Research Of TFP Calculation Based On Regression Model With Convex Constraint

The research of productivity is very important in economics. The focus of productivity research has changed from partial factor productivity to total factor productivity (TFP) after the 1930s. This indicates the beginning of modern productivity research, and it also promotes the productivity research onto a new altitude. From the research on TFP, we can analyze the origin of one country's economic growth based on the different contribution rate of different factors. Thus we can forecast whether a country's economic growth can last or not. What's more, labor productivity and capital productivity subtracted from total productivity gives TFP. Namely, the extra growth which can't be explained by labor and capital should be attributed to technology advancement. In view of these, no country will ignore the important role of TFP when making macroeconomic policy.On the ground of domestic and oversea existing research, this paper sums up the theory of TFP, and proposes a new TFP calculating method based on generalized linear regression model with convex constraint. The new method avoids some overstrict economic hypothesis which may lead to some unreasonable conclusion. The empirical result shows that the new method is more effective and reasonable when calculating China's TFP. The main content of this paper is as follows:The first part introduces the concept of total factor productivity, and elaborates the research background, significance, domestic and foreign research process of the TFP calculation. The second part reviews the generation of TFP calculating methods, and presents Cobb-Douglas productivity function and Solow residual method. In the third part, we propose a kind of generalized linear regression model with convex constraint and unknown dependent variable in order to avoid the invalid economic hypothesis. Then we give the least square estimation of the parameters using iterative projection between two convex sets, and give the maximum likelihood estimation of the parameters using EM algorithm. The fourth part calculates China's TFP and its growth rate respectively with traditional method and new method. The fifth part establishes index system to evaluate national science and technology innovation ability. It can show the China's innovation ability level among the selected countries. Based on the rank we can compare traditional method with new method. The last part analyzes contribution rate of each input factor to economic growth quantitatively based on new method's result since 1980.