The larger companies are winning more business. But how will
they fit into a restructured industry?
Put 45 energy service companies (ESCos) into a $1-billion market, and they...
selects customers that are non-coincident with class peak. (Remember that class peak is used as a proxy for pre-existing LSE load.) Strategy 3 successfully selects customers who, while having a relatively poor peak-day load factor, are generally non-coincident with the class peak. If these customers can be identified, then it may prove advantageous to remove them from the profiled class load shape.
What the Experiments Show
Tables 1 and 2 report the results for all of the experiments for January and July simulations. The tables show the only strategy that produces consistent outcomes across seasons was Strategy 1 applied to rates GD and G. Recall that customers are targeted for removal based on January data, the strategy of removing high load-factor customers from these general service rates consistently produced a subgroup with a higher load factor than the original profile. It also degraded the load factor of the profile when recalculated after removal of customers with high load factors.
Intuitively this result makes sense. As a "catchall" category, general service rates apply to customers that do not "fit" other categories. As such, there is greater diversity with the sample and an increased chance to cull customers with much different load patterns than is suggested by class profile. Moreover, since general service customers can be sizable, it is likely that the incremental cost of interval metering may prove small relative to the benefits of culling this customer.
The experiments yield another result that may appear surprising. For the January data only, any strategy applied to any of the four of the rate classes tended to lower the load factor of the customers remaining in the load profile. Mathematically, it is possible to remove a sub-class of customers with a poorer-than-average load factor from a larger sample and still find that the load factor of the remaining customers degrades. Overall, the January data suggests that any cherry-picking strategy will produce a poorer load profile for remaining customers. This tendency was not evident in reviewing the July data, however.
Of course, the cost implications of migration will depend on cost causality rules that likely will vary across control area. However, customers with high load factors are generally thought to be cheaper to serve than low load-factor customers. This fact likely will remain true as we move forward with retail choice.
If effective cherry-picking develops, then one question remains: How quickly can load profiles of the remaining customers be updated? The update is necessary so that the load profile accurately reflects the usage patterns of the remaining customers. If adjustments are not made, then the entity designated as supplier of last resort will have to make up the difference between estimated cost and actual system consumption.
Are the findings reliable? Are they over- or understated?
The variation in the population about sample means should not surprise anybody. The degree of variation is the issue here. In fact, the use of Central Vermont's load research may well underestimate the problem (other things being equal) because of the availability and penetration %n6%n of off-peak and TOD rates on