Backtesting of value at risk estimation methods for energy commodity trading: evaluating performance and identifying the optimal approach

This paper focuses on the application of Value at Risk (VaR) estimation techniques in the context of energy commodity trading, with a specific emphasis on the NYMEX (NG) and Title Transfer Facility (TTF) natural gas index. The primary point of this study is to break down and assess the VaR assessment strategies and recognize the most reasonable methodology. Historical price data for these commodities were collected, and forward prices were generated using a Long Short-Term Memory (LSTM) forecasting model. A portfolio comprising Future and Forward contracts for both NG and TTF index was constructed and three VaR estimation approaches: Variance-Covariance, Historical Simulation, and Monte Carlo Simulation were employed. Backtesting was then performed using the metric Quadratic Probability Score (QPS) to determine the optimal approach. Sensitivity analysis was also conducted to evaluate the variation in the estimated VaR values. The VaR values were determined using all three methods at different confidence interval and their backtesting was performed. The study found that the Monte Carlo Simulation approach outperforms other methods for estimating VaR in risk-averse trading scenarios with lowest QPS. It also underscores the need to consider individual contract risks within a portfolio, suggesting that diversification alone might not suffice in mitigating those risks.