Advanced

Modeling and Forecasting Volatility in Copper Price Returns with GARCH Models

Goodwin, Dominice LU (2012) NEKH01 20121
Department of Economics
Abstract
This thesis offers a study on how well the standard GARCH(1,1) model, the GJR-GARCH(1,1) model and the QGARCH(1,1) model, were able to model (in-sample) and forecast (out-of-sample) the volatility of copper spot price returns in four equally large subsamples within the period July 21, 1993 to 22 Mars, 2012. The results shows that the GARCH models' ability to model the conditional variance (in-sample) was highly satisfactory for the three subsamples in which there was found significant presence of ARCH effects. In out-of-sample forecasting the GARCH models dominated a Random Walk model across all four subsample (17 to 29 % lower Mean of Absolute Errors for the GARCH models). In three out of the four subsamples the GARCH models also... (More)
This thesis offers a study on how well the standard GARCH(1,1) model, the GJR-GARCH(1,1) model and the QGARCH(1,1) model, were able to model (in-sample) and forecast (out-of-sample) the volatility of copper spot price returns in four equally large subsamples within the period July 21, 1993 to 22 Mars, 2012. The results shows that the GARCH models' ability to model the conditional variance (in-sample) was highly satisfactory for the three subsamples in which there was found significant presence of ARCH effects. In out-of-sample forecasting the GARCH models dominated a Random Walk model across all four subsample (17 to 29 % lower Mean of Absolute Errors for the GARCH models). In three out of the four subsamples the GARCH models also dominated the unconditional variance (8 to 19 % lower Mean of Absolute Errors for the GARCH models). In one subsample the GARCH models produced about 8 to 10 % higher MAE than the unconditional variance. The results did also indicate that no one of the three GARCH models used in this thesis could generally be recommended over the others for the purpose of forecasting the one-step ahead one day variance in copper spot price returns. This conclusion is based on the facts that the differences between the GARCH models in the out-of-sample forecasting was small, and that it was difficult to predict which model that would have the slightly better forecasting performance. This conclusion may be interpreted as that the standard GARCH(1,1) model was sufficient. (Less)
Please use this url to cite or link to this publication:
author
Goodwin, Dominice LU
supervisor
organization
course
NEKH01 20121
year
type
M2 - Bachelor Degree
subject
keywords
GARCH(1, 1), GJR-GARCH(1, QGARCH(1, forecasting volatility, Copper
language
English
id
3049334
date added to LUP
2012-09-27 12:51:59
date last changed
2012-09-27 12:51:59
@misc{3049334,
  abstract     = {This thesis offers a study on how well the standard GARCH(1,1) model, the GJR-GARCH(1,1) model and the QGARCH(1,1) model, were able to model (in-sample) and forecast (out-of-sample) the volatility of copper spot price returns in four equally large subsamples within the period July 21, 1993 to 22 Mars, 2012. The results shows that the GARCH models' ability to model the conditional variance (in-sample) was highly satisfactory for the three subsamples in which there was found significant presence of ARCH effects. In out-of-sample forecasting the GARCH models dominated a Random Walk model across all four subsample (17 to 29 % lower Mean of Absolute Errors for the GARCH models). In three out of the four subsamples the GARCH models also dominated the unconditional variance (8 to 19 % lower Mean of Absolute Errors for the GARCH models). In one subsample the GARCH models produced about 8 to 10 % higher MAE than the unconditional variance. The results did also indicate that no one of the three GARCH models used in this thesis could generally be recommended over the others for the purpose of forecasting the one-step ahead one day variance in copper spot price returns. This conclusion is based on the facts that the differences between the GARCH models in the out-of-sample forecasting was small, and that it was difficult to predict which model that would have the slightly better forecasting performance. This conclusion may be interpreted as that the standard GARCH(1,1) model was sufficient.},
  author       = {Goodwin, Dominice},
  keyword      = {GARCH(1,1),GJR-GARCH(1,QGARCH(1,forecasting volatility,Copper},
  language     = {eng},
  note         = {Student Paper},
  title        = {Modeling and Forecasting Volatility in Copper Price Returns with GARCH Models},
  year         = {2012},
}