Managing electricity price modeling risks

Managing electricity price modeling risks

In the paper ‘ Managing electricity price modeling risk via ensemble forecasting: The case of Turkey ’, Dr Ezgi Avci Surucu and her co-authors focus on how ensemble modeling is a solution for managing market price risks in electricity day ahead markets.  The paper offers important implications to theory and practice of surveillance of complex auction markets. It systematically examines the interplay of different informational and strategic factors in multi-unit auction markets that are characterized by risks in transparency. From the policy perspective, Avci’s research shows that integration of big data analytics and domain-specific knowledge improves decision-making in surveillance of complex auction markets.

Ezgi Avci Surucu defended her dissertation in the Senate Hall at Rotterdam School of Management,Erasmus University Rotterdam on Friday 25 May 2018. Her supervisor is Prof. Wolf Ketter and his co-supervisor is Prof. Eric van Heck. Other members of the Doctoral Committee are Prof Derek Bunn (LBS), Dr Jan van Dalen (RSM), Ronald Huisman (ESE), Shmuel Oren, Haas School of Business.

About Ezgi Avci Surucu

Ezgi Avci Surucu (1983) obtained his Bachelor’s degree in statistics and her master’s degree in Industrial Engineering from Middle East Technical University, Ankara, Turkey. She is performing a consulting role for several years for Turkish industry and government providing recommendations about measurement, analysis and improvement of the firms/organizations' management systems. Her recent research interests are auction market design, big data analytics, energy information systems, data-mining methods, machine learning algorithms and monitoring and surveillance of markets. During her PhD Ezgi was a visiting research fellow at London Business School and the Oxford Institute for Energy Studies and has presented her research at IAEE International Conference, European Conference on Operational Research, Commodity and Energy Markets Conference, International Symposium on Environment and Energy Finance Issues, Operations Research and Industrial Engineering Congress, Euro Working Group for Commodities and Financial Modelling Conference

Paper Abstract

There are two ways of managing market price risk in electricity day ahead markets, forecasting and hedging. In emerging markets, since hedging possibilities are limited, forecasting becomes the foremost important tool to manage spot price risk. Despite the existence of great diversity of spot price forecasting methods, due to the unique characteristics of electricity as a commodity, there are still three key forecasting challenges that a market participant has to take into account: risk of selection of an inadequate forecasting method and transparency level of the market (availability level of public data) and country-specific multi-seasonality factors. We address these challenges by using a detailed market-level data from the Turkish electricity day-ahead auctions, which is an interesting research setting in that it presents a number of challenges for forecasting. We reveal the key distinguishing features of this market quantitatively which then allow us to propose individual and ensemble forecasting models that are particularly well suited to it. Our findings support the additional benefits of ensemble forecasts especially according to an ex-ante (more realistic) decision making setting and in line with the previous findings indicating ensemble modeling is less uncertain and more accurate than the ex-ante best individual model.

Policy implications

  • For energy regulators and policy makers, using ensemble models can be useful to manage electricity price modelling risk for ex-ante policy impact assessment and lead to better policy decisions. Electricity price forecasts are used by energy regulators as one of the main input variables for ex-ante policy impact assessment (Shahidehpour et al., 2002). In forecasting, uncertainity is reflected in the forecast error and the source of risk arises from the unobservability of full information set underlying the individual forecasts which could be differently affected by statistical properties of the related price series (Timmermann, 2006). Therefore using ensemble forecasting could mitigate this risk related to decision making of a policy maker (Bunn, 1985).
  • Considering fractal dynamics of price could improve decision making of policy makers. We find that for both individual and ensemble models, in most of the cases T3 (night tariff time zone) has the highest MAPE values. This can be explained by the lower predictability level of this price series compared to T1 and T2, and the stronger impact of seasonality factors on demand during the night. Thus if fractal dynamics of price is prominently different for some time zones during the course of a day, ensemble modeling is less risky than individual models in terms of the risk of selection of an inappropriate individual forecasting model. Further, choosing ensemble models based upon the fractal dynamics of each time zone could improve policy maker's forecast accuracy.
  • Energy regulators could enhance predictability level of prices, especially for off-peak load periods by increasing transparency level of the market through disseminating data on primary resource based available installed capacity and planned generation schedules. We find that models with exogenous variables generally perform better than models without exogenous variables. This finding is in line with the previous literature and arises because of the high cross-correlation between demand, margin and price. Further this result also shows the appropriate selection of our input variables. On the other hand we find that T3 price series has lowest level of predictability and no long term correlation indicating that marginal bidders bid at their marginal costs (Sapio, 2004). This means if an energy regulator wants to enable power agents to forecast the prices accurately for off-peak hour, it needs to publish prior information on primary resorce based available installed capacity (the active power capacity that a generation unit can provide to the system) and final daily production program (firmlevel) in order to give a signal of the possible future supply stack for each hour and technology of the marginal generator.
  • Download the paper on Elsevier Energy Policy
  • Download the dissertation