Date of Submission


Date of Award


Institute Name (Publisher)

Indian Statistical Institute

Document Type

Doctoral Thesis

Degree Name

Doctor of Philosophy

Subject Name

Quantitative Economics


Economic Research Unit (ERU-Kolkata)


Bhattacharya, Nikhilesh (ERU-Kolkata; ISI)

Abstract (Summary of the Work)

Very often in eoonometrio enslysis one adopts the classical lineer regression model. The classical linear regression model is given by If, in addition, e is assumed to be normally đistributed, the model is called classical normal1 linear regression mode1.Ordinary least squares (0LS) methods of estimation and hypothesis testing are besed on this ndal, d eveluton copy of CV POFO But the assumptions on ɛis and- xs may not be fulfilled in reality; or, in other words, the model may not be correctly specified. Cne class of problems arises when some of the regressors are omitted from the equation and/or scme additional regressors are yrongiy included in the model, or when the algebraic form of the regression equation is misspecified. In such cases QLS method Kould fail to give satisfactory estimates of the regression coefficients.Another class of problems is created when E(Ee) o1, . Ceneralised least squares techniques are called for in such situations. Problems also arise when the regressors (X) are stochästic. There is little trouble if x is stochastic but fully independent of 8. However, if the regressors and disturbances are correlated, OLS estimates cease to be unbiased. The danger is particularly great if the regressor values and the disturbances in the same ob servational equation are correlated. In this case, OLS estimates of Bs are not even asymptotically unbiased. This kind of complication arises in two important situations :(à ). where the regressors are observed with errors and(b) where the equation is embedded in simultaneous equetion models where several current endogedous variables are determined through the simultaneous interactions of the structural relationships in the model.This study, is largely concerned with1. Problems of nmission of regressors from a single e quation regression model leading to autocorrelation among the disturbances2. MSE criterion in the context of specification error analysis with stochastic regressors and3. Handling of errors in variable models with trending r autocorrelated errors. Below we give a summary of different chapters in the thesis.Chapter l. A survey of previous researehes. This chapter gives a brief survey of existing literature on three main problems of econometrics (single equation methods) tn provide a background to the investigations reported in this thesis- The problems are those arising due to(a) Omission of relevant regressors from a regression equation and misspecification of algebraic forms(b) Autocorrelation of disturbances(c) Errors in variables.(a) Omission of relevant regres sors from a regression equation and misspecification of algebraic forms : The survey has been organised under the following beads(1) The consequences of using OLS procedures for estimating the regression coefficients of a misspecified model.(11) Applications of specification analysis.111) Different tests of misspecification and their appli- cationse(iv) The residual variance criterion.(v) The method of using least squares to approximate unknown regression functions.(vi) Conse quences of misspecification in simultaneous e quation systems.


ProQuest Collection ID:

Control Number


Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.


Included in

Mathematics Commons