FUKUSHIGE Mototsugu Nagoya City University, Department of Economics, Associate Professor, 経済学部, 助教授 (10208936)
OKAMURA Kumiko Toyama University, Department of Economics, Associate Professor, 経済学部, 助教授 (20281016)
TANABE Kunio Institute of Statistical Mathematics, Department of Prediction and Control, Prof, 予測制御研究系, 教授 (50000203)
|Budget Amount *help
¥4,800,000 (Direct Cost : ¥4,800,000)
Fiscal Year 1997 : ¥2,100,000 (Direct Cost : ¥2,100,000)
Fiscal Year 1996 : ¥2,700,000 (Direct Cost : ¥2,700,000)
For many economists, it is common practice to use a parametric specification of production functions (PF), such as Cobb-Douglas or Translog production function, when they analyze the production technology. However, these parametric specifications are subject to their own functional biases. In this paper, to partially remedy these shortcomings, we propose Baysian non-parametric estimation method of a PF,as well as analyzing the functional biases of some of the specifications in the CES family.
Employing this method, the input space was partitioned into cells, and the production level for each cell was estimated by Baysian method. The assessed production surface was a collection of piece-wise constant "steps" over these cells. In the estimation process, the relationships between the steps have been modeled as a prior information to the model, so that the smoothness of the function is maintained.
In this research we used micro data (Census of Manufactures, MITI) in estimating the production
function. This data set include 22 industries and about 50,000 records in each year.
With the census data, surveyed for units of operation, we want to examine the input-output technical relationship (s) for manufacturing industries listed in the Japan Standard Industrial Classification code.
We want to estimate the production surface (function ) from the micro data which will affects production technology more properly. Within the knowledge of the authors, this kinds of micro data for manufacturing production and its analysis are rare.
We have employed the Baysian non-parametric density estimation (BNDE) method for the data analysis. It is important to note that we do not parameterize the estimating function such as Cobb-Douglas and Translog form.
Using the program developed for estimating the density, we estimated the most efficient production surface. From this most efficient production surface, we can analyze the distribution of inefficient production units for each "class" of production, in which the classes are defined based upon the levels of inputs. Less