Advanced

Identifying an Appropriate Risk Model for Quantifying Foreign Exchange Portfolio Exposure

Strömberg, Albert LU and Rior, Maxine LU (2016) EXTM10 20161
Department of Economics
Abstract
In this master thesis we investigate the predictive performance of several well-known Value at Risk (VaR) estimation methods, given a specific company setting. The thesis is conducted at a Swedish financial institution that wants to be able to measure foreign exchange (FX) risk induced by cross-currency transactions carried out by the company. The goal of the thesis is to provide empirical evidence for the best model to predict risks stemming from the FX portfolio of the financial institution. In addition, a risk assessment tool is developed for the company to use in daily risk monitoring. The tested VaR models are applied to daily foreign exchange data from seven currency pairs as well as a portfolio consisting of the same currency pairs.... (More)
In this master thesis we investigate the predictive performance of several well-known Value at Risk (VaR) estimation methods, given a specific company setting. The thesis is conducted at a Swedish financial institution that wants to be able to measure foreign exchange (FX) risk induced by cross-currency transactions carried out by the company. The goal of the thesis is to provide empirical evidence for the best model to predict risks stemming from the FX portfolio of the financial institution. In addition, a risk assessment tool is developed for the company to use in daily risk monitoring. The tested VaR models are applied to daily foreign exchange data from seven currency pairs as well as a portfolio consisting of the same currency pairs. The VaR methods include historical simulation, parametric as well as extreme value theory approaches, sometimes in combination with volatility estimators. Model performance is evaluated using both quantitative and qualitative backtests according to Basel standard. The backtests enable a ranking of the models in order to find the best-performing one. The thesis concludes that the use of a volatility estimator almost always provides better-working models. The backtests indicate that the final, best model uses the historical simulation approach with an observation window size set to 1,000 trading days combined with an EWMA estimator with a decay factor of 0.85. (Less)
Please use this url to cite or link to this publication:
@misc{8883129,
  abstract     = {In this master thesis we investigate the predictive performance of several well-known Value at Risk (VaR) estimation methods, given a specific company setting. The thesis is conducted at a Swedish financial institution that wants to be able to measure foreign exchange (FX) risk induced by cross-currency transactions carried out by the company. The goal of the thesis is to provide empirical evidence for the best model to predict risks stemming from the FX portfolio of the financial institution. In addition, a risk assessment tool is developed for the company to use in daily risk monitoring. The tested VaR models are applied to daily foreign exchange data from seven currency pairs as well as a portfolio consisting of the same currency pairs. The VaR methods include historical simulation, parametric as well as extreme value theory approaches, sometimes in combination with volatility estimators. Model performance is evaluated using both quantitative and qualitative backtests according to Basel standard. The backtests enable a ranking of the models in order to find the best-performing one. The thesis concludes that the use of a volatility estimator almost always provides better-working models. The backtests indicate that the final, best model uses the historical simulation approach with an observation window size set to 1,000 trading days combined with an EWMA estimator with a decay factor of 0.85.},
  author       = {Strömberg, Albert and Rior, Maxine},
  keyword      = {Foreign exchange,Risk management,Value at Risk,Expected Shortfall,Historical simulation,Volatility clustering,Volatility-weighted historical simulation,Age-weighted historical simulation,EWMA,GARCH(1,1),Normal distribution,Student t-distribution,Extreme value theory,Peaks over threshold,Backtesting,Christoffersen test,Kupiec test,Lopez I test},
  language     = {eng},
  note         = {Student Paper},
  title        = {Identifying an Appropriate Risk Model for Quantifying Foreign Exchange Portfolio Exposure},
  year         = {2016},
}