Analýza predikce cen a tržního ohodnocení přeshraničních kapacit pro regulační zálohy

Abstract

In recent years, the balancing capacity market in Germany has undergone several changes in framework conditions which influences bidding behavior. Additionally, due to other factors such as the energy crisis in 2022, the prices of balancing capacity increased in level and variability. Thus, the importance of predicting balancing capacity prices increases, for TSOs for budgeting and to assess market design on the one hand and for bidders to maximize profits on the other. Therefore, the goal of this thesis is to develop accurate forecasts of balancing capacity prices for aFRR and mFRR. The prediction takes place on two forecast horizons: short-term at gate closure time and medium-term 28 days ahead. Additionally, the developed multiple regression models were adapted to predict the market value curve for the exchange of balancing capacity as envisaged by the inverted market-based method in the electricity balancing guideline. Market value indicators were chosen as points of this curve to forecast, and the model was tested for aFRR on the border between Austria and Germany. Overall, the models outperform the chosen autoregressive benchmark forecasts for nearly all forecasted variables despite high autocorrelation in the balancing capacity prices, especially for the short-term price prediction. Thus, the short-term price prediction of all analyzed markets is very accurate with positive aFRR yielding the best accuracy. The prediction on the medium-term horizon is less precise but still shows good accuracy. However, all models cannot accurately predict extreme prices. Forecasting the market value indicators yields a notably lower accuracy than the price prediction due to lower autocorrelation. Best results were achieved when predicting the market value for no exchange of balancing capacity and especially the volume of crosszonal capacity for which the market value reaches zero. For most predictions, the direction of economic exchange is forecasted accurately. Higher forecast accuracy for negative than for positive aFRR was found for the market value indicators due to higher autocorrelation in negative direction. The most important regressors of all models are autoregressive factors, bid curve parameters for balancing capacity (like maximum and minimum prices) and unavailabilities (especially of hydro power plants). For the medium-term price prediction, time values such as month or time of day and clean spreads gain importance. The market value indicator models heavily rely on parameters of demand and excess power for Austria and Germany as well.

In recent years, the balancing capacity market in Germany has undergone several changes in framework conditions which influences bidding behavior. Additionally, due to other factors such as the energy crisis in 2022, the prices of balancing capacity increased in level and variability. Thus, the importance of predicting balancing capacity prices increases, for TSOs for budgeting and to assess market design on the one hand and for bidders to maximize profits on the other. Therefore, the goal of this thesis is to develop accurate forecasts of balancing capacity prices for aFRR and mFRR. The prediction takes place on two forecast horizons: short-term at gate closure time and medium-term 28 days ahead. Additionally, the developed multiple regression models were adapted to predict the market value curve for the exchange of balancing capacity as envisaged by the inverted market-based method in the electricity balancing guideline. Market value indicators were chosen as points of this curve to forecast, and the model was tested for aFRR on the border between Austria and Germany. Overall, the models outperform the chosen autoregressive benchmark forecasts for nearly all forecasted variables despite high autocorrelation in the balancing capacity prices, especially for the short-term price prediction. Thus, the short-term price prediction of all analyzed markets is very accurate with positive aFRR yielding the best accuracy. The prediction on the medium-term horizon is less precise but still shows good accuracy. However, all models cannot accurately predict extreme prices. Forecasting the market value indicators yields a notably lower accuracy than the price prediction due to lower autocorrelation. Best results were achieved when predicting the market value for no exchange of balancing capacity and especially the volume of crosszonal capacity for which the market value reaches zero. For most predictions, the direction of economic exchange is forecasted accurately. Higher forecast accuracy for negative than for positive aFRR was found for the market value indicators due to higher autocorrelation in negative direction. The most important regressors of all models are autoregressive factors, bid curve parameters for balancing capacity (like maximum and minimum prices) and unavailabilities (especially of hydro power plants). For the medium-term price prediction, time values such as month or time of day and clean spreads gain importance. The market value indicator models heavily rely on parameters of demand and excess power for Austria and Germany as well.