An Agent-Based Model of Electricity Consumer: Smart Metering Policy Implications in Europe

: EU Regulation 2009/72/EC concerning common rules for internal market in electricity calls upon 80% of EU electricity consumers to be equipped with smart metering systems by 2020, provided that a positive economic assessment of all long-term costs and benefits to the market and the individual consumer is guaranteed. Understanding the impact that smart metering systems may have on the electricity stakeholders (consumers, distribution system operators, energy suppliers and the society at large) is important for faster and effective deployment of such systems and of the innovative services they offer. For this purpose, in this paper an agent-based model is developed, where the electricity consumer behaviour due to different smart metering policies is simulated. Consumers are modelled as household agents having dynamic preferences on types of electricity contracts offered by the supplier. Development of preferences depends on personal values, memory and attitudes, as well as the degree of interaction in a social network structure. We are interested in exploring possible diffusion rates of smart metering enabled services under different policy interventions and the impact of this technological diffusion on individual and societal performance indicators. In four simulation experiments and three intervention policies we observe the diffusion of energy services and individual and societal performance indicators (electricity savings, CO 2 emissions savings, social welfare, consumers’ comfort change), as well as consumers’ satisfaction. From these results and based on expert validation, we conclude that providing the consumer with more options does not necessarily lead to higher consumer’s satisfaction, or better societal performance. A good policy should be centred on effective ways to tackle consumers concerns.


Introduction
. The deployment of smart metering systems in Europe is driven by EU legislation that views smart metering infrastructure as a tool to both enhance competition in retail markets and foster energy e iciency.Moreover, smart metering infrastructure is considered as key enabler to realising the full potential of renewable energy integration and provision of a secure energy supply.
. The EU Directive on internal energy market / /EC establishes common rules for internal market in electricity and urges an access of consumption data and associated prices to the electricity consumers.The information on electricity costs should be provided frequently enough in order to create incentives for energy savings and behavioural change.Such information provision could also create innovative services to e ectively enable active participation of consumers in the electricity supply market.Implementation of smart metering infrastructure is expected to facilitate this process.Directive (EU) / /EC along with the Recommendation / /EU calls upon % EU electricity consumers to be equipped with smart metering systems by , providing the economic assessment of nation-wide smart metering roll-out is positive.Therefore, smart metering systems, by providing feedback to the households on their electricity consumption, play an important role in the achievement of energy savings.The e ect of feedback on consumer's behaviour has been reported in many pilot projects (Box & Draper ), indicating that the potential for smart metering systems per se to trigger consumer engagement and behavioural changes is rather limited: "Information on consumption will not work without a motivation to conserve, which may be provided by other instruments like financial incentives, goal setting or personal commitment" (Fischer ).Thus, smart metering systems are enabling technologies, which need to be coupled with innovative end-user services to achieve better energy management through the means of rewards, automation and information .Moreover, technological di usion is accompanied by technology concerns, in particular data privacy and security.Such concerns played an important role in some EU national smart metering roll-outs, such as the Netherlands, where the consumers were granted the possibility to refuse the smart meter or accept it under "administrative-o " option .To this end, distribution system operators (DSOs) and energy suppliers will need to take e ective measures in motivating and engaging the consumers in managing their electricity consumption by o ering innovative services, while e ectively tackling concerns that may hinder the deployment of such services.

.
Agent-based modelling is a widely applicable tool for capturing the behaviour of socio-technical systems that generate emergent phenomena in a bottom-up manner (Van Dam et al.
; Chappin ; Epstein & Axtell ).It has also been widely used to study the di usion of new and green technologies and more recently to study the di usion of smart metering technologies (Kowalska-Pyzalska et al. ; Zhang & Nuttall ; Rixen & Weigand ).In particular, Zhang & Nuttall ( ) have developed an agent-based model of a market game involving two parties: residential electricity consumers and electricity suppliers.The aim is to evaluate the e ectiveness of UK policy on promoting smart metering in the UK retail electricity market.They choose the Theory of Planned Behaviour-TPB (Ajzen ) to formalize the behaviour of residential electricity consumer agents.This theoretical choice is driven by the consideration that TPB emphasizes the role of psychological (attitudes), sociological (subjective norms) and environmental factors (perceived behavioural control) in the consumers' decision making process.However, a limitation of the proposed model is consumer's personality characterization.They suggest that consumer's intention to perform certain behaviour is essentially driven by consumer's personality trait "price sensitivity".However, the range of "beliefs" that jointly determine a person's intention to perform a behaviour is certainly broader.Building on the work of Zhang & Nuttall ( ), and broadening consumer characterization, in the present paper we develop an agent-based model of electricity consumers, interacting with the energy supplier through a series of electricity contracts, each characterized by a di erent type of service o ered to the consumer.We are interested in exploring possible di usion rates of smart metering enabled services under di erent policy interventions.The model can be used as a tool to gain insight into diffusion patterns of energy services (represented by a contract) and associated switching rate among contracts.Furthermore, related influencing factors are also observed in the transition to sustainable and cost-e icient energy consumption.The remainder of the paper is structured as follows: Section describes the model, including agents 'properties and actions and characterisation of electricity contracts.Section illustrates the policy intervention whose impact is observed under di erent experimental set-up, defined in Section .We conclude discussing the policy implications of our findings and future considerations.

Model overview
. An agent-based model is developed in this paper that includes a number of household agents (i.e,.electricity consumers) and a portfolio of electricity contracts o ered by the electricity supplier.An overview of the model is presented in Figure .The agents and their interactions with the electricity systems (through the contracts) represent a socio-technical system, where the social subsystem consists of a network of consumers, each of them having a contract with the electricity supplier.Each contract is characterized with a type of end-user service (defined in the contract and enabled by the smart meter) and time duration.Agents gain experience with a certain type of contract and have a memory retaining knowledge on that experience.They also communicate this experience to other agents, which may influence their decision on the type of contract to be adopted (Figure ).
. Agents' behaviour may be influenced by governmental policy (e.g., national roll-out of smart metering systems with opt-out option for the consumers), national/local authority initiatives (e.g., environmental campaign) or business case driven initiatives from the DSO/supplier.While policies and institutions are influenced and shaped by actors' behaviours (DSO, consumers, markets, etc.) and change over time, for the purpose of this model they are assumed to be exogenous and fixed.
. The socio-technical system as a whole evolves based on the decisions of individual agents.These decisions influence the overall system level performance indicators defined as: -adoption of contract types, -energy savings, -CO 2 emissions savings, -comfort change and -social welfare; which will be further detailed in Section .below.The model is simulated for a period of years with time steps of one month and in each simulation run, the system behaviour is a combined result of the actions of all agents.

Characterization of agents .
Agents have personal goals and preferences determined by their own personal values.According to the literature (Steg et al. ; Steg & de Groot ), we can distinguish between self-transcendent values that refer primarily to collective consequences and self-enhancement values, which refer primarily to personal costs and benefits.Self-transcendence values include altruistic values that focus on societal well-being and biospheric values that focus on protecting the environment.Self-enhancement values include egoistic values, which focus on enhancing personal resources (e.g., wealth), and hedonic values, which focus on improving the way one feels (Steg et al. ; Steg & de Groot ).
. These self-enhancement and self-transcendence values characterise the agents' weight factors w e , w h , w b and w a and describe the agents' relevance (Menanteau & Lefebvre ) towards four criteria: financial savings, comfort change, CO 2 savings and social welfare.The weights are randomly assigned to each agent, following a uniform distribution [0, 1], as defined in Table below.The highest weight factor determines the "archetype" each agent belongs to (e.g., agents belonging to egoistic archetype have highest weight factor for the egoistic criterion), which indicates that the agents are heterogeneous with respect to archetype.The weight factors are normalized such that the sum of the weights equals .At the same time, these four values represent the four criteria against which agents evaluate contracts.The four criteria are detailed in the Appendix.

.
Each time step, agents get experience with a certain contract and based on that experience they develop an attitude towards the contract they have.They also memorise the experience they have with all contracts that they have had earlier in time and communicate this memorised experience to other agents, thus influencing their decisions.When an agent is not satisfied with her current contract, she decides to opt for a new one.The satisfaction level is measured relative to the threshold-attitude, as an agent specific variable (for further details, see Appendix).In case the agent is satisfied with her contract, but the contract has ended, the agent considers the present contract in the portfolio of contracts to be evaluated in the next time step.The decision on what

Agents' activities
. The main activity is that of the household agent, as summarized in Figure .Agents have a certain contract α j with the electricity supplier.Each contract communicates range of values (a cmin , a cmax ) relative to the criteria mentioned above, and expected to be achieved with that contract.The average of that range is the communicated score of contract α j on criterion c, i.e.: .
We normalize the scores as in Equation to be able to combine them across criteria when evaluating single contract.
) and a c,minmin = min k (a cmin ) are the best and worst communicated score of α j on criterion c among all k communicated contracts.This way, a c,norm,communicated will always be a value between and .

Updating memory .
Positive consumers' experience would certainly pave the path towards di usion of more advanced smart metering services, which addresses both the acceptance and e ective use of the smart metering system.While EU currently progresses towards nation wide deployment of smart metering systems, the real impact of smart metering enabled services and consumers experience with smart metering systems is still uncertain and limited.However, some observations on potential impacts (energy/financial, CO 2 savings, comfort change, etc.) of using smart metering systems (smart meter and feedback device) is already reported in the literature (see Appendix) and we therefore use a range of such impacts for deriving the experienced score.In this context, each time step agents gain experience with contract α j and the experienced score on each criterion c is derived as a random value from the communicated range, i.e.
a c,experienced = rand(a cmin , a cmax ) ( ) Next, the experienced score is normalized, as follows: where a c,norm,experienced (α j ) is the normalized experienced score of contract α j on criterion c and a c,maxmax and a c,minmin are defined as in Equation .
. Consumers update existing values in their memory with the current experience they have by calculating the average of the past experienced and the new experienced score, as in Equation .At the initial time step and in case of no previous experience: a c,memory(αj ) = a c,norm,communicated(αj ) .
where a c,norm,experienced (α j ) is calculated as in Equation and a c,memory (α j ) is the updated memory value.
. Next, each agent calculates the attitudes towards contract α j by multiplying the scores on criterion c with the criteria-specific weight factor w c and summing the result, i.e.: with w c being the weight factor, as a measure of relative importance the agent gives to criterion c and it is randomly drawn from uniform distribution between and .
. The total attitude towards contract α j is summation of all individual attitudes relative to each single criterion c, where c ∈ e, h, b, a: Based on this attitude and agent-specific satisfaction threshold, the agents decide whether to switch to a different contract or keep the same one they currently have.
Choosing contract from portfolio of contracts o ered by the supplier .

Personal preferences
Agents may consider switching to di erent contract for one of the following reasons: ) Being dissatisfied with the current one and ) Expiration of current contract.In each of these cases, agents ask for new contracts from the supplier and perform evaluation of portfolio of contracts received.If the agent was satisfied with the current contract and the same has ended, she reconsiders it for evaluation, along with the new ones received.
. The decision making process is modelled as multi-criteria problem and is presented in Figure .Technology concerns, in terms of data privacy and security, health, etc. were evident in many EU pilots and national rollouts of smart metering systems.In the case of the Netherlands, for instance, the original legal obligation to accept the meter was revoked due to data privacy concerns.This resulted in granting the consumer with possibility to either refuse the meter or accept the meter but block the remote reading facility (so called "administrative o " option).To introduce such concerns in our model, we characterise each agent with "techno-tolerance threshold" and each contract with a level of perceived concerns ("techno-risks").For each contract from the portfolio received, an agent considers only those that have a "techno-risks" value below her "techno-tolerance threshold".
. The total attitude towards each contract under evaluation is summation of all individual attitudes relative to each single criterion c, where c ∈ e, h, b, a: .

Social influence
Household agents belong to a social network, used to simulate share of agents' perceptions.The network is generated such that each agent communicates with n other peers in a network of agents.Half of those peers belong to the same archetype âĂŞ people tend to associate with others who are similar with according to the principle of "homophily" (McPherson et al. ) and the rest are randomly chosen from di erent archetypes.This determines the level of heterogeneity in our model and it is defined as number of peers, each agent communicates with, belonging to di erent archetype than her own one.

.
Based on the experience each agent shares (through personal interactions, social media, etc.) with these n peers, agents update their attitude, as follows: where A i,c,experienced (α j ) presents the attitude of the i-th neighbour among n neighbours the agent communicates with and it is calculated as in Equation ; w SN stands for the susceptibility factor, as a measure of importance the agent gives to the opinion of her peers and A c,communicated (α j ) represents the personal attitude the agent has towards alternative α j and relative to the score on criterion c, calculated as in Equation .
. The agent behaviour, in terms of electricity contract adoption, thus depends first on her personal attitude towards a certain contract and second on how much the attitude of her peers di ers from her own personal attitude.In this regard, behaviour change according to the second factor is a ected by the individual susceptibility.
. Finally, the agent comes up with an overall attitude A (α j ) towards contract α j , i.e.: .
The contract for which the agent has the highest attitude is her preferred one; that is its final decision on which contract α j to accept.The decision making can be formulated as follows:

Characterization of contracts .
Existing type of contracts along with potential future ones o ered by major EU suppliers has been considered in our model and all the agents have the same contract options.Most of them can be grouped in seven types of contracts, according to the type of service provided, namely: A. Indirect feedback with own historical and peer comparison once a year: this type of feedback provision allows for historical analysis of consumer' electricity consumption and peer comparison at the end of each year and it does not require adoption of a smart meter;  ): this type of feedback provision allows for demand response to electricity price using home automation (as in contract F), however, including self-consumption of electricity produced at consumer's premises.

Policy Interventions
. As mentioned before, the model in this paper explores the Directive (EU) / /EC on internal energy market for electricity and in particular, the recommendation (EU) / /EU on smart metering deployment.In this context, three possible policy interventions are presented.
. Mandatory smart metering policy: There is governmental policy in place, which mandates the DSO to install smart meters to all electricity consumers.This situation resembles the situation in most of the EU countries, where the consumer is required to accept the smart meter and can choose one of the contracts indicated in Table . .Voluntary smart metering policy: This policy mandates the DSO to carry on nation-wide smart metering deployment, nevertheless, the consumer can choose to refuse the meter (contract A) or opt for "administrative o " (contract D).This represents the situation in some EU countries where data privacy concerns resulted in introduction of the "opt-out" and "administrative-o " option for the consumer.The types of contracts o ered to the consumer in this policy are indicated in Table . .Environmental smart metering policy: The conditions in this policy option are the same as in the voluntary policy and the agents are entitled to the same contracts (see Table ).In addition, environmental campaign, launched by national/local authority is assumed to take place at a certain time step ( e.g., th month in our model).As a result, we assume that consumer will become more sensitive to environmental issue and therefore their biospheric weight will increase.We hypothesize an increase of biospheric weight by %, thus for some agents becoming the most salient weight.This would ultimately imply change of archetype for these agents and increase of biospheric agents.

.
We are interested in monitoring possible di usion rates of smart metering enabled services under the three di erent policy interventions presented above.

Simuation and Data Analysis
. The model has been implemented in NetLogo (Wilensky ) and extensively verified using both single and multi-agent testing (Van Dam et al. ), whereas the programming language R was used for the data analysis .
Experimental set-up .Data analysis was performed by building experimental set-up relative to the following variables: heterogeneity, policy, initial contract distribution and contract duration.
. We have built four experiments, relative to the initial contract distribution and contract duration (see see Table ).The contract types considered in our study reflect possible smart metering enabled services to be deployed in EU and we are interested to observe the impact of initial contract distribution on the final system level contract adoption.Scenario and are conservative assumption in this regard and reflect the current EU situations, whereas Scenario and depicts future conditions, where variety of smart metering services will be

Contract duration
All agents have the least technologically advanced contract (B in the mandatory policy and A in the voluntary and environmental policy) months Equal contract distribution months All agents have the least technologically advanced contract (B in the mandatory policy and A in the voluntary and environmental policy)

Equal contract distribution Indefinite
Table : Experimental set-up available to the end-user.Similarly, the idea of having months (as minimum contract duration observed in most of the EU member states) and indefinite contract duration is centred on our interest to analyse the e ect of "lock-in" periods, during which the consumer would need to pay a penalty for leaving the contract or other switching barriers, in case of fixed contract duration.Each experiment is tested for each policy separately. .

Data analysis .
The numerous individual decisions of agents to switch to a certain type of contract influences system level outcomes: adoption of contract types, average financial savings and CO 2 savings, comfort change and social welfare.Since we are interested in the system level performance that emerges from lower-level properties and processes, we focus our attention on these indicators and analyse the patterns due to change of policy and scenario.Parameter values vary between runs due to the stochastics used during agents' initialization and model execution (see Table ).To be able to arrive at realistic assessment of patterns observed in the simulated system evolution, one needs to do a statistical analysis of the results of many runs.

Average contract distribution .
Figure depicts the average contract distribution among agents for each policy intervention and experiment performed, discussed in the following section.In the sections therea er, the average and standard deviation of financial savings, comfort change, CO 2 emissions savings and social welfare, is analysed, based on the runs performed.They are visualized in Figure to Figure .
. Regarding the contract distribution, we can observe and explain the following: . In all experiments performed, there is a major di erence in the adoption level of contract A ("opt-out" option and feedback once a year) and contract D ("administrative-o ") when comparing the mandatory policy on the one hand, and the voluntary and environmental policy on the other hand.This di erence is caused by the fact that contract A and D are not available in the mandatory policy.
. There are no significant di erences in the adoption level of the contract types between the voluntary and environmental policy within a scenario even though one would expect that the environmental policy and the associated increase of biospheric consumers would yield a higher share of more advanced contracts.This can be explained by the fact that the techno-tolerance threshold, as currently modelled, does not vary with the archetype, i.e. it has the same value for each archetype.As a result, increase in the number of biospheric consumers does not necessarily lead to increased adoption of more technologically advanced contracts.
. More technologically advanced contracts, such as contract F and G are highly adopted in all policies.This is caused by the fact that these are the best-scoring contracts for out of the criteria that agents consider in their choice.
. For the voluntary and environmental policy, consumers have a much higher adoption level of contract C when contracts are initially equally distributed amongst agents ( nd and th scenario), than when all agents have initially contract A ( st and rd scenario).This is explained by the fact that some people tend not to switch away from their initial contracts since they remain satisfied over the course of the simulation.
. Figures -illustrates the impact of agents' behaviour on the four criteria mentioned above: financial savings, CO 2 emissions savings, comfort change and social welfare, both the average value across the whole simulation period and all simulation runs (continuous line) and the spread around the average value (colour shaded area).These results are discussed in the following sections.
. Heterogeneity, as modelled in this paper does not prove to have impact on the average contract distribution.This is due to the fact that egoistic, biospheric and altruistic agents have objectives which pull in the same direction, in terms of contract type preference, i.e. agents who belong to these three archetypes will behave similarly, whereas hedonic agents will act di erently.As such, more technologically advanced contracts that would yield higher energy and financial savings, would also result in higher CO 2 savings and increased social welfare.

Analysis of the system level performance indicators .
In Figure financial savings are lowest in the rd scenario, due to highest average adoption of contract B (feedback once per two months) in the mandatory policy and both contract A (feedback once a year with no smart meter) and B in the voluntary policy.Similarly, highest financial savings can be observed in the th scenario, as a result of highest adoption of contract G, as most technologically advanced contract and lowest adoption of both contracts A and B across all three policies (see Figure ).There is no significant di erence in the financial savings between the st and nd scenario.This means that when the contract duration is months it does not make a di erence what the initial contract distribution is.The contract duration does however have an impact if the contracts last indefinitely.
. In scenario , highest financial savings are observed in the environmental policy and lowest in the voluntary one, owing to higher adoption rate of contract F and G in the environmental policy (in comparison with the first two policies) and lower adoption rate of contract A and B. In scenario , and , the financial savings are highest in the mandatory policy due to high adoption of contract G and low adoption of contract A.
. Contrarily to the financial savings, one can observe lowest comfort reduction in scenario and highest in scenario .The same reasoning holds, as in the analysis of the financial savings, i.e. lower adoption level of contract A and B, combined with higher adoption of contract G in scenario results in highest comfort change for that scenario.
. Similarly to the financial and CO 2 savings, highest increase in social welfare is observed in scenario (for the same reasons mentioned above).Nevertheless, the di erence between this indicator and the financial and   Along with the system level performance, i.e. financial and CO 2 savings, comfort change and social welfare, we also observed the agents' satisfaction, in terms of their perception towards potential risks associated with single contract type (data privacy and security, health, etc.), but also agents' overall satisfaction due to specific attitude threshold, set up in the initialization.
. Figure illustrates agents' satisfaction due to their perceived smart metering technology risks, as di erence between agent specific tolerance threshold and technology risk associated with the adopted contract.On this note, one may observe that agents may be satisfied with their choice, in terms of overall contract performance, nonetheless at the expense of the technology risks perceived with the chosen contract.The tolerance threshold is initialized at the beginning as a random number drawn from a predefined set.In future model development, the tolerance threshold shall vary in accordance to the experience the agent has with the smart metering technology or the impact one may have from media or experiences in her social network.
. Furthermore, one may observe from Figure that average agents' tolerance satisfaction is worst in scenario , owing to higher adoption of contracts F and G (as most technologically advanced contracts and thus perceived technological risks) and lower adoption of contract A. Similarly, highest tolerance satisfaction is observed in scenario due to lower adoption of contract F and G and higher acceptance of contract A, in comparison with other scenarios.Moreover, in all scenarios, the average techno tolerance satisfaction is the least for the environmental policy.
. Energy savings, CO 2 emissions reduction and social welfare score highest for the environmental policy (Figures , and ) in scenario , which results in lowest techno-tolerance satisfaction, due to higher adoption of contract F and G and lower adoption of contract A. Similarly, all performance indicators perform worst for the voluntary policy, in scenario (Figures -), due to lower adoption level of contract G, in comparison with the mandatory and environmental policy and higher adoption level of contract A, in comparison with the first policy.This also leads to highest techno-tolerance satisfaction for the voluntary policy.Figure depicts the average attitude satisfaction, as di erence between the agent's general attitude regarding the certain contract and the agent specific attitude threshold.
. One may observe from Figure that the average attitude satisfaction is similar among scenarios, and with no significant di erence among the three policies.What is evident however, is the negative average satisfaction level in all the scenarios and for all policies, which indicates high average switching rate.

Conclusions
. The research question that the model aimed to answer is how smart metering technologies (and thereby en-  .The most remarkable outcome is that granting the consumers with opt-out and "administrative-o " option for smart metering system resulted in increased number of consumers opting for a less technologically advanced contract (i.e., contract A or D).Given our initial assumptions on population preference distribution, one may conclude that addressing consumers concerns (such as data privacy and security) by granting them with more options, does not necessarily lead to higher energy and CO 2 savings and ultimately higher consumers' satisfaction.During expert validation, this pattern has been recognized as realistic.
. This e ect remained strong even under the environmental policy, where despite significant number of agents becoming more environmentally concerned (i.e., having a higher weight for the biospheric criterion) system level indicators, such as energy and CO 2 savings remained lower, in comparison with the mandatory policy.
This result is due to significant adoption of contracts that do not require data sharing with DSOs (e.g.contract A and D in the environmental policy) and thus do not present additional benefits, such as energy/financial savings due to dynamic pricing.Furthermore, technologically advanced contracts that yield higher benefits may also be subject to technology concerns, as perceived by the consumers.As a result, average techno-tolerance satisfaction appears to be lowest for the environmental policy, which is associated with high level of perceived technological risk by the consumers for the adopted (more technologically advanced) contracts.Hence, it can be argued that giving consumer more choices do not necessarily produce better system level results, which is in line with psychological research.Schwartz & Ward ( ) claims that an abundance of choice is likely to produce worst decisions because people tend to simplify their choices to the point that the simplification hinders their capability to opt for a good choice.In this regard, policies may need to target the information to the right segment of population in order to avoid information overload.

.
Di erently from the average techno-tolerance satisfaction, the average total-attitude satisfaction does not vary greatly among policies and scenarios.For instance, agents experience highest techno-satisfaction in Scenario (due to highest adoption level of contract A); nevertheless, the total-attitude satisfaction remains similar as in the rest of the scenarios.This can be associated with the high diversity of the agents' population (in terms of agents' archetype). .
Furthermore, there is no significant di erence in the average total-attitude satisfaction among the policies; nevertheless, it appears to be slightly higher in the mandatory policy.This may be due to the increased adoption level of contract A in the voluntary and environmental policy, i.e., agents opt for less attractive contracts due to technological risks perceived for more attractive ones.Therefore, it can be argued that providing the consumer with more options (voluntary policy) does not necessarily lead to higher consumer's satisfaction.Indeed, giving consumers too many options (in our case contracts) to choose from leads the consumer feeling less satisfied even a er taking the decision (Schwartz & Ward ).
. Hence, risks perceived by the consumers shall be approached as an early attention point, and in particular, the ways in which their energy consumption data is used, by whom, and for which purposes.Information strategies (e.g., environmental campaigns in the environmental policy) for increasing consumers' awareness (e.g., towards environmental benefits) do not seem su icient to e ectively di use the full potential of smart metering services, even in the case of targeting specific segment of population (e.g., biospheric archetype).Associated risks perceived by the consumers (e.g.data privacy and security) still remain hurdle towards adoption of more advanced contracts.Therefore, policy interventions need to simultaneously address adoption barriers and openly communicate potential concerns and treat them e ectively (e.g.reassuring the consumer that she cannot be disconnected without notice, ensuring that "administrative o " actually means no metering data is being exchanged, etc.).
. To conclude, our results show some interesting policy implications.A good policy should be designed so as to adequately inform (right and complete information) the consumer on the advantages and disadvantages of the o ered technological solution.The consumer can therefore feel more comfortable in accepting more advanced technologies and lower her technological concerns (e.g., lower techno concern for well-designed information campaign âĂŞ not presently included in our current model).Consequently, this could lead to the consumer feeling more satisfied.

Future Considerations
. The famous quote by P.E.Box states that "All models are wrong, but some are useful" (Box & Draper ).While there are clear useful insights to be drawn from the current work, there are four important limitations that we wish to address in the future.

.
First of all, total-attitude satisfaction stays below in all the scenarios, i.e. agents are constantly dissatisfied, which results in high switching rate.This can be explained by two facts: first, in our study, the experience the agents get with each contract is modelled as exogenous variable, each time step randomly drawn from a predefined set of values (defined in the contract) for each indicator, whereas the evaluation of the current contract shall reflect upon learning e ects from past experiences and adapt the current experience accordingly (e.g. through adaptive set for each indicator).Second, the attitude threshold and techno-tolerance threshold are exogenous, fixed at the initialization of the model.Fixed techno-tolerance threshold means consumers disregard the contracts that are below their techno-tolerance threshold.Such an approach prevents the agent to consider more "technologically risky" contracts at the expense of better outcome (in terms of energy savings, environmental impact, etc.).The perception for more "technologically risky" contracts may change over time, owing to the experience an agent has with her contract, which will ultimately result in adaptive techno-tolerance threshold.Similarly, attitude threshold shall consider agents' learning and adaptation and therefore be reflexive and reactive to the environment.This certainly deserves attention in future developments of the model.
. Third, future model development shall consider more reflexive and reactive institutions, as well as explore institutions emerging from agents' behaviour.Finally, improving the model with empirical data constitutes a major route for further development of the model.

Appendix
In the following, the agent-based model developed in this paper is described using the ODD (Overview, Design concepts, Details) protocol (Grimm et al. ).

Purpose
The purpose of this paper is to explore the di usion potential of smart metering enabled services among a population of interacting electricity consumer and to evaluate the impact of such di usion on individual and societal performance indicators.

Entities, state variables and scales
Central entity of the model is electricity consumer, representing individual household.Consumer agents are characterised by: weight factors, memory retaining knowledge on the experience with the contract they currently own, aspiration threshold on the perceived technological risks and aspiration threshold on the overall attitude ("threshold-attitude") towards certain contract.The weight factors w e , w h , w b and w a describe the criteria's relevance (Menanteau & Lefebvre ) and the highest one determines the social aggregation i.e. "archetype" each agent belongs to.The weight factors are normalized such that the sum of the weights equals .
At the same time, the four values of the weight factors represent the four criteria against which agents evaluate and score contracts: Financial savings, resulting from energy savings: E average,saved [kW h] is the average monthly energy saved and price [ A C kW h ] is the electricity price.Thermal comfort change: Social welfare -Demand factor is used as a proxy of how e iciently the customer is using electricity during defined time period (e.g.month), i.e.: is the maximum load used in a given time period (i.e.month) and P peak [kW ] is the peak power during time period of one month, corresponding to the contracted capacity of the household .Low demand factor means less system capacity is required to serve the connected load.
Temporal resolution of the model is monthly time step, whereas the spatial resolution is abstract.Contracts have a fixed position in a rectangular plane and agents move within the space by changing contracts.Finally, average technology satisfaction and average attitude satisfaction have been analysed in each policy and among scenarios to understand the link between system level performance (in terms of financial savings, CO 2 emissions reduction, etc.) and the agents' satisfaction level.A distinct pattern of distribution of contracts, system level outcomes and satisfaction level emerges.

Initialisation
Model initialisation follows three successive steps: creating household agents, generating the social network and setting the state variables of the model.Each agent is assigned random weight for each of the four criteria.This means that also the archetype she belongs to is randomly assigned.Next, the peers each agent communicates with are randomly chosen from her social network.
Finally, the state variables (mentioned in Table ) are initialised for each agent and contract type.Weighting factors, threshold-attitude, susceptibility and heterogeneity are assumed not to change during the simulation period, whereas the memory varies according to the experience the agent has in each time step.The technotolerance threshold is initialised at the beginning of the simulation and randomly chosen from a predefined range.Contracts are assumed to be initially equally distributed among the agents and further experiments with di erent initial contract distribution are also tested in Section .Each contract is characterised with contract duration, communicated range for each criterion and perceived technological risk.The communicated ranges for financial savings assumed average energy savings between and % from contract A to G, respectively (Box & Draper ; Van Elburg ).The analysis of electricity prices for households is based on prices for the medium EU standard household consumption band, namely one with annual electricity consumption between and kWh (Eurostat).We have considered an average annual household consumption of kWh/year or kWh/month in our analysis.The average price of electricity for household consumers in the EU-(the prices for each EU Member State are weighted according to their consumption by the household sector for ) was EUR .per kWh (Eurostat).
We express comfort change as a temperature deviation relative to the target thermostat settings.Comfort cannot be defined absolutely; however the World Health Organization's standard for warmth indicates • C as suitable temperature for healthy people who are appropriately dressed.For those with respiratory problems or allergies, they recommend a minimum of • C as a target thermostat setting and gradually reduce it to a minimum of • C in the case of contract G.As for the CO 2 emissions savings, the EU Covenant of Mayors reports a value of .
(t CO2/M W he) as standard emission factor and .
(tCO2 − eq/M W he), as LCA emission factor , for EU (Covenant of Mayors ).In our model, we have used a value of .t CO2/M W he as standard reference factor.
Finally, demand factor was used as a proxy of social welfare.One shall note here, that the social welfare, in our model, is mainly linked to security of supply, since lower demand factor leads to more flatten household load profile and thus contribute towards enhanced energy usage and reduced outages in the neighbourhood.We have used peak load reduction between % due to demand response and maximum % due to both demand response and renewable energy self-consumption for contract G (Darby & McKenna ). Table provides an overview of the communicated ranges for each contract type.

Figure :
Figure : Overview of model entities and their relationship.

Figure
Figure : Social welfare.

Figure :
Figure : Average attitude satisfaction.
The model was run in an experimental setup of runs for each parameter combination in order to be able to explore the spread in the outcomes, which is caused by randomly determined factors like the social network layout, weight factors of household agents, agents' experience with certain contract, agents' susceptibility and technology threshold.The parameterization for the simulation experiments is given in Table.Empirical values were not available for most of the parameters and as a result, synthetic data were used, based on expert judgment and the same were extensively varied.Nevertheless, wherever a source is given, the parameter value is empirically based.Each experiment starts with N agents with randomly generated weights.The highest weight of an agent determines the archetype she belongs to.
T set[%]is the target thermostat temperature set by the agent and T f inal[C]is the actual temperature occurred due to behaviour change.CO 2 emissions reduction:a b [t co2 ] = E average,saved [kW h] * e rfE average,saved [kW h] is the average monthly energy saved and e rf [ tco2 kW h ] represents emissions reference factor .

Table :
Communicated ranges for electricity contracts the reasoning behind the contract adoption patterns and influencing factors (incentives, social influence, etc.).
Contract Financial savings[A C] Comfort change [%] CO emissions savings [t] Social welfare [%] • C; and for the sick, disabled, very old or very young, a minimum of • C (Organisation ).According to a study by housing expert Richard Moore (see Boardman et al. ), comfortable indoor temperature lies within a range of [18 • C − 21 • C].We have considered a temperature of