<fmt:message key='jsp.layout.header-default.alt'/>  
 

DSpace@UM >
Faculty of Science >
PhD Theses : Science >

Please use this identifier to cite or link to this item: http://hdl.handle.net/1812/583

Title: L1–Garch models: Parameter estimations, performance measures and its applications
Authors: Shamsul Rijal Muhammad Sabri
Keywords: L1–Garch models
Parameter estimation
Performance measure
RJB test
Issue Date: 2008
Publisher: University Malaya
Abstract: Since the failure of the high frequency data appears in many researches, economists have explored some alternative methods to overcome heteroscedastic modeling. One such model that has been introduced recently is robust L1–GARCH family. These tools have been introduced initially due to the statistical properties and their performance in parameter estimates. In this thesis, we further explore the performance of the L1–GARCH particularly, in estimating conditional variances. The estimation is enhanced by first employing an exploratory data analysis before proceeding with L1–GARCH. To better understand the behavior of the estimates as well as the performance of these methods, simulation studies were carried out. The results obtained suggest that, ordinary GARCH(1,1) performs well in estimating conditional variances in the absence of outliers or contaminants in the data. However, L1–GARCH(1,1) outperform the GARCH(1,1) in estimating the conditional variances in the presence of outliers. Another problem that persists with heteroscedastic modeling is that of the goodness-of-fit test. In GARCH models, the most common adequacy test used is the classical Jarque-Bera(JB) test. This test however, is known to be extremely sensitive to outliers and hence a single outlier may lead to failure of normality assumption. To overcome this, we introduce robust JB (RJB) measures that are (i) less sensitive to the presence of outliers and (ii) able to detect the departure from the usual normal distribution (symmetric heavy tailed). In assessing the performance of and , the test statistics are compared, in the presence of outliers and symmetric heavy tailed alternatives; here we conduct simulation studies to calculate the power of rejecting the null hypothesis of the test (the data is normal distributed). Our simulations demonstrate that the are able to yield (i) good result to overcome the presence of outliers of the data and (ii) as efficient as other robust statistics that were introduced by others previously when the existence of assumption other than normal distribution. We also apply robust L1–ARCH model to compute the uncertainty of inflation of ASEAN– 5 countries. For completion, we examine the relationship between uncertainty of inflation and their economic growth using robust regression models. We find overwhelming statistical evidence supporting the hypothesis that increased inflation uncertainty lead to slow down in economic growth. This positive association between inflation uncertainty and growth is consistent with earlier studies, conducted for the major industrialized countries. This finding is in line with Friedman’s hypothesis that suggests uncertainty concerning regime changes depresses real economic activity. Throughout this thesis, the SPLUS programming language is used to run (a) simulation tests towards RJB tests; (b) estimating parameters of L1–GARCH model; (c) comparison of efficiency of conditional variance between ordinary GARCH(1,1) and L1–GARCH(1,1) and (d) estimating parameters of L1–ARCH model for inflation uncertainty amongst ASEAN-5 countries.
Description: Thesis (PhD) -- Faculty of Science, University of Malaya, 2008.
URI: http://dspace.fsktm.um.edu.my/handle/1812/583
Appears in Collections:PhD Theses : Science

Files in This Item:

File Description SizeFormat
L1-GARCH Models Parameter Estimations Performance Measures a.pdfThesis Text948.06 kBAdobe PDFView/Open


This item is protected by original copyright



Your Tags:

 

  © Copyright 2008 DSpace Faculty of Computer Science and Information Technology, University of Malaya . All Rights Reserved.
DSpace@UM is powered by MIT - Hawlett-Packard. More information and software credits. Feedback