Virtual Energy Networks for Distribution Constraints: Experimental Evidence from Peer-to-Peer Trading to Soak Up Midday Solar

How to cite

Myers, Erica; Nauze, Andrea La; Menezes, Flavio. 2025. "Virtual Energy Networks for Distribution Constraints: Experimental Evidence from Peer-to-Peer Trading to Soak Up Midday Solar." AEA RCT Registry. AEARCTR-0017484. https://doi.org/10.1257/rct.17484-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Erica Myers* , Andrea La Nauze , Flavio Menezes

* Primary investigator

Registry ID
AEARCTR-0017484
DOI
https://doi.org/10.1257/rct.17484-1.0
First registered
Last updated
Status
On going
Trial dates
2025-05-20 – 2026-06-30
Countries
Private
JEL codes
Q41, Q42, L94, C93, D47, Q48
Registry record
http://www.socialscienceregistry.org/trials/17484

Abstract

Peer-to-peer energy trading—often implemented through virtual energy networks (VENs)—has expanded in Australia, the UK, the Netherlands, Germany, and Japan as rooftop solar adoption accelerates and retail customers seek greater control over their energy use. While P2P trading is frequently promoted as a way for households to arbitrage price gaps between retail prices and feed-in-tariffs, in high-solar contexts such as Australia it also has the potential to serve an important network function: absorbing excess daytime solar generation to reduce reverse power flows, voltage rise, and curtailment at the distribution level. We partner with an established VEN platform to group participating households into “virtual transformers,” each containing a mix of solar and non-solar homes.We randomize into treatment and control groups and virtual transformers. Treated clusters are assigned default trades in which solar producers sell to off-takers at a price between the retail rate and feed-in tariff, effectively splitting the gains from trade generated by that price gap. Control clusters remain in a baseline condition—active on the platform but not engaged in trading—before receiving the intervention at a later date. We also layer dynamic, event-day incentives onto the treated clusters by subsidizing daytime electricity to near zero on forecast high-solar days, enabling us to assess whether stronger, time-targeted incentives further increase local solar absorption and ease distribution-network constraints. Finally, we combine the experimentally estimated demand elasticities with engineering models of distribution networks to quantify how increased daytime load shifting would affect hosting capacity, reverse-flow frequency, and the long-run need for transformer and feeder upgrades.

Intervention

The study evaluates a peer-to-peer (P2P) electricity trading intervention implemented through an existing Virtual Energy Network (VEN) platform. Participating households include electricity off-takers (primarily non-solar households) and solar producers with rooftop photovoltaic systems. Participants are grouped into geographically proximate “virtual transformers,” each containing a mix of solar and non-solar households intended to mimic local distribution-level constraints.

The core intervention enables default P2P electricity trades within each virtual transformer. For treated participants, electricity demand from off-takers is first matched to local solar generation from designated producers within the same virtual transformer, following a predetermined hierarchy of primary, secondary, and tertiary trading partners. Trading prices are set between the prevailing retail electricity rate and the feed-in tariff, splitting the arbitrageable surplus evenly between buyers and sellers. Any residual demand or excess supply not met within the virtual transformer is cleared through the platform’s existing state-level “community trading” mechanism or through the standard retail market.

In addition to baseline P2P trading, the intervention includes time-targeted event-day incentives. Event days were determined using historical irradiance data. For each virtual transformer, we identified an irradiance threshold—typically around the 80th percentile—such that the expected number of high-solar days was 5 per month. A day-ahead irradiance forecast triggered an event day whenever all participant locations within a virtual transformer were predicted to exceed that transformer-specific threshold. Text messages were sent out the evening before indicating event days.

Control participants remain active on the platform but do not engage in P2P trading during an initial baseline period, after which they receive access to the same trading intervention.

Experimental design

The study uses a randomized controlled trial with clustered treatment assignment. Households are first randomly assigned at the participant level to either treatment or control status, with stratification by battery ownership, geographic location, and timing of enrollment. Following this assignment, participants are grouped into geographically proximate “virtual transformers,” with stratification to ensure a mix of solar producers and off-takers and adequate trading capacity within each cluster.

Treatment is clustered at the virtual-transformer level. Treated clusters begin P2P trading after a one-month baseline period, while control clusters remain in a baseline condition—visible on the platform but not engaging in trading—for approximately four months before crossing over into treatment. To maintain functional trading groups, a small number of solar producers may be reassigned across virtual transformers when initial groupings lack sufficient counterparties; these adjustments do not affect the treatment assignment of off-taker households.

Within treated virtual transformers, default trading relationships are established using a predetermined hierarchy that assigns each off-taker to one or more solar producers. Prices for each trading pair are fixed according to the intervention rules and do not vary endogenously within the experiment. Event-day incentives are layered onto the treated clusters during the treatment period and are triggered based on transformer-specific irradiance thresholds derived from historical data and day-ahead forecasts.

Design details

Randomization method
Randomization was conducted by the research team using a computer-generated random number process.
Randomization unit
Participant-level randomization with cluster-level implementation
Clusters
34 virtual transformers
Observations
24 hours x 300 participants x 182 days
Arms
17 treatment clusters and 17 control clusters
Primary outcomes
Hourly electricity demand for off-takers, Hourly electricity supply / exports from solar producers
Secondary outcomes
Persistence in Responsiveness to Event Days (next-day / next-week spillovers), Default Trades vs. Organic Trades on the Platform (Correlations with engagement and prices of non-default trades including system size, weather, make up of virtual transformer), Net export (reverse power flow) at the virtual transformer level (Hourly net export, 95th/99th percentile export during solar peak, probability net export exceeds a threshold (proxy for constraint), Implied increase in hosting capacity (modeled using experimentally estimated elasticities + engineering simulations), change in required distribution upgrades (reduced need for transformer upgrades)

Cited by

No citation data available for this registration yet.

Keywords