Journal of Energy Markets

Risk.net

Forecasting natural gas price trends using random forest and support vector machine classifiers

Francisco Castañeda, Markus Schicks, Sascha Niro and Niklas Hartmann

  • The prediction of future trends in the TTF natural gas market in the Netherlands using financial indicators, here inspired by the results achieved by Khaidem et al in their research work "Predicting the direction of stock market prices using random forest" on stock prices listed on the NASDAQ and the Korean Stock Exchange markets.
  • A feature engineering research, extending the initial set of financial indicators with potential natural gas price drivers, such as weather, seasonality, electricity prices, coal prices and natural gas storage levels in Europe.
  • The proposal of an alternative prediction target, transforming first the original gas price time series using a lag-1 difference between consecutive days, to reduce the effect of long-term trends in the prediction process.

Price forecasting using statistical modeling methods and data mining has been a topic of great interest among data scientists around the world. In this paper, different machine learning approaches are applied to forecasting future yearly price trends in the natural gas Title Transfer Facility market in the Netherlands. The study compares two models: random forest and support vector classifiers. The identification of potential natural gas price drivers that improve the model’s classification is crucial. The forecast horizon was set in a range from 10 to 60 trading days, considering that shorter time horizons have greater importance for trading. The results reflect values up to 85% of the area-under-the-curve score as a reaction of the models to the four different feature combinations used. This invites continued research on the multiple opportunities that these new technologies could create.

To continue reading...

You need to sign in to use this feature. If you don’t have a Risk.net account, please register for a trial.

Sign in
You are currently on corporate access.

To use this feature you will need an individual account. If you have one already please sign in.

Sign in.

Alternatively you can request an individual account here: