By James Carson

RisQuant Energy

JBCarson@RisQuant.com

Detailed tables that crashed the blogging editor can be found on a pdf on my website.

http://www.RisQuant.com

Issue. The question of forward bias in the power markets has been probed by analysts, managers and traders for years. The answer goes to the heart of power market efficiency. While bias has been noted many times, questions about the statistical significance of the bias persist.

This article explores that bias with a focus on statistical significance for four disparate locations for Day Ahead (DA) versus Real Time (RT) Locational Marginal Price (LMP) in each of these ISO/RTOs: PJM (WH: Western Hub), ISO-New England (CT Zone), Midwest ISO (IL Hub) and NY-ISO (NY-G, Hudson Valley). 

Summary. There is evidence for slight systematic DA versus RT bias in the LMP markets examined. Overall, the LMP markets examined are “efficient” in the sense that there is very little systematic bias.

Highlights:

• CT and NY-G showed overall long term bias at the p<0.01 level.
• The DA market averaged $2.02 over the RT for CT and $2.22 for NY-G.
• IL showed bias at the p<0.10 level, with DA exceeding RT by $0.84, while WH showed zero bias.
• The individual months by years showed scant significance anywhere.
• The months overall and years overall showed little significance anywhere.

WH’s zero bias can be attributed to the extensive analysis that has been done in that market as a consequence of its prominent role as a forward trading vehicle. In part, this confirms the superior transparency of the PJM Western Hub market.

Method. Most analyses compare the average DA with the average RT overall by various time slices using an equality of means test. This analysis, however, tested whether the mean pairwise difference between RT and DA prices significantly differed from zero. This variance reduction technique significantly sharpened the analysis.

Analysis Procedure: Examined prices were daily mean “on peak” LMPs using the NERC definition for the Eastern Interconnect. That is: hour ending 8:00 through hour ending 23:00 for Monday through Friday while excluding six holidays.

• The pairwise daily differences were averaged for each month, always beginning with the first January in the year after the DA market had been started.
• The pairwise daily differences were averaged across the years by month and across the months by year.
• Estimated the standard deviation for each period.
• Estimated the standard error of the mean for each period (sdev/sqrt(n)) where ‘n’ equals the number of on peak days in each period.
• Estimated the T-Statistic for each period. This is the pairwise mean divided by the standard error of the mean.