Browsing by Subject "Regression analysis."

Now showing 1 - 12 of 12

Open Access
Comparison of several estimators for the covariance of the coefficient matrix
(1995) Orhan, Mehmet
The standard regression analysis assumes that the variances of the disturbance terms are constant, and the ordinary least squares (OLS) method employs this very crucial assumption to estimate the covariance of the disturbance terms perfectly, but OLS fails to estimate well when the variance of the disturbance terms vary across the observations. A very good method suggested by Eicker and improved by White to estimate the covariance matrix of the disturbance terms in case of heteroskedeisticity was proved to be biased. This paper evaluates the performance of White’s method as well as the OLS method in several different settings of regression. Furthermore, bootstrapping, a new method which very heavily depends on computer simulation is included. Several types of this method are used in several cases of homoskedastic, heteroskedastic, balanced, and unbalanced regressions.
Open Access
Detecting structural change when the change point is unknown
(1995) Başçı, Sıdıka
There are various tests which are used to detect structural change when the change point is unknown. Among these widely used ones are Cumulated Sums (CUSUM) and CUSUM of Squares tests of Brown, Durbin and Evans (1975), Fluctuation test of Sen (1980) and Ploberger, Krämer and Kontrus (1989). More recently, Andrews (1990) suggests Sup F test and shows that it performs better than the above stated tests in terms of power. The problem with these tests is that they all assume stable variance although the regression coefficients change while moving from one regime to the other. In this thesis, we relax this assumption and suggest an alternative test which also allows heteroskedasticity. For this aim, we follow the Bayesian approach. We also present some of the Monte Carlo study results where we find that Bayesian test has superiority over the above stated tests in terms of power.
Open Access
Evaluation of the Goldfeld-Quandt test and alternatives
(1994) Tomak, Kerem
In this study, the widely used Coldfeld-C^uandt test for lieterosk('da.sticity in the linear regression model is evaluated. VV(' reduce the dimension of the data spa.ce that is needed lor tin' computaticui of tlu' t('sts. VVe tlu'ii compa.r(‘ the pi'rformaiK'es of tin' Likelihood Ratio and tin* Cloldh'ld-C^uandt tests by using stringency measure. The problem of analytically non-tractable distribution function in the case of the Likelihood Ratio test is overcome by employing Monte Carlo methods. It is observed that the Likelihood Ratio test is better in most of the cases than the Goldfeld-Quandt test.
Open Access
Modelling the relationship between productivity, employment and wages in Turkish small and medium sized enterprises,1981-1998
(2001) Demirel, Görkemli
This thesis analyzes the empirical relationship between wages and productivity as well as the relationship between wages and employment in Turkish manufacturing industry. Unlike the previous studies done for manufacturing industry, in this study the size definitions of manufacturing industry, sectoral distribution and the sectoral division between public and private sector are considered. In the empirical part of the thesis, first wage and productivity and wage and employment relationships are estimated by using OLS method. After finding out both wage-productivity and wage-employment relationships are significant, descriptive growth rate comparisons are made for the period of 1981-1998. The main conclusion that emerges from both analyses is that relationship between variables of interest is valid. Wages, productivity and employment relationship have important policy implications regarding especially on Turkish small and medium sized enterprises.
Open Access
Nonstationary factor model applications of Elastic Net
(2013) Konak, Deniz
In this thesis, we adopted Elastic Net estimators for selecting true number of factors in factor models with stationary and nonstationary factors. Elastic Net is a member of shrinkage estimators family. As a member of shrinkage estimators family, elastic net estimators are stable to changes in data and in general they do not over parametrize the models. These two properties of elastic net estimators makes elastic net more favourable than information based criterion penalty methods for estimating true factor number. Since Principal Components Analysis (PCA) based algorithms always tends to give only single factor for nonstationary data sets, we use Sparse Principal Components Analysis (SPCA) algorithm which is a regression-type optimization formulation of PCA. Simulations show the performance of Elastic Net estimator for estimation of true factor number with stationary and nonstationary factors cases .
Open Access
A powerful test for unit root and an application to GNP of seven OECD countries
(2000) Ustundag, Aliye
This thesis uses a powerful test, Dickey-Fuller Generalized Least Squares (DF-GLS), to see whether unit root exists or not in real GNP of OECD Countries - Australia, Canada, Germany, Japan, Italy, U.K. and U.S. - for the years between the first quarter of 1960 and the second quarter of 1998 by using quarterly data that takes 1995 as base year. For this purpose a simple model with a deterministic component plus an error term, which is assumed to be AR (1), is used. The results of the regressions show the existence of unit root for all of the considered countries. Furthermore, we give flnite sample performances of Augmented Dickey-Fuller (ADF) test and DF-GLS tests which Elliott et al. (1996) conducted by using Monte Carlo experiment.
Open Access
Regression by selecting best feature(s)
(2000) Aydın, Tolga
Two new machine learning methods, Regression by Selecting Best Feature Projections (RSBFP) and Regression by Selecting Best Features (RSBF), are presented for regression problems. These methods heavily make use of least squares regression to induce eager, parametric and context-sensitive models. Famous regression approaches of machine learning and statistics literature such as DART, MARS, RULE and kNN can not construct models that are both predictive and have reasonable training and/or querying time durations. We developed RSBFP and RSBF to fill the gap in the literature for a regression method having higher predictive accuracy and faster training and querying time durations. RSBFP constructs a decision list consisting of simple linear regression lines belonging to linear features and/or categorical feature segments. RSBF is the extended version of RSBFP such that the decision list consists of both simple, belonging to categorical feature segments, and/or multiple, belonging to linear features, linear regression lines. A relevancy heuristic has been developed to determine the features involved in the multiple regression lines. It is shown that the proposed methods are robust to irrelevant features, missing feature values and target feature noise, which make them suitable prediction tools for real-world databases. In terms of robustness, RSBFP and RSBF give better results when compared to other famous regression methods.
Open Access
Robust regression and applications
(1996) Kıracı, Arzdar
This study analyzes the effect of outliers in the regression analysis with the help of a written program in the programing language of GAUSS. The analysis relies on the subject of Robust Regression, which is explained and supported by experiments and applications. The applications contain examples to show the superiority of this technique.
Open Access
Robust regression, HCCM estimators, and an Empirical Bayes application
(1999) Orhan, Mehmet
This Ph.D. thesis includes three topics of econometrics where the chapters of the whole study are devoted to robust regression analysis, research on the estimators for the covariance matrix of a heteroskedastic regression and finally an application of the Empirical Bayes method to some real data from Istanbul Stock Exchange. Some robust regression techniques are applied to some data sets to show how outliers of a data set may lead to wrong inferences. The results reveal that the former studies have gone through some wrong results with the effect of the outliers that were not detected. Second chapter makes a thorough evaluation of the existing heteroskedasticity consistent covariance matrix estimators where the Maximum Likelyhood estimator recently promoted to the literature by Zaman is also taken into consideration. Finally, some empirical study is carried out in the last part of the thesis. The firms of ISE are categorized into sectors and some estimation is done over an equation which is very common and simple in the finance literature.
Open Access
Spurious regression problem in Kalman Filter estimation of time varying parameter models
(2010) Eroğlu, Burak Alparslan
This thesis provides a simulation based study on Kalman Filter estimation of time varying parameter models when nonstationary series are included in regression equation. In this study, we have performed several simulations in order to present the outcomes and ramifications of Kalman Filter estimation applied to time varying regression models in the presence of random walk series. As a consequence of these simulations, we demonstrate that Kalman Filter estimation cannot prevent the emergence of spurious regression in time varying parameter models. Furthermore, so as to detect the presence of spurious regression, we also propose a new method, which suggests penalizing Kalman Filter recursions with endogenously generated series. These series, which are created endogenously by utilizing Cochrane’s variance ratio statistic, are replaced by state evolution parameter Tt in transition equation of time varying parameter model. Consequently, Penalized Kalman Filter performs well in distinguishing nonsense relation from a true cointegrating regression.
Open Access
Testing Marshall-Lerner Condition: a non-parametric approach
(2000) Yücel, Mustafa Eray
This study examines the determinants of trade flows for six developed countries. Volume of imports (exports) is regressed on relative import (export) price and domestic (world) real income using non-parametric kernel estimation techniques. On quarterly data, Local Constant Least Squares (LCLS) and Local Linear Least Squares (LLLS) estimates of trade elasticities are obtained. Using pointwise and point estimates of these elasticities, the Marshall-Lemer Condition is checked for our sample countries. The condition is satisfied for two of our six sample countries. Although the existing controversy on the subject has not been solved using non-parametric regression techniques, a new room is opened for further investigation by presenting the time-behaviour of trade elasticities.
Open Access
A time series analysis of the Japanese yen with monthly data
(1996) Eltaş, Metin
The purpose of this thesis is to obtain a function which will help in using the exchange rate between the Japanese Yen (Yen) and the United States Dollar (Dollar) as an investment alternative. A three-step method is followed throughout this study. Yen and the set of five countries' exchange and interest rates is searched at the first step. Mullticolinearity and nonstationarity problems are observed at this stage. At the second step the data set is converted into a stationary form by taking the first differences. Then regression is applied and no significant correlation is found. At the final step relation between Yen and three subgroups from the data set are examined and no significant relation is found again. This thesis concludes by explaining the outcomes of our analyses.