Customers' price elasticity in delivering event-based flexibility in France

2024-01-02

2024-04-30

Electricity Consumption in kWh per half-hour: We will analyze the electricity consumption (kWh per half-hour) of participating customers during and around Eco-sessions.

We will randomize customers to recieve different incentives for curtailing energy demand.

To ensure the robustness and applicability of our findings, we block units into experimental blocks of 5 units. We first find exact matches on the categorical variables (geographic locations, housing type, whether or not they answered the energy consumption survey) and then find the closest match on the numerical variable using Mahalanobis distance. Blocking begins by creating a measure of multivariate distance between all possible pairs of units. Within each category, the algorithm optGreedy calls an optimal-greedy algorithm, repeatedly finding the best remaining match. The optGreedy algorithm breaks ties by randomly selecting one of the minimum-distance pairs. Once the groups of 5 closest units are created, we randomly allocate a number from 1 to 5 to each unit within each group. Finally, we recode units 3 to 5 to be in the low incentive group. If there are more than 5,000 in units 1 and 2 (medium and high), I randomly select the extra groups and allocate all units in these groups to be in the low incentive group. Some groups do not have 5 matches. These units are randomly allocated to one of the 5 groups.

The following variables are used to create the groups:
- Geographic locations at the sub-national level
- Housing type
- Average annual consumption - we will create a measure of distance between all possible pairs of units
- Whether or not they answered the survey about energy consumption patterns

Household/customer

No

We will invite 157k eligible customers to participate in Eco-sessions. We will divide the customers that accept the invitation into 3 groups: 10k, 5k, 5k.

157k eligible customers with half-hourly data on energy use prior, during, and after the trial.

10k
5k
5k

In this section, we use the standard errors we obtained from the differences-in-differences analysis of UK Saving Sessions 2022-2023 in Jacob et al. (2023) to estimate the Minimum Detectable Effect Size for Eco-sessions in France. We first give a rough estimation of how the change in group size affects the standard errors of the analysis, and then calculate the MDES based on a power of 80% and significance level of 5%. Based on treatment groups of 5,000 customers, we calculate a MDES of 0.08 kWh per half-hour. We calculated this MDES by “scaling up” the standard error of the difference in differences between Octopus and Bulb customers in Table AT.4 from by Jacob et al. (2023) based on the smaller sample sizes (5,000 per treatment group) involved in the French Eco-sessions. We then multiplied these scaled up standard errors by 2.8, according to the rule of 2.8. This MDES (0.08 kWh per half-hour) is lower than the average demand reduction of 0.1 kWh per half-hour for signed-up customers and 0.17kWh per half-hour for opted-in customers during Winter 2022-2023 Saving Sessions in Great Britain identified in Jacob et al. (2023). Note that the standard errors we have used come from a difference-in-differences (DiD) estimator used in Jacob et al. (2023). We expect the RCT design will produce smaller standard errors than a DiD. However, we have two reasons to expect a smaller treatment effect. We will compare different treatment incentive levels in the main analysis on electricity consumption reduction instead of a simple treated versus control group. In addition, the French Eco-sessions (seven hours of the day) will be longer than the Great Britain Saving Sessions (which were typically one hour, and at most two). Based on the better design (RCT) but smaller treatment effect, we expect to be sufficiently powered to capture a treatment effect as small as twice the effect found during saving sessions.