Kutzker, Tim (2020). Specification Testing in Econometric Models. PhD thesis, Universität zu Köln.
|
PDF
Kutzker.pdf Download (1MB) | Preview |
Abstract
The thesis consits of three independent articles. First, specification tests for the m-dimensional spatial autoregressive (SAR) panel model are provided. Therefore, we derive the limiting distribution of the specification test statistics and examine size and power properties in finite sample simulations. In the empirical application we analyze the Euro Stoxx 50 returns. Regarding this, a 3-dimensional SAR panel model incorporating global dependencies, dependencies inside industrial branches and local dependencies is assumed. The investigation shows the tests’ ability to detect inaccurate Value-at-Risk forecasts. Secondly, we propose a new non-parametric test for detecting relevant breaks in copula functions. We assume that the data is driven by two non-equal copulas C1 and C2. Under the null hypothesis, the copula difference within an appropriate norm is smaller than a certain positive adjustable threshold . Within the alternative hypothesis, the copula difference exceeds the fixed value. The test is based on a cumulative sum approach of the empirical copula with sequentially estimated marginals. We propose a bootstrap procedure to compute critical values. The Monte Carlo simulation study indicates that the test results in a reasonable sized and powered testing procedure. A real data application of the DAX30 up to cross sectional dimension N = 30 shows the test’s ability to detect relevant break points. Finally, we propose a novel consistent specification test for quantile regression models where we allow the covariate effects to be quantile dependent and nonlinear. To achieve this, we parameterize the conditional quantile functions by appropriate basis functions, rather than parametrically and hence allowing to test for functional forms beyond linearity while retaining the linear cases as special cases. Due to the dependence on the quantile itself covariate-quantile relations can differ for distinct quantiles. The induced class of conditional distribution functions can finally be tested with a Cramér-von Mises type test statistic. We derive the theoretical limit distribution and propose a practical bootstrap method. To increase the power of our test, we suggest a modified test statistic using quantile regression splines. A detailed Monte Carlo experiment shows that the test results in a reasonable sized testing procedure with large power. An application to conditional income disparities between East and West Germany over the period 2001 − 2010 indicates that there are still significant differences across the quantiles of the conditional income distributions, when conditioning on age.
Item Type: | Thesis (PhD thesis) | ||||||||
Creators: |
|
||||||||
URN: | urn:nbn:de:hbz:38-123231 | ||||||||
Date: | 18 August 2020 | ||||||||
Language: | English | ||||||||
Faculty: | Faculty of Management, Economy and Social Sciences | ||||||||
Divisions: | Faculty of Management, Economics and Social Sciences > Economics > Econometrics and Statistics > Professorship for Statistics and Econometrics | ||||||||
Subjects: | Economics Mathematics |
||||||||
Uncontrolled Keywords: |
|
||||||||
Date of oral exam: | 18 August 2020 | ||||||||
Referee: |
|
||||||||
Refereed: | Yes | ||||||||
URI: | http://kups.ub.uni-koeln.de/id/eprint/12323 |
Downloads
Downloads per month over past year
Export
Actions (login required)
View Item |