|
| |
|
|
A Bootstrap Comparison of Robust Regression Estimators
Kalina, Jan ; Janáček, Patrik
The ordinary least squares estimator in linear regression is well known to be highly vulnerable to the presence of outliers in the data and available robust statistical estimators represent more preferable alternatives. It has been repeatedly recommended to use the least squares together with a robust estimator, where the latter is understood as a diagnostic tool for the former. In other words, only if the robust estimator yields a very different result, the user should investigate the dataset closer and search for explanations. For this purpose, a hypothesis test of equality of the means of two alternative linear regression estimators is proposed here based on nonparametric bootstrap. The performance of the test is presented on three real economic datasets with small samples. Robust estimates turn out not to be significantly different from non-robust estimates in the selected datasets. Still, robust estimation is beneficial in these datasets and the experiments illustrate one of possible ways of exploiting the bootstrap methodology in regression modeling. The bootstrap test could be easily extended to nonlinear regression models.
|
|
|
Robust estimation of autocorrelation function
Lain, Michal ; Hudecová, Šárka (advisor) ; Hlávka, Zdeněk (referee)
The autocorrelation function is a basic tool for time series analysis. The clas- sical estimation is very sensitive to outliers and can lead to misleading results. This thesis deals with robust estimations of the autocorrelation function, which is more resistant to the outliers than the classical estimation. There are presen- ted following approaches: leaving out the outliers from the data, replacement the average with the median, data transformation, the estimation of another coeffici- ent, robust estimation of the partial autocorrelation function or linear regression. The thesis describes the applicability of the presented methods, their advantages and disadvantages and necessary assumptions. All the approaches are compared in simulation study and applied to real financial data. 1
|
|
|
L1 Regression
Čelikovská, Klára ; Maciak, Matúš (advisor) ; Hlubinka, Daniel (referee)
This thesis is focused on the L1 regression, a possible alternative to the ordinary least squares regression. L1 regression replaces the least squares estimation with the least absolute deviations estimation, thus generalizing the sample median in the linear regres- sion model. Unlike the ordinary least squares regression, L1 regression enables loosening of certain assumptions and leads to more robust estimates. Fundamental theoretical re- sults, including the asymptotic distribution of regression coefficient estimates, hypothesis testing, confidence intervals and confidence regions, are derived. This method is then compared to the ordinary least squares regression in a simulation study, with a focus on heavy-tailed distributions and the possible presence of outlying observations. 1
|
|
|
Robust Regression Estimators: A Comparison of Prediction Performance
Kalina, Jan ; Peštová, Barbora
Regression represents an important methodology for solving numerous tasks of applied econometrics. This paper is devoted to robust estimators of parameters of a linear regression model, which are preferable whenever the data contain or are believed to contain outlying measurements (outliers). While various robust regression estimators are nowadays available in standard statistical packages, the question remains how to choose the most suitable regression method for a particular data set. This paper aims at comparing various regression methods on various data sets. First, the prediction performance of common robust regression estimators are compared on a set of 24 real data sets from public repositories. Further, the results are used as input for a metalearning study over 9 selected features of individual data sets. On the whole, the least trimmed squares turns out to be superior to the least squares or M-estimators in the majority of the data sets, while the process of metalearning does not succeed in a reliable prediction of the most suitable estimator for a given data set.
|
|
|
Diagnostics for Robust Regression: Linear Versus Nonlinear Model
Kalina, Jan
Robust statistical methods represent important tools for estimating parameters in linear as well as nonlinear econometric models. In contrary to the least squares, they do not suffer from vulnerability to the presence of outlying measurements in the data. Nevertheless, they need to be accompanied by diagnostic tools for verifying their assumptions. In this paper, we propose the asymptotic Goldfeld-Quandt test for the regression median. It allows to formulate a natural procedure for models with heteroscedastic disturbances, which is again based on the regression median. Further, we pay attention to nonlinear regression model. We focus on the nonlinear least weighted squares estimator, which is one of recently proposed robust estimators of parameters in a nonlinear regression. We study residuals of the estimator and use a numerical simulation to reveal that they can be severely heteroscedastic also for data generated from a model with homoscedastic disturbances. Thus, we give a warning that standard residuals of the robust nonlinear estimator may produce misleading results if used for the standard diagnostic tools
|
|
|
On Exact Heteroscedasticity Testing for Robust Regression
Kalina, Jan ; Peštová, Barbora
The paper is devoted to the least weighted squares estimator, which is one of highly robust estimators for the linear regression model. Novel permutation tests of heteroscedasticity are proposed. Also the asymptotic behavior of the permutation test statistics of the Goldfeld-Quandt and Breusch-Pagan tests is investigated. A numerical experiment on real economic data is presented, which also shows how to perform a robust prediction model under heteroscedasticity.
|
|
|
Robustness of High-Dimensional Data Mining
Kalina, Jan ; Duintjer Tebbens, Jurjen ; Schlenker, Anna
Standard data mining procedures are sensitive to the presence of outlying measurements in the data. This work has the aim to propose robust versions of some existing data mining procedures, i.e. methods resistant to outliers. In the area of classification analysis, we propose a new robust method based on a regularized version of the minimum weighted covariance determinant estimator. The method is suitable for data with the number of variables exceeding the number of observations. The method is based on implicit weights assigned to individual observations. Our approach is a unique attempt to combine regularization and high robustness, allowing to downweight outlying high-dimensional observations. Classification performance of new methods and some ideas concerning classification analysis of high-dimensional data are illustrated on real raw data as well as on data contaminated by severe outliers.
|
|
|
Robustness Aspects of Knowledge Discovery
Kalina, Jan
The sensitivity of common knowledge discovery methods to the presence of outlying measurements in the observed data is discussed as their major drawback. Our work is devoted to robust methods for information extraction from data. First, we discuss neural networks for function approximation and their sensitivity to the presence of noise and outlying measurements in the data. We propose to fit neural networks in a robust way by means of a robust nonlinear regression. Secondly, we consider information extraction from categorical data, which commonly suffers from measurement errors. To improve its robustness properties, we propose a regularized version of the common test statistics, which may find applications e.g. in pattern discovery from categorical data.
|
|
|
Nonlinear Trend Modeling in the Analysis of Categorical Data
Kalina, Jan
This paper studies various approaches to testing trend in the context of categorical data. While the linear trend is far more popular in econometric applications, a nonlinear modeling of the trend allows a more subtle information extraction from real data, especially if the linearity of the trend cannot be expected and verified by hypothesis testing. We exploit the exact unconditional approach to propose alternative versions of some trend tests. One of them is the test of relaxed trend (Liu, 1998), who proposed a generalization of the classical Cochran- Armitage test of linear trend. A numerical example on real data reveals the advantages of the test of relaxed trend compared to the classical test of linear trend. Further, we propose an exact unconditional test also for modeling association between an ordinal response and nominal regressor. Further, we propose a robust estimator of parameters in the logistic regression model, which is based on implicit weighting of individual observations. We assess the breakdown point of the newly proposed robust estimator.
|
guest ::
