Energy-Trader Perspective

Energy trading is conducted by asset-trading companies that use production, demand, and price forecasting to optimize the revenue created from energy production. Aside from production forecasts, energy traders analyze pricing and demand trends in order to determine the ideal strategy for bidding in the energy market (for solar and other sources). Additionally, energy traders assume responsibility for produced energy and inherit any risks associated with trading an asset on the market. Risks include revenue losses via under – and overcommitment of projected production (see Figure 14.2).

Total revenue is a function of the forecast/observed energy in addition to the DAM and RTM LMP (Luoma et al. 2012):

R = Eforecast-LMPdAM P (EObs. ^forecast) *LMPRTM (14.6)

Since there is no deviation penalty for producing an inaccurate day-ahead forecast, revenue is generally maximized by selling more energy in the

market with the higher LMP. Thus, if energy traders can accurately predict day – ahead demand and price, revenue-maximizing bidding strategies can be devised.

However, incorrect forecasts can cause management problems for the balancing authority. To incentivize accurate forecasts, a penalty can be implemented for inaccurate ones (a “deviation penalty”). When deviation penalties are implemented, optimal bids are driven toward expected power generation (Botterud et al. 2012). Overall, this leads to significant reductions in revenue. Equation 14.6 is extended to monetarily penalize incorrect forecasts:

R = Ef-LMPdam + (Eg – Ef) – LMPrtm – DEV• |E0 – Ef| (14.7)

where DEV is the deviation penalty rate. To be effective in discouraging speculation, DEV should be larger than either the RTM or the DAM price. With a deviation-penalty rate equal to twice the maximum of the RTM or DAM, Figure 14.3 shows that revenue is always maximized by bidding into the DAM with a perfect forecast.

The deviation penalty is conducive to illustrating energy-trader behavior. When the forecast is incorrect, the monetary outcome depends on the magnitude of the error and the price ratio. For instance, if LMPrtm < LMPdam (price ratio < 1), up to 80% of the maximum revenue can be obtained with overpredictions of up to 20%. In this situation, a trader may wish to bid into the DAM using an exceedance probability at a high b-level (see Section 14.2). Selling energy using high b-levels increases the likelihood that the observed production will fall below the bid, in which case the trader is required to compensate by purchasing energy in the RTM. However, since the price ratio is less than 1,

compensation energy can be procured at a net profit (excluding the penalty). Similarly, if LMPrtm > LMPdam (price ratio >1), procuring energy in the RTM is expensive and production overpredictions can be costly. For a price ratio of 5, an overprediction of only 10% will reduce total revenue to 0. Underpredictions of up to —20%, however, still yield a profit (see Figure 14.3). A trader therefore bids into the DAM at a low b-level to minimize the chance of overprediction.

In general, energy traders are interested in solar-forecasting components similar to those of interest to the ISO and utilities. For the DAM and RTM, these include a mean-power forecast and a characterization of uncertainty. However, for energy trading, uncertainty is especially important and is often expressed not by confidence bounds but by exceedance probabilities (Pp). In order to effectively devise a bidding strategy, energy traders must also be able to accurately predict the price ratio. For this reason, they are more interested in intrahour solar variability than are other stakeholders. Specifically, for high penetrations of solar energy, localized spikes in aggregate production on a node can drive prices down. Occasionally, production spikes can be so large that significant congestion occurs on the grid and the LMP becomes negative. If the energy trader can predict these rapid changes in produc­tion, LMP fluctuations can be predicted and result in large profits. Spatially, similar strategies are employed. Spatial variability for this application is the change in cloud cover over a node that affects average solar-power production. Coupled with an electricity-transmission model, information on the spatial variability of available irradiance can help the trader effectively forecast where energy surpluses and deficiencies are likely to occur. In this way, energy prices can be better predicted

and bidding strategies updated. Table 14.2 summarizes the basic solar-forecasting components that are desirable for energy traders.