法,可以獲得具一致性估計式,將採用Monte Carlo Simulation,以電腦模擬方式,分析各估計式的偏誤量情形。最後,運用工廠校正資料,進行實證分析。第二年將焦點放在分量參數迴歸模型,仍然利用縱橫資料分析生產效率。分量迴歸模型的特色,在於吾人可在不同分位量(quantile) 上,例如0.2、0.4、0.5、0.6、0.8、0.9 等,進行迴歸分析,不再侷限於傳統平均數迴歸分析。理論上,在不同分位量上,投入與產出的函數關係應有不同,研究者能夠更清楚瞭解廠商生產特性。同樣,為驗證本研究估計方法用於縱橫資料分析,估計式仍保有一致性,將採Monte Carlo Simulation 為之; 也會使用「工廠校正資料」,從事實證分析。第三年計畫將前兩年的研究主題結合,即合併考慮平滑係數與分量迴歸模型,利用縱橫資料分析廠商生產效率。由於結合兩種先進分析方法,期在研究方法與實證分析上,有所突破,故亦打算兼採Monte Carlo Simulation 與利用工廠校正資料進行實證分析。面臨主要困難,在於如何運用分量迴歸技巧,在縱橫資料架構下,估計平滑係數,估計式且須具備一致性和(或)有效性。 This proposal plans to spend three consecutive years to thoroughly examine the production efficiency and productivity of Taiwan’s manufacturing industry, using the newly developed methods of smooth coefficient models and quantile regression. In the first year, the smooth coefficient model will be applied to estimate the production function of the manufacturing sector with inputs labor and capital, say, under the framework of panel data with composed errors. The expenditure of research and development (R&D) of firms may be defined as the smooth variable. Hence, both the coefficients of labor and capital depend on R&D. Furthermore, their marginal products and returns to scale also vary with R&D. A firm spends a larger amount on R&D is expected that its marginal products of labor and capital and production efficiency will be higher than would otherwise. To show the proposed estimators are consistent Monte Carlo simulations will be performed. In the second year, a quantile regression model is applied to study the production efficiency of Taiwan’s manufacturing industry. Different from the previous works, the current study will extend the conventional quantile regression to a panel data setting. Under various quantiles, such as 0.2, 0.4, 0.5, 0.6, 0.8, 0.9, the estimated production coefficients are expected to reflect different production characteristics, which nests the method of ordinary least squares (OLS) or least absolute deviation (LAD) a special case. The OLS and the LAD methods provide information on the averaging behavior or central tendency of a distribution. They fail to offer useful information about the tail behaviors of that distribution. Monte Carlo simulations will be conducted to confirm that the derived estimators are indeed consistent. Finally, the smooth coefficient model and the quantile regression will be combined to investigate the production performance of Taiwan’s manufacturing firms in the context of panel data. The main challenge to be met is how to derive an appropriate estimation procedure that allows for employing the quantile regression approach to estimate the smooth coefficients, on the one hand, and leads to consistent and possibly efficient estimators, on the other. Therefore, Monte Carlo simulations will be adopted as an auxiliary means. |