Correlation Breaks Down When Most Needed
In a recent newsletter we addressed a question from a subscriber about the cost of being on hourly pricing and how the subscriber thought it was supposed to be cheaper than fixed rate electricity supply. In our response we detailed that most customers should not even consider hourly pricing, with the exception of a small group who can actively manage their consumption profile. Within that small group, the thought is that with optimal management, one can shift load from a high-priced hour to a lower-priced hour, thereby reducing overall electricity expenses. To accomplish cost-effective load shifting, however, you need good information and you need to act on it accordingly. But is there good information available? Or are we laboring under a misconception that there is good load intelligence available and we are just fooling ourselves? And if that is the case, are we really managing our energy budgets or are we just gambling?
Day-Ahead and Hourly Real Time
Day-ahead prices are posted after 4:00 PM the day before they become effective. These prices reflect what market experts think tomorrow’s price will be by way of their buying activity. Suppliers often, but not always, take positions for supply for each hour for the next day. After midnight, the hourly real time market kicks in, and any imbalances between what the suppliers have committed for supply and ultimate consumption is “settled” in the real time market at the hourly rate. The basic idea is that hourly-managing energy users will plan energy usage for the next day based upon day-ahead prices, shifting consumption from higher-priced hours to lower-priced hours. The aforementioned process would work great if day-ahead and real-time hourly prices were the same. Unfortunately, they aren’t – not by a long shot, especially when price parity is most sorely needed.
We investigated how well day ahead and real-time hourly prices correlated. The standard statistical correlation for evaluating how well sets of data correlate is R-squared (also known as the coefficient of determination). If R-squared is 1.0, then there is a direct correlation between the two sets of data; and if R-squared is zero, then there is no correlation between the two sets of data. Statisticians generally state that if R-Squared is above 0.8, it is assumed that the two data sets are correlated.
During most days, when there are no unusual demands on the electric grid (i.e., prices less than 10¢/kWh), R-squared is right about 0.8. And this is what you’d expect, as there is little information that alters the prediction of energy consumption from one day (day ahead) to the next (real-time hourly). R-squared values above 0.8 encompass 80+% of the days. But, on days when the system is stressed (e.g., January 7, 2014), R-squared falls to 0.25, which means there is little correlation between the two sets of data. Since this occurs on only a small percentage of days, it shouldn’t be a big concern. Being right over 80% of the time is a darn good batting average. Unfortunately, all days and hours aren’t equal when it comes to market pricing and your budget. When you look more closely on the days when the system is stressed, you will see that being wrong has a much higher cost – so much so, it can wipe out months (or years) of being on hourly pricing. Also, with day-ahead and real-time hourly having very low correlation a good deal of the time, it illustrates that energy experts have no idea what the price is going to be tomorrow if the weather gets hot, gets cold, or is a polar vortex rolls through, etc.
Hourly vs Hourly+1 Time Series
Because we found no correlation between day-ahead and hourly prices on days when the grid was stressed, we decided to look a bit deeper to see if we could find some correlation so that managers could get some forewarning of what the next hour’s prices were going to be. We looked at hourly prices as a time series. The most common example of time series is the weather: the best predictor of tomorrow’s weather is today’s weather. In our hourly challenge we compared the last hour’s settlement price to this hour’s settlement price. Interestingly, for many days we got R-squared values above 0.6 – not a great correlation but much better than 0.25. While this is better than using day-ahead prices, to take advantage of this one would have to employ a dynamic strategy rather than setting the operation plan the day before (after day-ahead prices are published). Other than those who are doing hourly energy management for sport, very few managers have the time to dynamically alter energy consumption from one hour to the next.
What about the Smart Grid?
Isn’t the Smart Grid going to make being on hourly less risky and more financially rewarding? We need to acknowledge that the Smart Grid has potential – but it isn’t there yet by any means in bringing economic benefit to the masses. Yes, there are a few exceptions of end users who are using sophisticated software management strategies to respond to hourly real time prices, but they are dwarfed by the number of end users who are on hourly pricing. Just because the Smart Grid has future potential, it doesn’t translate into reducing your energy budget or financial risk today.
Successful suppliers do not assume the risk of an hourly product, so why should you? If they can’t hedge their position, they take a pass on that opportunity. Take a good look at the risk you are assuming by being on an hourly product versus the ultimate reward you could achieve. If it was your money, would you be placing that bet?