The Day-Ahead Energy Market Forecasting in Russian Federation

The Day-Ahead Energy Market Forecasting in Russian Federation: a Case-Study of Siberia Alexander Filatov (Irkutsk State University, Far East Federal University, alexander. filatov@gmail. com) Evgenya Smirnova (Irkutsk State University, smirnovevgen-91@mail. ru) The Institute for East and Southeast European Studies, Regensburg, 5 -th of April, 2016

Introduction. Russian Reforms of Electricity Market a) Vertically Integrated Monopoly (Before 2003) b) Competitive Unbundled Structure (From 2003) Competitive Generation: 1. Territorial Generating Companies. 2. Wholesale Generating Companies. 3. Ros. Atom (Nuclear Plants). 4. Rus. Hydro (Hydro Plants). Transmission: Government-Granted «Regulated» Transmission Monopoly Competitive Distribution and Sale a) b)

Introduction. The Structure of Electricity Market Wholesale electricity and power market DAM Retail Plants and other consumers BM BCM DAM – Day-Ahead Market BCM – Bilateral Contract Market BM – Balancing Market Population

Introduction. Auction Clearing Bid Prices Electricity bought at DAM Unclaimed Supply Buyers Bids Equilibrium Price Unsatisfied Demand Sellers Bid Quantities The Price and Quantity Forecasting at DAM Provides: 1. The effective regimes of the power plants work. 2. The Improvement of the generating companies business planning. 3. The best option choice between trade operations at DAM, long-term bilateral contracts, and forward contracts that allow risk hedging.

The Data (14. 09. 2007 -31. 12. 2015, 3025 obs. ) Mean Std. dev. Min Max. DAM Price (RUR per MWh) 623 175 142 1249 DAM Quantity (thous. MWh) 350 161 18 662 Average Temperature (0 C) +3 15 -36 +35 Light Day Duration (min. ) 732 186 461 1009 Gas Price (RUR per 1000 м 3) 3931 1175 1564 8736 Oil Price (RUR per barrel) 2925 655 1029 4112 Dollar exchange rate (RUR) 34, 00 10, 68 22, 69 72, 06 Euro exchange rate (RUR) 44, 31 9, 70 33, 21 83, 87 GDP (billion RUR) 13849 3358 8135 19589

The Electricity Price Features 1. 2. 3. 4. High volatility. Spikes. Autocorrelation. Seasonality.

The Daily Price Forecasting at DAM Regressors: • t - time; • - dummies for days of week; - dummy for holidays; • - share of the working turbines at Sayano-Shushenskaya power station • - average day temperature; - light day duration; • - oil and natural gas prices; • - dollar and euro exchange rates; • - GDP of Russia The Regression Model: The Autoregression Model AR(1):

The Distribution of Errors Pearson Criterion: The Normal Distribution: = 225, 76 >> = 33, 41. The Logistic Distribution: = 77, 73 > = 33, 41. 1 2 3 9 10 18 19 20 Empirical 0, 0048 0, 0069 0, 0110 … 0, 2376 0, 1761 … 0, 0041 0, 0007 Logistic 0, 0019 0, 0039 0, 0080 … 0, 1785 0, 1657 … 0, 0012 0, 0006 0, 0003 Normal 0, 0008 0, 0026 0, 0076 … 0, 1580 0, 1508 … 0, 0004 0, 0001 0, 0000 -150 -100 -50 f (emp) 0 50 f (norm) 100 150 f (logist) 50 100 f (logist) 150

The Distributed Lagged Model. Koyck Lag Structure The general distributed lagged model: Let , 0 < λ < 1. Then: The final model: , Stage 1. Elimination of trend, seasonality and real factors except price of natural gas Stage 2. Koyck lag structure for the price of natural gas.

The Daily Quantity Forecasting at DAM The Model here – dummies for the half-year – DAM price. Logistic function for the long-term forecast of the DAM quantities: Quantity dynamics at DAM

The Hourly Price Forecasting at DAM Price, RUB per MWh 550 MA(3): MA(5): 500 450 400 350 300 250 200 150 Electricity Price in 2010, Fact and Forecast Hour Dummies Hour 1 (00: 00– 01: 00) Hour 2 Hour 3 Hour 4 Hour 5 Hour 6 – Hour 18 MA(3) – 0, 24* 3, 22* 13, 6** 28, 18** 41, 64** 54, 74** MA(5) 1, 31* 6, 83* 15, 56** 26, 26** 37, 11** 52, 79** Hour Dummies Hour 19 Hour 20 Hour 21 Hour 22 Hour 23 MA(3) 44, 54** 29, 10** 14, 72** 6, 16* 2, 10* MA(5) 39, 43** 29, 23** 17, 49** 8, 15** 2, 49*

Extrapolation Based on the Maximum Likeness Model Initial time series: The extrapolation for the period The initial vector: The likeness measure: The likeness function: The maximization: : Extrapolation: , Error: ;

Optimal Solution of the Maximum Likeness Method Spikes Elimination: MA(5), Stationarity: ε(t) 14 12 R 2 = 0. 4789 10 8 6 4 2 0 24 48 72 96 120 144 168 192 216 240 264 288 312

Danke für die Aufmerksamkeit!