Abstract:
In this thesis, we model the dynamic behaviour of natural gas consumption using continuous-time stochastic models to incorporate their significant advantages over the discrete-time models into the modeling process. In addition to offering a wide set of choices for the drift and volatility terms and yielding analytical solutions for any forecast horizon, continuous-time models can also be used in the pricing of contingent claims depending on natural gas consumption since they enable more reliable forecasts at high-frequency levels. Here, we also document that the per consumer natural gas consumption data exhibit stationarity, strong seasonality, mean reversion, and serial correlation. Hence, we study the application of a One-factor mean-reverting process and stochastic Gompertz diffusion model on the modeling of daily natural gas consumption in Istanbul, Turkey that will also incorporate the empirical observations. In the comparison of their forecasting performances which are tested via the backtesting method at different forecast horizons, we find out that the One-factor mean-reverting process is more advantageous than the Gompertz diffusion process. To illustrate the pricing implications of these models, we price two hypothetical contracts and find out that the results vary from one model to another and hence, the choice of model becomes crucial in the real world.