The Theory Basis and Practice of Econometric Diagnosis
Zhao Jin-wen · SSRN Electronic Journal · 1998
This is a monographic study on diagnosing econometric modeling. I systematically introduce the ideas and methods of statistical diagnostics to the whole process of classical modern econometric modeling which provide the basic frame and theoretical basis and factual evidence for the birth of Econometric Diagnostics. There are ten chapters in this dissertation. I organize them as follows: Chapter l is the review of econometric modeling. I give a short retrospect to its theories and methods and point out their shortcomings. At the same time, I propose the strategic plans that use the idea of diagnostics to the whole process of econometric modeling. A detailed description of data influence measures can be seen in Chapter 2. I introduce a crop of diagnostic statistics and criteria to judge the effect in whole process of econometric modeling. In Chapter 3, I discuss the influence analysis of estimation in experimental modeling and lay emphasis on the contamination issue of estimation. Particularly I propose a new approach to diagnose the multicollinearity-influential observations and verify the validity of the method with living examples. In addition, I propose the diagnosis and influence depiction for seasonal fluctuation for the first time. Furthermore, I study the data influential analysis on model transformation in Chapter 4 and pick out the influence observation group at the power transformation model of Chinese consumption data. In Chapter 5, I inquire into the influence analysis of variable selections in modeling through data, multicollinearity and model mis-specification. I develop the diagnosis theory of statistical hypothesis and propose new concepts of testing influence case (or group) and hide from view in Chapter 6 which are the base of scrutinizing the whole process of econometric modeling. Meanwhile I illustrate with the ordinary econometric tests as examples. In Chapter 7, I present the diagnosis techniques in time series modeling. I pay our attention on the test diagnosis, the identification of AO and AI outliers, and the influential diagnosis in model ordering. In Chapter 8, I discuss the diagnosis theory of Bayesian modeling and analyze the skills of diagnosing outliers, estimation and prediction influence, and inference diagnosis. I study the diagnosis of structural changes in Chapter 9 and specialize in diagnosing the constant test of parameters, Bayesian identification of change points and the techniques of plot diagnosis. Finally I discuss the first class theory of cointegration and its diagnosis in Chapter 10. The theory of diagnosis for cointegration is proposed first by myself. In addition, I'll develop the detail diagnosis theory of nonlinear econometrics and Monte Carlo simulation, which is a virgin soil now. I believe that the econometric diagnostics will be a new branch of knowledge that is cautious of theory and adequate system and worth much in practice.