Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Evaluating VaR and ES for commodities - both conventionally and with neural networks

Fang, David LU and Eile, Måns LU (2020) NEKN02 20201
Department of Economics
Abstract
As commodities are becoming more popular and accessible assets for speculative and hedging purposes, the limited research regarding risk management for said asset-class justifies further contribution to the deficient output. Many previous studies have highlighted the extraordinary high volatility, with non-linear and clustering characteristics associated with commodities. Hence, incorporating volatility forecasts in risk anagement
seems warranted. As the standard risk measurements for market risk in the last decades have been Value at Risk (VaR) and Expected Shortfall (ES), these metrics are evaluated based on a Volatility Weighted Historical Simulation with volatility forecasts provided by a GARCH(1,1) approach and a Recurrent Neural... (More)
As commodities are becoming more popular and accessible assets for speculative and hedging purposes, the limited research regarding risk management for said asset-class justifies further contribution to the deficient output. Many previous studies have highlighted the extraordinary high volatility, with non-linear and clustering characteristics associated with commodities. Hence, incorporating volatility forecasts in risk anagement
seems warranted. As the standard risk measurements for market risk in the last decades have been Value at Risk (VaR) and Expected Shortfall (ES), these metrics are evaluated based on a Volatility Weighted Historical Simulation with volatility forecasts provided by a GARCH(1,1) approach and a Recurrent Neural Network (LSTM) approach for oil, gold and soybean. The data period spans from 1990 to 2019 and the results indicate that both approaches work remarkably well in estimating both VaR and ES. In general, the GARCH(1,1) approach displays somewhat more accurate VaR estimates according to Kupiecs test. However, The LSTM approach does spread the violations more adequate according to Christoffersen’s test. Both approaches display very well specified ES estimates, with somewhat better test statistics for the GARCH(1,1) approach according to the ”Testing ES directly”-test proposed by Acerbi & Szekely. (Less)
Please use this url to cite or link to this publication:
author
Fang, David LU and Eile, Måns LU
supervisor
organization
course
NEKN02 20201
year
type
H1 - Master's Degree (One Year)
subject
keywords
Value-at-Risk, Expected Shortfall, Commodities, GARCH(1, 1), ANN, LSTM, Volatility forecasting, VWHS
language
English
id
9021126
date added to LUP
2020-08-29 11:16:29
date last changed
2020-08-29 11:16:29
@misc{9021126,
  abstract     = {{As commodities are becoming more popular and accessible assets for speculative and hedging purposes, the limited research regarding risk management for said asset-class justifies further contribution to the deficient output. Many previous studies have highlighted the extraordinary high volatility, with non-linear and clustering characteristics associated with commodities. Hence, incorporating volatility forecasts in risk anagement
seems warranted. As the standard risk measurements for market risk in the last decades have been Value at Risk (VaR) and Expected Shortfall (ES), these metrics are evaluated based on a Volatility Weighted Historical Simulation with volatility forecasts provided by a GARCH(1,1) approach and a Recurrent Neural Network (LSTM) approach for oil, gold and soybean. The data period spans from 1990 to 2019 and the results indicate that both approaches work remarkably well in estimating both VaR and ES. In general, the GARCH(1,1) approach displays somewhat more accurate VaR estimates according to Kupiecs test. However, The LSTM approach does spread the violations more adequate according to Christoffersen’s test. Both approaches display very well specified ES estimates, with somewhat better test statistics for the GARCH(1,1) approach according to the ”Testing ES directly”-test proposed by Acerbi & Szekely.}},
  author       = {{Fang, David and Eile, Måns}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{Evaluating VaR and ES for commodities - both conventionally and with neural networks}},
  year         = {{2020}},
}