Now or Later? Trading Wind Power Closer to Real-time: How Poorly Designed Subsidies Can Lead to Higher Balancing Costs.

• Author: Mauritzen, Johannes

1. INTRODUCTION

   Many deregulated electricity markets, including the common Nordic market, have traditionally relied heavily on a day-ahead market mechanism where trade is organized between 12 and 36 hours ahead of actual delivery. The installation of large amounts of intermittent power sources such as wind power poses serious problems for this type of market mechanism.

   The reason is that wind power cannot be scheduled and can only be forecast approximately. However this forecast becomes better the closer one gets to the time of delivery (Holttinen et al. 2006). Large amounts of wind power are therefore widely expected to lead to a heavier reliance on markets that trade closer to the time of delivery, like the Nordic hour-ahead market called Elbas (Nord Pool Spot, 2012).

   A growing literature has developed around the issue of dealing with intermittency in deregulated power markets. Particularly relevant to this article is Holttinen (2005) who, using a simulation model, estimates that producers could reduce their balancing costs by 30 percent by trading on an hour-ahead market like Elbas as opposed to a day-ahead market. Holttinen et al. (2006) uses two commercial wind power forecasting tools and data from a Finnish wind power farm to calculate a balancing cost of between 0.6 and 0.8 Euro per megawatt-hour (mwh) if power is bid in the market between 12 and 36 hours ahead of delivery. Yet to my knowledge, no empirical studies exist on the effects of intermittent energy and forecast error on the use of markets that trade closer to real time.

   Previous empirical studies investigating the effects of wind power on balancing costs have used total amount of wind power as a proxy for the effects of forecast error (Morthorst (2003), Forbes and Zampelli (2007)). However I will show that the use of total amount of wind power can introduce a potentially severe bias in the estimation since it does not take into account the asymmetric effects of positive versus negative forecast errors. In this paper, I use data on day-ahead forecasted as well as settled wind power in western Denmark in order to create measures of positive and negative forecast errors.

   Figure 1 shows the ex-ante expected relationship between the forecast error of wind power and the probability of trade on the Elbas market. Negative forecast errors--where actual wind power produced is less than that estimated a day ahead--can be expected to lead to a higher probability of trade on the Elbas market. As wind power producers realize that they will not be able to generate as much electricity as they had expected, they have a strong incentive to go on the Elbas market in order to make up for the shortfall. By doing so, they reduce the costs they incur in the balancing market.

   The same logic should apply for positive forecast errors--where more wind power is delivered than forecast. Wind power producers have an incentive to go on the Elbas market to sell the excess electricity in order to avoid incurring balancing costs.

   Yet an empirical estimate of the relationship gives a more nuanced picture, as figure 2 shows. A shortfall of wind power production has the expected result of increasing the probability of trade on the Elbas market. This result provides support to the idea that trading closer to realtime can reduce the balancing costs associated with having large amounts of wind power and other intermittent generation. However, a surplus of wind power has the unexpected effect of reducing the probability of trade on the Elbas market.

   The likely reason for this relationship is that subsidies and regulations intended to encourage investment in wind power have led to a perverse incentive for wind power producers to avoid the Elbas market when they produce surplus wind power. In particular, in 1999 Denmark introduced an incentive covering all non-utility-owned wind turbines built before 2003, where it obligated the transmission system operator to purchase all wind power produced from a given turbine for a 10-year period while freeing producers from incurring balancing costs. In turn the incentive for wind power producers covered by the incentive to trade excess wind power on the short-term Elbas market is essentially eliminated.

   I use linear probability models to estimate the effect of wind power forecast errors on the probability of market trading. The models are specified to allow for a quadratic relationship between forecast errors and probability of market trade. They also allow for a non-linearity at zero--that is, the slopes are allowed to differ between positive and negative forecast errors.

   It might be questionable whether such a simple linear model sufficiently captures the relationship between forecast errors and probability of trade. To give an idea of the appropriateness of the linear model, I also use a simple non-parametric estimation of the effect of forecast errors on the probability of market trade. This provides a visualization of the expected probability of trade given the forecast error. From this, the simple quadratic specification is shown to have a surprisingly good fit.

   To give the results a causal interpretation I rely on the assumptions that the wind forecast error variables are exogenous and independent. Though such assumptions can often be strong, I argue that in this case they are justifiable. I also discuss potential violations of these assumptions and test for their relevance.

   Finally as evidence for the claim that the unexpected negative relationship between positive forecast error and probability of Elbas trade can be explained by subsidy policy, I run a rolling windows regression with updated data through February of 2013. This shows that as the subsidies expired for a large amount of wind power over the course of 2012, the effect of positive forecast errors reversed and became positive, as originally expected.

2. THE NORDIC MARKET, DATA AND METHODOLOGY

   The Nordic market is a good testing ground for the effects of intermittency and forecast error on short-term market trading. The Nordic market is one of the oldest market-based electricity systems, dating back to the Norwegian electricity market reform of 1991, and is generally seen as being well functioning and efficient. For a thorough history and overview see Rud (2009). The Nordic market also has several market mechanisms: a day-ahead market called Elspot, a continuous hour-ahead market called Elbas, and balancing and regulating markets operated at a national level.

   Denmark, which became fully integrated in the Nordic market in 2003, has a relatively long history of feeding large amounts of wind power into its grid. In 2011 wind power made up approximately 27 percent of all electricity production in the country (Energistyrelsen, 2012). About 75 percent of this wind power is produced in the western Denmark price area, consisting of Jutland and Fyn.

   Significant wind power capacity was built up relatively early in Denmark because of generous subsidies. Prior to 2003 a fixed per kilowatt-hour (kwh) tariff was granted. This was lowered from 0.60 Danish Kroners (DKK) per kwh to 0.43 DKK per kwh in 2000 and the guaranteed length of the tariff was changed from 10 years to 22,000 full-load hours. Following Denmark's full entrance into the Nordic electricity market in 2003, the subsidy was changed to a feed-in tariff of 0.10 DKK over the market price. The announcement of this change led to a rush to install turbines in 2002 before the new, less generous subsidy regime came into effect. In 2008 the feed-in tariff was increased to 0.25 DKK following several years of lower-than-expected wind power investment.

   The hour-ahead Elbas market began operation in Sweden and Finland in 1999. Eastern Denmark joined in 2004 and western Denmark joined in 2007. Elbas has later been extended to Norway, Estonia and northern Germany. In addition the market has been linked with the Dutch-Belgian intraday market.

   The timing of trade in Denmark is shown in figure 3. Bids for the day-ahead Elspot market, (1) operated by the Nordic central exchange, Nord Pool, must be received by noon the day before delivery. Producers and consumers submit bids for every hour of the following day and from these bids Nord Pool establishes virtual demand and supply curves. Prices for each hour are determined by the intersection of these curves. In Denmark two price areas exist--Denmark East and Denmark West. Prices in these areas diverge from the Nordic-wide system price if congestion occurs on...