MKonline has issued new consumption normals up to 2020.
A quick overview
- New consumption normals are published
- Nomals are now being dynamically simulated
- Be advised to update our consumption normal every day
How are the assumptions made
Our analysts have carefully gathered data and conducted literature surveys on the expected power consumption development for all price areas in Europe. Based on this MKonline has made and implemented assumptions on expected growth per year in our consumption forecasting system.
Notably, the assumptions are to be understood as expected underlying growth. The simulated normal for 2015 and 2016 will therefore not “hit” the assumed percentage growth perfectly. The reason is twofold:
- Calendar effects: The accumulated effects of weekends, holidays, and leap years vary from year to year. The easiest to understand may be the leap year effect. It will add, ceteris paribus, 1/365=0.27% to the annual growth in 2016. Going on, comparing 2017 to 2016 you will “lose” an 0.27% annual growth. Hence the simulated annual growth in percent will by definition deviate from the assumed underlying growth assumption.
- Learning from recent history: We collect and adjust the consumption model to actual numbers as we go. If, for instance, the consumption model (without adjustment) deviates from the recent received actuals, a Kalman filter adjusts the forecasts to fit the latest received actuals best. This procedure is also included for estimation of the normals. Further, the “recent” adjustment is kept constant for 60 days, and then linearly phased out over the next 30 days.
In layman words, the Kalmanfilter will help pull the forecasts, and the normal simulation, back on track during periods where consumption grows/shrinks faster than we assumed ex ante, or we experience exogenous shocks.
Consumption normals are updated each day
We recalculate and update the normals every day. That is of course not beaches we change our mind with regard to assumed underlying consumption growth. It is because we dynamically adapt the normals to reflect the latest received actual numbers. As a rule of thumb: the normal for the period “today” until 90 days ahead will experience minor adjustments each day. For this reason you should renew your consumption normal form MKonline each day.
We will keep you posted each time we will change the assumptions on underlying assumption growth.
Note: As I write this. 26 February 2015, the new consumption normal for Turkey is not yet simulated.
How is the normal simulated?
The consumption normal is simulated by running more than 30 weather years through our proprietary consumption models. One for each price area, on hourly level, from 2006 until 2020. Note that using the average of each weather parameter as input would not work. This is because the consumption model is nonlinear and contains input related to spatial definition of each model area.
One has to distinguish between preceding normal and expected normal ahead.
- Preceding normals: For the preceding period we take into consideration the received actual consumption. The deviation between actual consumption and simulated normal consumption indicates how “uncommon” the weather has been, in terms of MWh over a period.
- Future normal: Looking ahead we include our assumptions on expected underlying consumption growth, instead of actual received consumption numbers. In the years to come the normal consumption therefore indicates what we should expect the consumption to be, given no prior information on the weather.
Note: The normal for the current year is partly simulated on received numbers so far and partly on assumptions for the remaining part of the year.
A word of caution:
Normal consumption is also referred to as temperature adjusted consumption. Please be advised not to directly compare MKonline normals to that of third parties. MKonline adjusts for a wide range of weather parameters which enter the consumption model, not only temperature. We also include and simulate calendar effects and exploit the information embedded in received consumption data until “today”. At the end of the day, the model properties and weather period over which one simulates the normal will determine the result.