Predictive Modelling Electricity Prices for Short-Term and Long-Term Horizons, the case of the Czech Republic

Accurate forecasts are a crucial aspect of many industries such as electricity and energy markets for which this paper studies. While many markets have gone through turbulent time, this study aims to apply Autoregressive Integrate Moving Average (ARIMA), Support Vector Regression (SVR) and Long-Short Term Memory (LSTM) models to predicting the daily wholesale electricity prices from the Czech day ahead market. This thesis employs the use of a rolling interval approach to predicting across the year 2022 for short-term horizons and long-term horizon, in which the short-term horizon is defined as the next day’s price, while the long-term horizon being the forecast for the multi-step of the next month. A variance reduction method of log transforming the data is used, while models are also used with the un-transformed data. As a result, Support Vector Regression Model outperformed the other models across both time horizons, while the ARIMA and LSTM model provided closely comparable results on average across the year.