Push it to the limit - Testing the usefulness of Extreme Value Theory in electricity markets
(2012) NEKP01 20122Department of Economics
- Abstract (Swedish)
- We set out to investigate whether the methodologies used in extreme value analysis are applicable in estimating Value at Risk (VaR) for the spot price returns of the European Energy Exchange (EEX). An initial inspection of hourly data reveals a volatile behaviour where returns of extreme proportions occur frequently. Applying the two traditional extreme value theory methods to the data, Block Maxima and Peaks Over Threshold, we find that while the latter perform better, it is highly sensitive to the chosen threshold and thereby the confidence level.
Suspecting that spikes in the original data series might have distorted the estimated parameters and consequently the measures of VaR, we perform a similar analysis on daily data obtained... (More) - We set out to investigate whether the methodologies used in extreme value analysis are applicable in estimating Value at Risk (VaR) for the spot price returns of the European Energy Exchange (EEX). An initial inspection of hourly data reveals a volatile behaviour where returns of extreme proportions occur frequently. Applying the two traditional extreme value theory methods to the data, Block Maxima and Peaks Over Threshold, we find that while the latter perform better, it is highly sensitive to the chosen threshold and thereby the confidence level.
Suspecting that spikes in the original data series might have distorted the estimated parameters and consequently the measures of VaR, we perform a similar analysis on daily data obtained through aggregating the hourly observations. Now we find a clear weekly dependence and an AR-GARCH-filtering approach suggested by McNeil and Frey (2000) is employed. Unfortunately, the risk measures based on daily data did not improve the overall picture as much as we had hoped, as they clearly still deviate from their theoretical values and could be rejected for most confidence levels.
Backtesting the daily data did improve the results and we found that a conditional AR-GARCH-model outperformed an unconditional one, however only slightly. Our findings would thus suggest that the two classical extreme value methodologies can be used to model the extreme tails of the return distribution, but that they are not as accurate as found in other electricity markets. In order to increase the accuracy, one would need to constantly update the model parameters. Furthermore, we believe that more advanced modelling, taking spikes and mean reversion of the data into account, could lead to improvements. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/3166413
- author
- Fulgentiusson, Henrik LU
- supervisor
-
- Birger Nilsson LU
- Rikard Green LU
- organization
- course
- NEKP01 20122
- year
- 2012
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- Electricity, Extreme Value Theory, Peaks Over Threshold, Block Maxima, Value at Risk.
- language
- English
- id
- 3166413
- date added to LUP
- 2012-11-16 14:20:46
- date last changed
- 2012-11-16 14:20:46
@misc{3166413,
abstract = {{We set out to investigate whether the methodologies used in extreme value analysis are applicable in estimating Value at Risk (VaR) for the spot price returns of the European Energy Exchange (EEX). An initial inspection of hourly data reveals a volatile behaviour where returns of extreme proportions occur frequently. Applying the two traditional extreme value theory methods to the data, Block Maxima and Peaks Over Threshold, we find that while the latter perform better, it is highly sensitive to the chosen threshold and thereby the confidence level.
Suspecting that spikes in the original data series might have distorted the estimated parameters and consequently the measures of VaR, we perform a similar analysis on daily data obtained through aggregating the hourly observations. Now we find a clear weekly dependence and an AR-GARCH-filtering approach suggested by McNeil and Frey (2000) is employed. Unfortunately, the risk measures based on daily data did not improve the overall picture as much as we had hoped, as they clearly still deviate from their theoretical values and could be rejected for most confidence levels.
Backtesting the daily data did improve the results and we found that a conditional AR-GARCH-model outperformed an unconditional one, however only slightly. Our findings would thus suggest that the two classical extreme value methodologies can be used to model the extreme tails of the return distribution, but that they are not as accurate as found in other electricity markets. In order to increase the accuracy, one would need to constantly update the model parameters. Furthermore, we believe that more advanced modelling, taking spikes and mean reversion of the data into account, could lead to improvements.}},
author = {{Fulgentiusson, Henrik}},
language = {{eng}},
note = {{Student Paper}},
title = {{Push it to the limit - Testing the usefulness of Extreme Value Theory in electricity markets}},
year = {{2012}},
}