Day-Ahead Market Prices of Electricity and Economic
Fundamentals: Preliminary Evidence from New York City
Kevin Forbes*Associate Professor,School of Business and EconomicsCatholic University of AmericaWashington, DCforbes@cua.edu
Ernest M. Zampelli Professor, School of Business and EconomicsCatholic University of AmericaWashington, DC
To neoclassical economists, prices are an essential tool in ensuring efficient resource allocation. Indicative of this, McDermott (2012) in a paper entitled, “The Regulatory Dilemma: Getting over the Fear of Price” makes the point that societal welfare is unlikely to be maximized when price signals are repressed. For example, a necessary condition of not wasting resources in the generation of electricity is the equalization of marginal cost across generating stations. It is almost inconceivable that a regulated utility would achieve this condition. However, profit maximization by firms in a competitive market will give rise to marginal cost equal to price and thus marginal generation costs will be equalized across generating stations.
Another example of where the use of the price mechanism is probably important is the goal of minimizing the cost of reducing carbon emissions in the power sector. Achieving this goal requires that marginal abatement costs be equal across all carbon sources within the power sector. If firms are profit maximizers and the carbon market is perfectly competitive, there is reason to believe that this result would be realized – firms would find it profitable to reduce carbon emissions until marginal abatement costs equal the price of carbon. With a uniform carbon price within the power sector, marginal abatement costs would therefore be equalized across generating units within firms and also across all firms and thus total abatement costs would be minimized, a result that would only occur by happenstance when policy makers mandate renewable energy quotas.
While markets may “work” in abstract theory, the California power crisis of 2000/2001, the oil price spike of 2008, and the 2008/2009 global financial crisis have significantly undermined confidence in using markets to allocate resources. A recent example of this aversion to using price to address important resource allocation issues is Pope Francis’ high degree of skepticism regarding using carbon prices as a tool to manage reductions in carbon emissions (Pope Francis, 2015, paragraph 171)
Forbes and Zampelli (2014) presented a novel test of electricity market informational efficiency. If day-ahead markets for electricity are efficient, then the prices in these markets will reflect the load forecast produced by the system operator but will also reflect the information and insights of market participants. For example, if the market is efficient, the day-ahead prices will reflect the impact of sporting event on electricity demand even if the system operator has not factored in the event into its load forecasts. If prices do accurately reflect expected demand, then a measure of the day-ahead price normalized for fuel cost should actually be useful in predicting load. Support for this hypothesis was presented using data from the PGE aggregation zone of the California ISO. In this paper, the hypothesis is tested using data from New York City. The preliminary results are consistent with the view that there are advantages in using markets to allocate resources. The results also indicate that there is considerable room for improvement in forecasting electricity loads.
Day-Ahead Load Forecasts in New York City
Electricity generation and transmission for New York City and the other zones in New York State are coordinated by the New York Independent System Operator (NYISO). This is accomplished largely by administering both a day-ahead and real-time markets for electricity. NYISO also posts a day-ahead load forecast for each of New York State’s eleven zones (http://www.nyiso.com/public/index.jsp)
The day-ahead load forecasts for New York City were evaluated over the period 6 August 2009 through 30 June 2013. With a persistence forecast as a reference, the mean-squared-error-skill-score (MSESS) of the forecast is 0.28, indicating that NYISO’s day-ahead forecast is superior to a persistence forecast that myopically projects the level of electricity consumption in period t to be equal to the consumption level in period t-1.[*] The root-mean-square-error (RMSE) of NYISO’s forecast over the sample period is 3.61 percent of the mean level of load. These metrics indicate a seemingly respectable level of forecasting performance. However, visual inspection of the data indicates that were a number of instances in which the forecast error exceeds 1,000 MWh (Figure 1).
For a forecast to be considered optimal, the forecast errors should have the property of white noise, i.e. the error should not be systematic. Consistent with the findings reported by Forbes and Zampelli for NYISO and eleven other electric power control areas (2014), the errors in NYISO’s load forecasts for New York City do not have the property of white noise but instead have a significant diurnal pattern (Figure 2).
Portmanteau (Q) tests for white noise were conducted for lags 1 through 100, 120, 144, 168, 192, and 200. In all cases, p-values were equal to 0.0000, thereby strongly rejecting the null hypothesis of white noise.
If one defines the forecast error as actual load minus forecasted load, it is reasonably clear that the real-time electricity market in New York City has an asymmetric response to the forecast errors (Figure 3).
Specifically, there is little apparent systematic response to negative errors but positive errors, i.e. actual load exceeding day-ahead forecasted load, appear to be a contributor to significant differences between the real-time and day-ahead prices. On average, the real-time price exceeded the day-ahead price by about four dollars per MWh over these hours with the price differential exceeding $100 per MWh during 378 hours. The price differential was larger than $500 per MWh in 38 hours and over $1000 per MWh in six hours. In our view, these price spikes represent instances in which the reliability of the electric power system was challenged because of an inadequate forecast.
Preliminary Out-of-Sample Forecasting Results
A time-series model of actual hourly load was estimated using data over the period 9 August 2009 through 30 June 2013. Explanatory variables in the model include NYISO’s reported day-ahead hourly forecasted load, a measure of the day-ahead hourly zonal price, day-ahead measures of forecasted weather, and measures that reflect seasonality. With the aim of achieving white noise in the residual error terms, the model also includes a number of ARMA disturbance terms.
The preliminary results are encouraging. A number of the key explanatory variables are highly statistically significant. For example, the coefficient corresponding to the day-ahead price measure is positive and has a p value less than .001. The model’s only apparent shortcoming is that white noise in the residual errors terms has yet to be achieved. Because of this shortcoming, we have focused our evaluation of the current model specification on very short term forecasts. Specifically, our first test consists of rolling eight hour ahead forecasts. The second test consists of rolling four hour forecasts. The third test consists of rolling two hour forecasts. In each of these cases the forecast is calculated one hour before the first forecast period.
The preliminary model was tested using out-of-sample hourly data over the period 1 July 2013 – 31 December 2014. The RMSE of NYISO’s day-ahead forecast over this period is 3.17 percent of the mean load. The RMSE for the rolling eight hour forecast is 2.3 percent. The RMSE for the rolling four hour forecast is about 1.9 percent. The RMSE for the rolling two hour forecast is about 1.55 percent. A comparison of Figure 4 with Figure 5 provides a visual sense of the improvement in forecast accuracy.
Summary and the Direction of Future Research
This research update has presented further evidence that day-ahead market prices of electricity reflect economic fundamentals. Out-of-sample evidence has also been presented that the accuracy of NYISO’s load forecasts for New York City can be radically improved. Application of the methods employed here to other electricity control areas has obvious implications for both efficiency and electric power reliability but would seem to require that decision makers overcome their fear of price.
Forbes, Kevin F. and Ernest M Zampelli (2014) “Do Day-Ahead Electricity Prices Reflect Economic Fundamentals?: Evidence from the California ISO, The Energy Journal, 35 (3): pp 129-144
McDermott, Karl A. (2012). “The Regulatory Dilemma: Getting Over the Fear of Price,” The Electricity Journal 25(9): 6–13. http://dx.doi.org/10.1016/j.tej.2012.10.011.
Pope Francis (2015) Laudato Si’ : On Care For Our Common Home. Available at http://w2.vatican.va/content/francesco/en/encyclicals/documents/papa-francesco_20150524_enciclica-laudato-si.html
Wilks, Daniel S. (2011) Statistical Methods in the Atmospheric Sciences, 3rdedition, Waltham MA: Academic Press
* Dr. Forbes is a USAEE Distinguished Lecturer.
** If electricity demand in say, hour 12, was 5,000 MWh, then the persistence forecast for hour 13 would be 5,000 MWh. One can calculate a MSESS with a persistence forecast as a reference. A perfect forecast would yield a MSESS equal to one. A forecast equivalent in accuracy to a persistence forecast would yield a MSESS equal to zero while methods that produce forecasts inferior to a persistence forecast would yield a MSESS that is less than zero. For more information on MSESS see Wilks (2011, p. 327-329).