Identifying customer patterns for energy services in a ...utility providers (Capgemini Consulting, 2015). First, customers are looking for more flexible and responsive suppliers of

1

Identifying customer patterns for energy

services in a dynamic price setting

Navid Sadat-Razavi Rotterdam School of Management

Erasmus University Rotterdam

A thesis submitted for the degree of

Master of Science

Business Information Management

12th July 2016

Master Thesis – Navid Sadat-Razavi

2

Student Information

Name: Navid Sadat-Razavi Student Number: 441482 Study Programme: Msc Business Information Management

Submission Date: 12.07.2016

Graduation Committee

Acknowledgements

University Coach: Ir. Derck Koolen Phd Candidate Department of Technology and Operations Management Rotterdam School of Management (RSM) Co-Reader: Dr. Yashar Ghiassi-Farrokhfal Assistant Professor Department of Technology and Operations Management Rotterdam School of Management (RSM) External Coach: Mark Schütz Managing Director Utilities and Strategy Transformation Capgemini Consulting

The work in this thesis was supported by Capgemini Nederland B.V. Their cooperation is gratefully acknowledged. Special thanks go to Mark Schütz (Managing Director Utilities), Jeroen van Daal (Principal) and Arie Hobbels (Consultant) at Capgemini.

The work in this thesis was supported by Qurrent Energie. Their cooperation is gratefully acknowledged. Special thanks go to Mark van Loon, Business Development Manager at Qurrent Energie.

The work in this thesis was supported by Vereniging Eigen Huis. Their cooperation is gratefully acknowledged.


3

Preface

The copyright of the master thesis rests with the author. The author is responsible for its contents. RSM is only responsible for the educational coaching and cannot be held liable for the content.

The author declares that the text and work presented in this Master thesis is original and no sources

other than those mentioned in the text and its references have been used in creating this Master thesis.


4

Executive Summary

The main aim of our study was to examine customer patterns of residential households in a dynamic price setting. With the development of new technologies such as smart meters, utility providers have the opportunity to inherently change the way they are communicating with their customers. One way the utility providers are aiming to improve the interaction with their customers is by introducing dynamically changing electricity prices, based on real electricity market prices. However, less is currently known about consumer preferences and response to dynamic prices. Given that different people can have a diverse set of values and preferences, it is assumed that their reaction to changing electricity prices is quite diverse. Therefore, we have asked ourselves how household characteristics and dynamic electricity prices influence household electricity consumption in dynamic price settings. We believe that clarifying the relationship of these two components with electricity consumption behavior can deliver valuable insights for future energy services.

We have conducted three separated analyses, investigating the influence of household attributes on households’ willingness to use electricity (RQ1), the influence of dynamic electricity prices on household behavioral electricity patterns (RQ2), and a household segmentation based on household attributes that are capable of reflecting behavioral patterns of electricity consumers (RQ3). Our data set was obtained from a pilot project of a Dutch energy supplier called ‘Qurrent Energie’, in which households were (and still are) provided with dynamic electricity prices. We are using a set of panel data regressions, a principal component analysis, and a k-means analysis to derive at our results.

Our findings have shown that the willingness to use electricity of households exposed to dynamic price settings is lower as compared to household exposed to the usual Time-of-Use price settings. Moreover, the willingness to use electricity has a positive relationship with household attributes, such as the number of occupants, the building size, age and type. In addition, we have found that, other than the availability of roof insulation, all household attributes have shown a weaker relationship with a households’ willingness to use electricity compared to our control group. Lastly, we have found that the significance of these household attributes is varying during different hours of the day. This gives reason to believe that households exposed to dynamic prices have generally become more price sensitive and follow fewer habitual patterns. Moreover, the analysis of this study has confirmed that dynamic prices have a significantly different, in our case less positive, relationship with electricity usage behavior compared to TOU prices. Additionally, our results are able to prove that the capability of households to change their consumption behavior based on changes in electricity prices only exists between 8AM to 5PM. We have found that the time window in which household are capable to change their behavior based on electricity prices is overlapping with the time-window in which the relative usage between treatment and control group is deviating from each other. We have interpreted this observation as instances of dynamic pricing encouraged load shifting behavior. In the last step, we have investigated whether the household attributes reflect lifestyle patterns of households, by conducting a cluster analysis. Our results show that clusters two and three are showing clearly distinguishable behavioral electricity usage patterns from the rest of our sample. Two conclusions can be made from these findings. First, by segmenting households based on their household attributes, we were able to isolate two groups that are potentially engaging with the introduced dynamic prices by changing their behavioral patterns. We can claim that cluster 2 reduced their load during the day, while cluster 3 engaged in load shifting behavior. Second, we can claim that our cluster analysis confirms our assumption that household attributes reflect electricity usage lifestyle patterns.


5

Table of Contents Executive Summary ............................................................................................... 41. Introduction ..................................................................................................... 8

1.1 Context Motivation ............................................................................................................... 9

1.2 Research Question .............................................................................................................. 10

1.3 Structure .............................................................................................................................. 102. Theory and Model .......................................................................................... 12

2.1 Demand Response ............................................................................................................... 12

2.2 Consumption Behavior and Behavioral Change ................................................................ 15

2.3 Summary of Concepts ......................................................................................................... 17

2.4 Household characteristics and Household Willingness to Use Electricity ........................ 18

2.5 Electricity Prices and Household Behavioral Consumption Patterns ................................ 20

2.6 Household Segmentation for DR improvements ................................................................. 213. Methodology and Results .............................................................................. 22

3.1 Data and Descriptives ........................................................................................................ 22

3.2 Analysis of Willingness to Use Electricity (RQ1) ............................................................... 26

3.3 Analysis of Relative Electricity Usage (RQ2) ..................................................................... 35

3.4 Analysis of Household Segmentation (RQ3) ....................................................................... 434. Discussion ........................................................................................................ 52

4.1 The Influence of Household Attributes on Willingness to Use Electricity .......................... 52

4.2 The Influence of Dynamic Prices on Usage Behavior ........................................................ 53

4.3 Clustering Usage Behavior Based on Household Attributes .............................................. 54

4.4 Bringing the Findings Together .......................................................................................... 55

4.5 Limitations and Recommendations for Future Research ................................................... 565. Conclusion ....................................................................................................... 58

5.1 General Conclusion ............................................................................................................ 58

5.2 Managerial Implications ..................................................................................................... 59

5.3 Academic Implications ........................................................................................................ 596. References ...................................................................................................... 617. Appendix .......................................................................................................... 66

7.1 Qurrent Energie Dashboard Screenshots ........................................................................... 70


6

List of Tables

Table 1. Incidence of Household Attributes in Reviewed Studies and their Impact on Energy Consumption .......................................................................................................................................... 19 Table 2. The Real Time Electricity Price Composition of ‘Qurrent Energie’ ...................................... 23 Table 3. Summary Statistics of Household Attributes – Treatment Group .......................................... 25 Table 4. Summary Statistics of Household Attributes – Control Group ............................................... 25 Table 5. Summary Statistics of Willingness To Use Electricity ........................................................... 29 Table 6. Correlation Matrix ................................................................................................................... 29 Table 7. Panel Data Regression Results RQ1 ....................................................................................... 31 Table 8. Testing Influence of Household Attributes between Treatment and Control Group .............. 34 Table 9. Comparison of Regression Results ......................................................................................... 39 Table 10. Statistical Significance of Differences in Regression Coefficients between Treatment and Control Group ........................................................................................................................................ 40 Table 11. 24h Panel Data Regression Results of Dynamic Prices ........................................................ 42 Table 12. PCA Results .......................................................................................................................... 46 Table 13. Summary Statistics of Prices and Relative Usage ................................................................. 66 Table 14. Summary Statistics of Prices and Relative Usage ................................................................. 66 Table 15. 24h Panel Data Regression Results Equation 1 – Treatment Group ..................................... 67 Table 16. 24h Panel Data Regression Results Equation 1 – Control Group ......................................... 68 Table 17. PCA Analysis of All Available Household Variables .......................................................... 69


7

List of Figures

Figure 1. Classification of DR Programs (Albadi & El-Saadany, 2008) .............................................. 13 Figure 2. Heuristic model of environmentally relevant behavior (Matthies, 2005). Translated by Fisher (2008). .................................................................................................................................................... 17 Figure 3. Comparing the Willingness to Use Electricity between Treatment and Control Group ....... 32 Figure 4. Comparison of Regression Results between Treatment and Control Group ......................... 33 Figure 5. Dynamic Pricing Scheme with Min and Max. ...................................................................... 37 Figure 6. TOU Pricing Scheme. ............................................................................................................ 38 Figure 7. The Average Relative Load Profiles of Treatment and Control Group. ............................... 38 Figure 8. and 9. Within Group SSE of Actual and 250 Randomized Data Sets against 15 Cluster Solutions ................................................................................................................................................ 45 Figure 10. and 11. The difference of Within Group SSE of Actual and 250 randomized Data sets against 15 Cluster Solutions................................................................................................................... 46 Figure 12. Visualization of Five-Cluster Solution of K-Means Analysis Along the Two Strongest Principal Components ............................................................................................................................ 47 Figure 13. Overview of the Distribution of Clusters ............................................................................ 48 Figure 14. Average Household Occupancy for Each Cluster ............................................................... 49 Figure 15. The Distribution of Building Types for Each Cluster ......................................................... 49 Figure 16. Average Building Size for Each Cluster .............................................................................. 50 Figure 17. Distribution of Terrain Type per Cluster ............................................................................. 50 Figure 18. Differences in the Relative Electricity Usage Across the Identified Clusters ..................... 51


8

1. Introduction

The international energy outlook (IEO) 2016 has forecasted a sharp increase in demand for energy until 2040. According to the IEO, the total world consumption of energy has been, and will further increase from 549 quadrillion British thermal units (Btu) in 2012 to 815 quadrillion Btu in 2040 (U.S. Energy Information Administration, 2016). The increasing shortage of natural resources such as oil, the increasing environmental pollution and related threats of global climate change have triggered a new debate about the sustainable nature of natural resources, and policies to restructure the energy sector.

One specifically disrupted energy sector is the electricity industry. Several utility providers throughout Europe are experiencing financial losses due to new market conditions (The Economist, 2013). Coming from a historical position of a quasi-regulated electricity market, it was the responsibility of utility providers to ensure a stable state of the grid and a reliable supply of electricity in the country. Hence, the focus of utility providers lied primarily on operational excellence and the security of electricity supply, creating an inflexible attitude towards dependent households (Spiegel, 2014). Developments of recent years have changed the electricity landscape, and thereby the success of the traditional business model of utility providers, namely, the EU policy led energy transition and changing electricity consumers.

As a result of the policy-led energy transition towards renewable electricity, the European electricity landscape is rapidly transforming into an industry that makes it difficult for conventional utility providers to remain profitable. The growing share of renewable electricity in the European energy mix is increasing uncertainties in electricity production planning (Capgemini Consulting, 2015). Renewable electricity sources are, unlike traditional sources, volatile in nature and strongly dependent on external factors such as weather conditions (DeMeo et al., 2007). Hence, renewable energy production has become increasingly difficult to forecast, due to the dependency on weather conditions.

As electricity consumers are becoming increasingly IT-savvy and used to interactive, smart digital services from other industries, several new trends are evolving in respect to customer demands towards utility providers (Capgemini Consulting, 2015). First, customers are looking for more flexible and responsive suppliers of energy (Capgemini Consulting, 2015). Today, consumers interact with utility providers approximately 9 minutes per year, which is a considerably small window of interaction (Accenture, 2015). Second, customers are increasingly interacting with their energy ecosystem, thereby raising the need for a two-sided communication with utility providers. New renewable energy technologies like PV-solar panels are enabling households to produce their own energy, making them ‘Prosumers’ rather than consumers (Grijalva & Tariq, 2011). The increasingly decentralized production of energy will make households more independent from utility providers (Grijalva & Tariq, 2011). Therefore, utility providers need to reposition themselves by identifying value-added services for energy consumers that can address one or more issues evolving from the energy transition.

Consequently, utility providers need to focus on process optimizations and the balancing of energy generation levels on the one hand, and the improvement of customer relationships on the other hand. Our study is going to focus on new business models that can evolve through a better communication and relationship with electricity consumers.


9

1.1 Context Motivation

The energy transition trend is coinciding with the digital revolution, another significant trend that is spanning across various industries (Capgemini Consulting, 2015). The emergence of smart technologies, big data analytics, and information technology have created new means of organizing operations and customer relationships (Westerman et al., 2014).

1.1.1 The Smart-Grid

These technological developments have a significant impact on the European electricity grid, providing new possibilities in the energy sector. Smart devices have started to connect residential households with utility providers and the rest of the grid, producing big amounts of previously unavailable data. This increased amount of data is enabling new entrants as well as incumbents to reshape internal operations, gain new insights about market conditions, renewable electricity production and consumer behaviors in nearly real-time (Capgemini Consulting, 2015). Hence, the notion of the smart-grid provides an entirely new perspective on the challenges of the electricity industry. The digitalization of energy services can improve utility providers’ understanding of increasingly uncertain energy production and consumption, while encouraging a better way of communication with electricity consumers and providing electricity consumers with new information about their use of electricity.

In the past, several initiatives have been targeted towards residential households as end-consumers of electrical energy. These so called ‘Demand Side Management’ programs have been partly successful, but had their limitations due to lack of the resources to properly record consumers’ electricity consumption behaviors (Capgemini Consulting, 2015). However, recent technological developments have brought a new perspective to this topic that significantly contributed to our motivation to investigate the difficulties faced by the energy sector.

1.1.2 Demand Side Management and Demand Response

With changing customer needs, utility providers need to understand how the demand side of the grid can be managed and supported. Demand Side Management (DSM) refers to a set of measures that can be used to optimize the demand side of the energy system (Palensky & Dietrich, 2011). DSM ranges from the improvement of devices and materials in households to improve energy efficiency, up to real-time adjustments in demand patterns through improved communication technologies (Palensky & Dietrich, 2011). Part of the later is the concept of Demand Response (DR) programs. Essentially, Demand Response (DR) is building upon the behavioral traits of energy consumers, by offering incentives to change or shift electricity usage by communicating with energy users in a comprehensive manner (Darby, 2012). The communication of electricity prices with energy consumers can contribute to increased knowledge of households to comprehend the electricity market situation.

The development of smart metering devices has opened a new world of possibilities to the energy sector. These devices provide detailed information about the customers of utility providers, and their energy consumption patterns (Borenstein et al., 2002). Moreover, the introduction of smart technologies, like smart meters and smart appliances to households has provided consumers with enhanced abilities to control and supervise their consumption patterns in a systematic manner (Jacobsson & Bergek, 2004). Additionally, it has become increasingly easy to reach out to individuals in order to enhance a two-way communication. Hence, the development of smart devices, and smart meters in specific, has opened up


10

new opportunities for utility providers to appropriately reach out to consumers and understand their behavior.

Utility providers are starting to introduce dynamic prices to their energy contracts with residential households, with the aim to accurately match fluctuating energy supply with increasingly uncertain demand patterns. Due to the introduction of smart metering devices, this has become realizable. However, in order to properly take advantage of this improved way of communication, it has to be understood how different types of households react to dynamic prices. Our findings will help utility providers to understand which households are likely to react to dynamically changing electricity prices, and during what time of the day the electricity prices are an effective tool to influence electricity demand. In order to fully understand how next generation DR programs have to be designed, the energy sector needs to obtain a more comprehensive overview of the influence DR components such as dynamically changing electricity prices have on behavioral consumption patterns of households.

1.2 Research Question

Demand Response programs assume perfectly rational behavior of consumers. However, consumers in the real world often lack information processing due to limited information (Hermsen et al., 2016). The introduction of smart metering devices is providing end-users of electricity with more extensive information about their own electricity usage, and more detailed information about the current electricity market situation through market-based dynamic electricity prices. The newly available information is likely to increase consumer informedness and encourage reflective and rational decision making, for those consumers capable of taking advantage of these information. As a consequence, consumer patterns of some households are likely to change in the future. It is the aim of our study to investigate these changing consumer patterns, in order to understand how energy services in dynamic pricing settings need to be arranged.

Therefore, our research question is as follows:

How can utility providers identify customer patterns for energy services in a dynamic price setting?

1.3 Structure

The theory section is going to clarify the underlying concepts used for this study. Furthermore, we will provide a conceptual overview of Demand Response programs, explain why we will focus on Real Time Pricing Demand Programs in specific, and in which relation this program stands compared to the behavioral theories.

Following the theory section, the Model section is going to provide an in-depth discussion about the academic work done on our specific subject. Afterwards, a conceptual model and a set of Hypotheses is presented that we wish to answer during the course of our study.

Furthermore, we will present our research setting and the statistical methods used in this study. Our data set was obtained from a pilot project of a Dutch energy supplier called ‘Qurrent Energie’, in which households were (and still are) provided with dynamic electricity prices. We are using a set of panel data regressions, a principal component analysis, and a k-means analysis to derive at our results.


11

We will discuss our findings independently, and explain their value when combining all parts. Lastly, a concluding part will bring all aspects of our study together and provide a short overview of the most essential parts of this research.


12

2. Theory and Model

2.1 Demand Response

2.1.1 Context

Electric utilities and power network companies have been pushed to reorganize their operations from vertically integrated mechanisms to open market systems (Bhattacharya et al., 2001). Additionally, the European Union enforced climate and energy-related policies expects all members to produce 20% of all generated energy from renewable resources (European Parliament, Council of the European Union, 2009). Due to the volatile nature of renewable resources, the uncertainties in energy generation are increased. Furthermore, energy consumption patterns have become increasingly uncertain, as the lifestyle of households is diversifying (Capgemini Consulting, 2015). Part of the change in consumer patterns are newly formed micro-grids, local energy grids of a small groups of households or a community capable to operate anonymously. Micro-grids are introducing the capability of households to self-generate electricity and directly communicate with the rest of their community in order to optimize the use of renewable electricity. Micro-grids have added a new level of complexity to the electricity environment of residential household, and require detailed information about electricity generation and use in order for households to be useful. The introduction of smart metering devices to the smart grid enables utility providers to supply the demanded information to households through informational electricity usage feedback on the one hand, and dynamic electricity prices on the other hand. Informational electricity usage feedback is capable of informing households about their consumption patterns, while a better overview of electricity prices enables households to gain a better understanding of the electricity market situation. Demand Response programs are a cheap alternative for balancing the electricity systems and have received increasing attention due to new technology-enabled ways to communicate with electricity end-users (Torriti et al., 2010).

2.1.2 Definition

Demand Response (DR) refers to a wide range of actions that can be taken on the customer side of the electricity meter, in order to respond to specific conditions within the electricity system (Torriti et al., 2010). Essentially, Demand Response (DR) is building upon the behavioral traits of energy consumers by offering incentives to change or shift electricity usage (Darby, 2012). More specific, demand response can be defined as deviations from the usual electricity usage of a residential household in response to changing electricity prices over time (Albadi & El-Saadany, 2008).

Program Classifications

The different DR programs can generally be classified into two main categories, namely Incentive-Based Programs (IBP) and Price-based Programs (PBP) (Albadi & El-Saadany, 2007). Figure 1 visualizes the categorization of different DR programs.


13

Incentive-Based Programs (IBP)

Incentive-Based Programs (IBP) are focused on providing an actual, direct incentive to the participating customers. Moreover, IBPs can be subdivided into Classical, and Market-Based IBPs. Classical IBPs provide customers with participation payments for their participation in the Program. For instance, residential households and small commercial customers can sign up for programs in which utilities are capable of remotely shutting down specific equipment on demand (direct control) or give their consumers specific load targets to adhere to (interruptible / curtailable programs). Furthermore, in market-based IBPs participants are rewarded money for the performance towards preset targets of the utility provider. Demand bidding allows participating customers to bid load reduction on the electricity wholesale market, while Emergency DR programs are providing financial rewards for measured load reductions during emergency situations.

Furthermore, Capacity Market Programs offer capacity payments for customers willing to restrain from electricity consumption when directed. Lastly, the Ancillary Services Market Programs offer customers the opportunity to offer load reductions in the intraday energy markets (Contreras et al., 2016).

Priced-Based Programs (PBP)

In PBP programs, consumers are offered dynamic pricing rates over time. In PBP programs electricity prices are significantly higher during peak-periods than during off-peak periods. Time of Use (TOU) pricing is changing the unit price of electricity during specific time periods at a fixed rate. Moreover, Critical Peak Pricing (CPP) is imposing a premium on electricity during special peak times, while Extreme Day Pricing (EDP) is imposing a premium on electricity during special high-demand days. Logically, Extreme Day CPP (ED-CPP) is a combination of both, in which electricity prices are increasing during special peak times on high-demand days. Lastly, in Real Time Pricing (RTP) schemes, customers are informed about, mostly hourly changing, varying electricity prices on a day-ahead or hour-ahead basis (Contreras et al., 2016). It has been widely agreed that RTP programs are the most effective DR programs for electricity markets as they are able to effectively communicate real market situations of electricity markets (Bloustein, 2005). Moreover, given the recent technological developments and an advanced metering infrastructure in Europe, effective communication of prices and detailed load profiles between utility providers and electricity consumers have become possible.

Dynamic pricing schemes are usually based on retail prices and reflect real-time system costs, thus, encouraging energy consumers to reduce or shift energy consumption during high wholesale price periods. Consequently, dynamic prices in a Real Time Pricing (RTP) scenario provide the best available information about the marginal value of electrical energy at a location, during a specific point in time. In order to successfully apply dynamic prices on a broad scale, households need to be provided with hourly meters capable of recording and communicating a household’s electricity usage per hour (Contreras et al., 2016).

Figure 1. Classification of DR

Programs (Albadi & El-Saadany, 2008)


14

The APX Electricity Market

The APX (NL) electricity market is an independent exchange for electricity trading on the spot market. Distributors, producers, traders and industrial end-users can trade on this market for day-ahead transactions and intra-day transactions (APX Group, 2016). On the day-ahead market, trading is conducted on the day before the delivery of the traded good. The participating entities can submit their orders electronically. Based on the submitted data for demand and supply of electricity, the electricity prices are calculated for every hour of the following day (APX Group, 2016). Hence, the hourly electricity prices are fluctuating based on the overall markets demand and supply for electricity. Consequently, this means that the dynamic prices of customers in RTP DR programs are influenced by usual electricity consumers exposed to TOU prices. This creates a dependency of the two different electricity consumer types, and implies that as the transition towards the use of dynamic prices progresses, the pricing structure of day-ahead market based dynamic prices will dramatically change.

Customer Response

Three general reactions of customers as a response to DR programs have been observed (US Department of Energy, 2006). First, energy consumers can engage in ‘Peak Shaving’, the reduction of electricity during critical peak periods, when prices are high. Second, energy consumers might engage in ‘Load Shifting’, by shifting part of their energy consuming activities to times of lower prices. Lastly, energy consumers can differ from their usual energy consumption patterns by engaging in on-site generation of electricity. Although the actual behavior might not change with the latter response type, from the utility perspective energy usage patterns of these consumers will change significantly (Albadi & El-Saadany, 2008).

Overview

Demand Response programs assume perfectly rational behavior of consumers. However, consumers in the real world often lack information processing due to limited information (Hermsen et al., 2016). The introduction of smart metering devices is providing end-users of electricity with more extensive information about their own electricity usage, and more detailed information about the current electricity market situation through market-based dynamic electricity prices. The newly available information is likely to increase consumer informedness and encourage reflective and rational decision making, for those consumers capable of taking advantage of these information. As a consequence, consumer patterns of some households are likely to change in the future.

The underlying question of DR programs is why some households are inelastic to electricity prices, while others are likely to respond. Another question is how varying levels of price sensitivity can be explained. The consumer choice theory and the theory of habitual behavior are aiming to explain rational decision making and irrational decision making in the form of habits, respectively. Both theories are explained in section 2.2. Subsequently, the scope of our study is going to be explained based on these theories and the concept of DR programs.


15

2.2 Consumption Behavior and Behavioral Change

2.2.1 Consumer Choice Theory

Consumer choice theory presents a set of principles that are able to explain consumers’ decisions to buy a given product (Deaton & Muellbauer, 1960). More specific, the theory aims to clarify the relationship between consumer preferences and time- and budget-constraints, in order to explain how consumers arrive at a final product decision (Deaton & Muellbauer, 1960). Consumer preferences allow consumers to weight different sets of goods according to the total satisfaction of consuming that product or good (Deaton & Muellbauer, 1960). Consumers making choices are generally exposed to financial- and time-related constraints that limit their ability to freely pick a product based on sole satisfaction (Salvatore, 2008). Financial constraints contain income constraints, the extent to which the relative buying power of one consumer changes towards a given product, and budget constraints, the general change in price for a product. Moreover, time-related constraints display the fact that consumers are not capable of freely choosing when to consume a product.

Taking this concept into the energy sector, varying income levels of household will generally lead to high-income households to consume more energy than low-income households, according to consumer choice theory. However, the preferences of a household to use electricity during a specific hour of the day can increase the price elasticity of that household during that time. Moreover, in a scenario with hourly changing electricity prices, electricity at 5PM can be seen as a substitute product to electricity at 6PM or 7PM. However, as storage options are currently limited and not widely implemented, a time-related constraint is influencing the substitution effect. Households that have to consume electricity at the time of purchase are not capable of substituting electricity at 7PM entirely for electricity at 5PM, they are merely able to shift part of their consumption patterns to a financially more satisfying time. Nevertheless, the consumer choice theory reveals that dynamically changing electricity prices should theoretically encourage consumers to shift part of their electricity consumption to times in which electricity is cheap. Lastly, it is worth mentioning that the height of the price variations plays an important role when applying the consumer choice theory to the energy sector (Schleich & Klobasa, 2013). While modest price variations will only attract households with lower incomes, higher price variations should be capable of attracting a bigger group of households.

The consumer choice theory displays how purchasing decisions are made at the individual level. According to this theory, it is reasonable to assume that consumers will react to electricity price variations over time. Additionally, the possibility to see the price development in advance should encourage some households to exchange electricity during high-price times with electricity at low-price times. Moreover, the above mentioned effect is likely to change its strength with different income levels, and during different times of the day. More specific, time-related constraints will enforce the need for certain amounts of electricity to be consumed during specific moments of the day, which leads to inconsistent price sensitivities of individuals during a day.

Much of the time-related constraints and the earlier mentioned price-inelasticity explained in the consumer choice theory can be regarded as habitual patterns of households. Although much of the electricity consumed at highly demanded times might be reducible or shiftable, daily routines and lifestyle patterns and switching costs prevent individuals from doing so. It is therefore necessary to


16

understand how habits lead to irrational decision making and how these patterns can be identified and broken by individuals, in the context of household energy consumption.

2.2.2 Habitual Behavior

Several approaches to creating a psychological model of energy consumption behavior as a basis for successful behavioral change have been developed. However, one model has been especially successful in forming a heuristic model of environmentally relevant behavior, by reviewing theory and findings from the entire discipline. This integrated model of environmentally relevant behavior is helpful in explaining why and how energy consumption related feedback influences individuals’ behaviors (Fischer, 2008; Matthies, 2005). The model by Matthies (2005) is displayed in Figure 2 below and differentiates between two general types of actions.

First, habits (environmentally detrimental habits) are actions that are not reflected upon and are performed similarly on a regular basis. Second, conscious decisions (the area above environmentally detrimental habits) are active decision making processes that can break habits through new evaluations of an individual’s values (Fischer, 2008).

By ensuring that individuals are provided with information capable of showing them that their actions are leading to an environmental problem and that this problem can be resolved by behavioral change, environmentally detrimental habits can be broken and replaced. This process is called ‘Norm Activation’ (Fischer, 2008). After the norm activation process, an individual proceeds to a process of evaluating different motives on how to act. In general, the heuristic model of environmental behavior distinguishes three motivations; personal environmental motives, social motives (expectations of others), and other motives (e.g. costs of new behavior). After weighing the importance of these three motives individually, a person proceeds by performing a more- or less environmental friendly actions. Although not specifically mentioned, the introduction of new information as a reminder of current behavioral patterns is necessary for an individual to break habitual behaviors (Fischer, 2008).

Therefore, the delivery of energy consumption feedback and information as a reminder of current behaviors, such as electricity prices, fills an ‘information vacuum’, which enables them to react and make more informed choices (Buchman et al., 2014). Naturally, consumers can have varying preferences for the type of feedback delivered to them, and might show varying levels of responsiveness. Consequently, the context in which feedback is delivered is equally important as the message itself. Taking the daily routines of households as an example, the same energy consumption feedback, such as the electricity price, delivered to a household might find varying levels of attention and acceptance during the afternoon, as opposed to the early morning or evening times, where other activities might be prioritized.

Information and feedback will only be significantly influencing energy consumption behavior in households that deem a specific form of information as relevant and valuable. Additionally, this insight emphasizes the importance of the context, and the type of person a piece of information is delivered to.


17

Figure 2. Heuristic model of environmentally relevant behavior (Matthies, 2005). Translated by Fisher (2008).

2.3 Summary of Concepts

The electricity consumers’ buying choice is dependent on personal preferences and budget- and time-constraints (Deaton & Muellbauer, 1960). Consequently, different types of households will base their evaluations whether to buy electricity or not on different personal preferences, electricity prices, and time-constraints that could potentially enforce them to make a rational decision. Additionally, behavioral energy consumption patterns of households can depend on habits and routines of households that are performed unconsciously (Fischer, 2008). In order for the energy consumers to break habitual behavior and change behavioral energy consumption patterns, they need to be more informed about dynamic pricing mechanisms (Matthies, 2005). Real Time Pricing Demand Response Programs (RTP DR) are building upon these behavioral aspects of energy consumers by communicating hourly dynamic prices that reflect the real marginal cost of electrical energy of the system (Contreras et al., 2016). By providing hourly changing prices for electricity to residential households, energy consumers have the increased ability to make buying decisions based on their preferences, budget- and time-constraints. However, the willingness to use electricity might heavily depend on households’ traits as preferences and preferred time of electricity usage are likely to vary. Furthermore, as the hourly dynamic prices in RTP DR programs are often delivered on a day-ahead basis, it is likely that some households are willing to break their habitual behaviors or change their behavioral patterns due to the constant exposure to their electricity usage behavior. For the same reason, behavioral electricity usage patterns of some individuals are likely to have changed due to the exposure to dynamic prices. Additionally, the evaluation of preferences and budget- and time-constraints of these households are likely to vary during different hours of the day, giving reason to believe that dynamic prices in RTP DR Programs might not successfully influence energy consumers’ behavioral patterns throughout the day. Lastly, if households’ attributes indeed influence energy consumers’ willingness to use electricity, and dynamic prices are only successfully influencing electricity usage patterns during specific times of the day, utility providers would be well advised to segment their consumer base, in order to improve their understanding of customer patterns.

The following sections are going to deepen the discussion about the current academic findings of three main topics of interest. Namely, the influence of household attributes on the willingness to use electricity in dynamic pricing environments, the influence of dynamic prices on behavioral electricity usage patterns, and the segmentation of electricity consumer groups based on household attributes.


18

2.4 Household characteristics and Household Willingness to Use Electricity

Household attributes have been proven to play an important role in explaining how willing energy users are to conserve energy (Nababan, 2015). Moreover, next to the influence of electricity prices on absolute electricity consumption, the effect of building characteristics and socio-demographic information have been a thoroughly studied field (Hayn et al., 2014; Schleich & Klobasa, 2013). In their study on Time-of-use (TOU) pricing and residential electricity demand in Germany, Schleich and Klobasa (2013) have found that the size of a households building is positively influencing energy consumption, while the number of appliances positively influences the electricity consumption in the period from May to October (Schleich & Klobasa, 2013). Furthermore, another study has found that the number of occupants of a household, the building size of a household, the building type, and the number of bedrooms positively influence the electricity consumption (Yohannis et al., 2008). Additionally, a study characterizing domestic electricity consumption patterns in Ireland has found that the number of occupants, the building size and the building type are positively influencing residential electricity consumption (McLoughlin et al., 2012).

However, contradicting findings have also been reported in other studies. A study investigating determinants of residential electricity consumption has found the number of occupants and the building size to be positively associated with electricity consumption, while the building type and the building age are not associated with electricity consumption (Kavousian et al., 2013). Furthermore, a study investigating short- and long-run price elasticities has found that the income is positively influencing peak and off-peak residential electricity consumption, while the household size does not show any significant influence (Filippini, 2011). Additionally, another study has found that the building type of a household is not influencing electricity consumption, while the building age has a direct influence over the electricity consumption (Statistik Austria, 2011).

Moreover, it is vital to note that the previously mentioned studies investigated the influence of household attributes in a flat or time-of-use pricing environment, and not in a dynamic pricing environment. However, a recent study investigating the influence of the building size and building type has found that the building size does not show a significant relationship with electricity consumption, while the building type has shown a significant influence (Alberini et al., 2011). Unfortunately, this study did not make an attempt to understand if the role of building characteristics and socio-demographic information is significantly different compared to TOU pricing settings. Additionally, our study aims to complement the recent studies investigating dynamic price settings by investigating what type of different groups and behaviors exist. Lastly, we contribute to the current academic findings by investigating the influence of household attribute on an hourly basis, which can improve the targeting of households if certain household attributes show not to be significant during each hour of the day.

Consequently, the partly contradicting results of former studies need to be reevaluated. One plausible explanation could be that the influence of building characteristics and socio-demographic information is varying depending on the time of the day, country under examination, and the type of the pricing mechanism. Another possibility is that different pricing environments influence the previously mentioned relationship. The aim of this study is to cover the currently persistent gap in the academic literature.

The following Research Question (RQ1) arises from our discussion:


19

How do household-level attributes influence a household’s willingness to use electrical energy throughout the day?

With the following Hypotheses:

H1: The attributes of a household significantly influence the willingness of a household to use electrical energy.

H1a: The number of occupants in a household has a significant positive relationship with the willingness of a household to use electrical energy.

H1b: The age of a household’s building has a significant positive relationship with the willingness of a household to use electrical energy.

H1c: The size of a household’s building has a significant positive relationship with the willingness of a household to use electrical energy.

H1d: The type of a household’s building has a significant influence on the willingness of a household to use electrical energy.

H1e: The roof insulation type of a household’s building has a significant positive relationship with the willingness of a household to use electrical energy.

H2: The relationship between the attributes of a household and the willingness of a household to use electrical energy is weaker in the dynamic pricing environment compared to the TOU pricing environment.

Source Variables of Interest

No. Occupants Building Size Building Age Building Type Other

Schleich and Klobasa (2013) + + No. Appliances

Yohannis et al. (2008) + + + + No. Bedrooms

McLoughlin et al. (2012) + + + + Composition

Kavousian et al. (2013) + + insignificant insignificant

Filippini (2011) insignificant + Income

Statistik Austria (2011) + + - insignificant

Alberini et al. (2011) insignificant +

+ = positive influence on energy consumption

- = negative influence on energy consumption

insignificant = no influence on energy consumption Table 1. Incidence of Household Attributes in Reviewed Studies and their Impact on Energy Consumption


20

2.5 Electricity Prices and Household Behavioral Consumption Patterns

Previous studies have convincingly suggested that dynamic pricing strategies would encourage the price-responsive demand to balance supply and demand of electricity (Borenstein et al., 2002). The first ever hourly Real Time Pricing (RTP) program for residential customers, however, was only conducted in 2011 in Chicago. The results of the demand estimates of the respective study suggested that dynamic prices only influence the residential electricity demand during peak hours. More specific, it was found that participating residential households were engaged in peak shaving behavior, but not in load shifting behavior (Allcott, 2011). Other studies examining electricity prices have found similar results, indicating that electricity prices only affect peak demand not off-peak demand (Schleich & Klobasa, 2013). Additionally, a study estimating the impact of time-of-use (TOU) pricing on Irish electricity demand has found that, while prices are affecting peak demand, they are not triggering load shifting behavior (Di Cosmo et al., 2014).

To the contrary, another paper investigating the impact of TOU pricing on electricity consumption has found that peak and off-peak consumption are negatively affected by increasing electricity prices (Filippini, 2011). Additionally, another study has found that demand based TOU electricity tariffs have decreased peak demand and shifting electricity demand from peak to off-peak periods (Bartush et al., 2011).

Next to widely contradicting results of the influence of electricity prices on electricity demand, Di Cosmo et al. (2014) have found significant evidence that the influence of electricity prices is different across household groups (Di Cosmo et al., 2014).

Several implications can be made from the earlier mentioned findings. First, it is likely that energy usage related behavioral patterns of households are affected by dynamic prices, because the constant update with prices should increase household informedness and awareness of energy usage, and should therefore potentially break habits and previous consumption patterns. Second, it is reasonable to assume that dynamic prices are having a significantly different impact on the consumption behavior of residential households compared to TOU prices, due to the constant reevaluation of preferences, and time- and budget – constraints needed. Although this assumption is self-evident, it yet needs to be proven. Third, given the contradicting findings of electricity price influence on demand, it is possible that electricity prices are only affecting households during specific time of the day. Or in other words, it is reasonable to assume that residential households are only capable of changing their behavioral electricity usage patterns during certain moments of the day. Understanding at what times of the day households are capable of changing consumption patterns would be vital for the improvement energy services. Fourth, given the finding that the influence of electricity prices is varying across household groups, it is vital to explore how households could be segmented.


How do dynamic prices influence a household’s capability to change electricity usage behavior?

With the following Hypotheses:

H3: The dynamic electricity price has a significantly different influence on the relative daily electricity usage, compared to the TOU electricity price.


21

H3a: The dynamic electricity price during the day (08:00-16:00) has no significant relationship with the relative daily electricity usage during day-times.

H3b: The dynamic electricity price during peak-times (17:00-19:00) has a negative significant relationship with the relative daily electricity usage during peak-times.

H3c: The dynamic electricity prices encourage the households to shift loads.

2.6 Household Segmentation for DR improvements

As visible from previous elaborations and discussion, studies investigating the effect of electricity prices on energy consumption behavior have been presenting variations in effect, sizes and significance of different electricity pricing types. Additionally, studies have argued that the effectiveness of energy information is often not generalizable across cultures and demographic groups (Fischer, 2008). Households with higher income, higher education levels, and higher electricity use are more reactive to energy consumption behavior change than other groups (Wilhite & Ling, 1995; Vine et al., 2013). In contrast, other feedback studies could not find a clear link between household-level characteristics and price effectiveness (Brandon & Lewis, 1999). Consequently, the question arises whether specific, predefined consumer groups have varying levels of capability to react to dynamic prices.

Earlier attempts to appropriately segment energy consumer groups have either focused on usage-based clustering or on the impact of socio-demographic factors and the equipment with electric appliances and new technologies (Hino et al., 2013; Kwac et al., 2014; Hayn et al., 2014). A detailed segmentation of household electricity usage provides the opportunity to better reflect on future energy systems and might be useful to create new load profiles (Hayn et al., 2014).

Creating household segments based on near-static characteristics can potentially help utility providers to model households’ electricity consumption and behavioral load patterns (Hayn et al., 2014). Therefore, it is reasonable to assume that the earlier investigated household attributes are capable of segmenting electricity consumers with varying levels of capability in simple but an effective manner.

Moreover, cluster analysis is a central element in marketing and widely used for market segmentations and the identification of consumers with similar needs and behaviors (Hayn et al., 2014). With increasingly detailed energy consumption and household data available, cluster analysis has received increasing attention in academic studies of the energy sector. Next to academic contributions discussing the use of supervised clustering techniques for electricity load profiles and consumption patterns, several studies have suggested to follow unsupervised clustering techniques, similar to those used in market segmentation research (Hayn et al., 2014; Hino et al., 2013; Kwac et al., 2014).


How can utility providers sufficiently segment their energy consumers into groups with varying capabilities to react to dynamic prices?

With the following Hypothesis: H4: A segmentatin based on household attribtues is capable of identifying different electricity usage patterns


22

3. Methodology and Results

3.1 Data and Descriptives

Our analysis is based on household-level hourly consumption data of the Dutch energy supplier ‘Qurrent Energie’. The data were collected during the month of March 2016, as part of a pilot project on dynamic pricing schemes of the company. During the pilot project, some of the participating households were confronted with hourly electricity prices that were fixed to the electricity price variations of the APX energy market. All households receiving dynamic electricity prices were exposed to the same price per hour. In order to assess whether dynamic prices led to a change in household electricity consumption, the response of the households on the dynamic prices was recorded in hourly blocks, as the net usage of electricity in kilowatt per hour.

Next to the electricity usage, the participants answered a survey about specific attributes and characteristics of their households. The results enable us to gain a deep understanding of the differences and similarities between household groups, as well as the opportunity to investigate the presence of significantly different household responses to hourly changing prices. In specific, the survey asked the households about the number of occupants, the type of the house, the insulation of the house, the location, the size, the heating type, and the use of solar panels.

Additionally, all participating households had access to a web-based dashboard provided by ‘Qurrent Energie’. This web-based dashboard allows customers of the company to get a better, real-time understanding of their energy-related activities. The customers of ‘Qurrent Energie’ are able to display their past energy consumption, to understand their PV panel production, and access the dynamic prices for electricity, if applicable to the household, through the web application. Moreover, the web-application of the company can be used via phone, tablet and conventional computers. Screenshots of the web application can be found in the Appendix in section 7.1.

3.1.1 Sample

Our sample consists of 225 participating households, with 75 households used for the actual pilot project, and an additional 150 households as the control group. The survey concerning the household attributes was answered by 73 households of the actual pilot project, and 35 households of the control group, making the first part of our study (RQ1) less generalizable. All households were exposed to the same applications, with the only difference being the dynamic pricing scheme that was only shown to the treatment group. The control group was exposed to a time-of-use electricity price during the time of the study, and consisted of usual ‘Qurrent Energie’ consumers that were not specifically included in the project. An overview of all included variables, the reason for their presence and their computation can be found in the following section.


23

3.1.2 Variables

Measurement of Electricity Usage

Absolute Electricity Usage (kWh)

Energy consumption was measured as the ‘Net Electricity Usage’ in kilowatt per hour. The total amount of energy consumed by a household subtracted by the total amount of energy produced by a household within an hour, in kilowatt per hour.

Independent Variable – Electricity Price

We have investigated the influence of dynamic and TOU electricity prices on electricity usage. Our treatment group has received hourly changing electricity prices, while our control group received TOU electricity prices.

The dynamic electricity prices are the APX electricity market prices plus any additional surcharges that are generally applicable to electricity end-users. Hence, the final dynamic price of this study is composed as:

Dynamic Price = (APX market price + BS + ET + SES + GOO) * (1 + VAT)

The TOU electricity prices are composed of a marginal fee of the electricity provider ‘Qurrent Energie’ that reflects the forward price and a risk premium. Moreover, the TOU electricity price changes between peak- and off-peak times.

Time-Of-Use Price = ((forward price + risk premium) + BS + ET + SES + GOO) * (1 + VAT)

Fee Description Amount

Balancing surcharges (BS)

Energy suppliers in the Netherlands are charged with fees for the use, management, and balancing of the national grid. This amount

is directly forwarded to the households.

€0,0071 / kWh

Energy Tax (ET)

Households in the Netherlands are taxed on their energy use per kilowatt hour.

€0,1007 / kWh

Sustainable Energy Surcharge (SES)

An additional tax on household energy consumption, used to stimulate investments into sustainable energy sources.

€0,0056 / kWh

Guarantee of Origin (GOO)

A fee charged by the Energy supplier to ensure that only energy from renewable resources is bought for Qurrent households.

€0,0028 / kWh

Value added tax (VAT) 21% of total Table 2. The Real Time Electricity Price Composition of ‘Qurrent Energie’

Household Attributes

The importance of household attributes as indicators of lifestyle patterns has been described by several studies investigating household energy consumption profiles (Hayn et al., 2014). More specifically, household attributes such as the number of occupants, the size of the building, and ownership of


24

electrically run household applications can indicate the height of a household’s energy consumption, but also its flexibility when aiming to decrease or shift energy consumption (Hayn et al., 2014).

Number of Household Occupants - One of the most evident household attributes is the number of persons living in a household (Hayn et al., 2014). Indications about total energy consumption, and also the capability to react to prices resulting from the higher probability that someone is inside the house can be made. The number of household occupants varies between one and six persons in the sample of our study.

Building Size – The building size of a household can potentially define load shifting capabilities of households, since the availability of more rooms, lights, appliances etc. enables individuals to decrease or shift consumption more easily compared to smaller households. The building size is measured as the total floor size in square meters (m2).

PV Panel Ownership – The ownership of PV panels is significantly influencing the net electricity usage (DV) of households during times where the sun is shining. Additionally, it is reasonable to assume that PV panel owners might be less price sensitive to the dynamic prices than non-owners. Hence, it is obligatory for our study to make a distinction between households that own PV panels, and those that do not. PV panel ownership is included as a dummy variable, indicating 1 as ownership and 0 as non-ownership.

Building Age – The age of a building can indicate the age of its electricity consuming appliances, such as fridges, freezers and washing machines. Additionally, younger households are likelier to possess smart technologies and more efficient appliances. The building age is taken into consideration as a categorical variable with four general categories:

1. 27 years or younger 2. Between 40 and 28 years 3. Between 50 and 41 years 4. 51 years or older

Electric Heating – The presence of electrically run heaters is accounted for with a dummy variable, indicating electric heating with 1, and other types of heating with 0.

Building Type – The building type a household is residing in, partitioned into five categories. The values of this categorical variable range from one to five.

1. Apartment (‘Appartement’) 2. Row House (‘Tussenwoning’) 3. Detached House (‘Vrijstaand’) 4. Corner House (‘Hoekwoning’) 5. Semi-detached House (‘Twee onder een kap’)

Terrain Type – Terrain Type is a categorical variable that indicates whether a household is located in an ‘Urban’, ‘Suburban’, or ‘Rural’ area. The categorization benchmark was set as follows:

1. Rural – less than 1000 inhabitants per km2 at household location


25

2. Suburban – between 1000 and 3000 inhabitants per km2 at household location 3. Urban – more than 3000 inhabitants per km2 at household location

Household Attributes – Descriptives

Table three and table four present the descriptive statistics of the household attributes. The average number of household occupants is 2.63 for the treatment group and 2.97 in the control group. Another feature in which both the samples differentiate from each other is the maximum number of occupants in a household, which is six for the treatment group and eight for the control group. The average building age of the sample of this study ranges between the categories two and three, which is similar to the control group, between 28 and 50 years. The average building size of the treatment sample is 158.2m2

and comparable to the control group with the minimum household size being 58m2 and the maximum household size being 550m2. Furthermore, the average building type is ranging between type 3 and type 4, which means that most households were either living in house type ‘Vrijstaand’ or in house type ‘Hoekwoning’. Additionally, 18.5% of all households are using electrically run heating in the treatment group, while 21.9% of the households in the control group are using electrically run heating. Lastly, the availability of roof insulation is at 85.3% in the treatment group, while it is at 76% in the control group.

Summary Statistics - Treatment Group

N Mean Median Min Max

Persons 85 2.63 2 1 6 Building Age 85 2.739 3 1 4 Building Size 85 158.2 140 58 550 Building Type 85 3.416 3 1 5 Heating Type 85 1.185 2 1 2 Roof Insulation 85 1.853 2 1 2

Solar Influx 720 37.74 1 0 259 Table 3. Summary Statistics of Household Attributes – Treatment Group

Summary Statistics - Control Group


Persons 35 2.967 3 1 8 Building Age 35 2.239 2 1 4 Building Size 35 153.5 123 50 600 Building Type 35 3.02 3 1 5 Heating Type 35 1.219 2 1 2 Roof Insulation 35 1.76 2 1 2

Solar Influx 720 37.74 1 0 259 Table 4. Summary Statistics of Household Attributes – Control Group

Weather Variable

Solar Influx – Potential PV production amount and sunshine intensity measured as the Solar influx in J/cm2. Solar influx was measured as hourly data from the Ministry of Climatology of the Netherlands


26

(KNMI). Moreover, solar influx information for taken from 20 different weather stations. Each household received the solar influx information from the weather station closest to the household.

The data recorded from the pilot project was analyzed in three steps in order to answer the three main research questions of this study. First, this study will conduct a set of panel data regression analyses with the aim to understand which household characteristics influence a household’s willingness to use electrical energy given different price levels (RQ1). Second, this study will conduct another panel data regression in order to determine how dynamic prices influence the relative consumption of households throughout the day in order to determine behavioral patterns and potential load shifting capabilities (RQ2). Third, a principal component analysis and an unsupervised cluster analysis will be conducted with the aim to sufficiently segment the households into distinguishable groups (RQ3). Lastly, the new household segments are evaluated in terms of their dynamic price responsiveness and DR recommendations would be given. The following sections are going to provide a description of the analysis types used in this study. Further details about the statistical methods can be found in the respective sections.

3.2 Analysis of Willingness to Use Electricity (RQ1)

3.2.1 Method

Panel Data Regression

Panel data regression is a statistical approach that measures the behavior of entities such as firms, industries or households and across time (Baltagi, 2013). Unlike time-series or cross-sectional regressions, a panel data regression has a double-subscript on its variables, enabling us to take into account cross-sections, as well as time-dimensions during the analysis (Baltagi, 2013). A panel data regression is denoted as:

Yi,t = αi + X’i,t β + εi,t i = 1 …. N; t = 1 …. T

with i denoting an observed entity, such as countries, firms or individuals, and t denoting time. Hence, the i subscript denotes the cross-sectional dimension, while the t subscript denotes the time dimension. Moreover, α is a scalar, X’I,t is the ith observation of explanatory variable X’ at time t. Furthermore, β denotes the coefficient estimate, or the main effect, of explanatory variable X on the dependent variable Y (Baltagi, 2013). Lastly, ε is an error term of the model, which captures variation in the dependent variable not explained by the independent variables.

Additionally, this study performs a Hausman test, in order to understand whether random effects or fixed effects need to be used for the panel data regression. This approach ensures the sufficient measurement of the interrelatedness of the variables of this study, while accounting for individual differences across households and time. The Hausman test is denoted by the following formula:

H = (βFE − βRE)′[Var(βFE) − Var(βRE)]−1(βFE − βRE)

Where βFE is the fixed effects estimate and βRE is the random effects estimate. The null Hypothesis is claiming that β is consistent across our panels. Therefore, fixed effects estimates have to be used in case H0 appears to be true, and random effects should be used in case we can reject the null Hypothesis.


27

Controlling for fixed effects obliges us to explore the relationship between the proposed individual and outcome variables within a household. Given that household level attributes have the same level of influence on the dependent variable or if omitted variables are included in the regression, controlling for fixed effects would be recommended for the panel data regression. A random effect model assumes that entity error terms are not correlated with the independent variables (Greene, 2008).

Assuming that each household has his very own price sensitivity and energy usage patterns, a panel data regression is capable of assessing the statistical significance of the individual variables within each separate panel. More specifically, panel data analysis allows a study to account for individual-level heterogeneity, making it possible to control for variables that are not measurable per se across households or across time (Baltagi, 2013). Hence, this type of regression analysis is capable of analyzing the change in demand response of each household separately over time. Panel data regression analysis has been used to examine Research Questions one and two of our study. Further details about the exact equations can be found in the respective sections.

Data Preparation

In order to answer the first research question, the collected hourly consumption data were merged with household-level data from the questionnaires and subsequently split into 24 separate data sets, one data set for every hour of the day, for the panel data regression analysis. Consequently, 24 separate panel data regressions were conducted for the data, in order to assess the relevance of the influence of household attributes on the price sensitivity of households for every hour of the day.

Equation 1

In order to investigate the influence of household attributes on the price sensitivity of households for every hour of the day, 24 separate panel data regression analyses have been conducted. This approach enables us to make statements about the significance of specific household characteristics throughout the day. By understanding how the influence of household attributes on household willingness to use electricity is changing, we are able to explain which households are likely to respond to price signals during a specific moment in time. The following dependent variable was computed to represent the ‘willingness of households to use electricity’:

Willingness to Use Electrical Energy (Price sensitivity)

WTU = Usagei,t / Pricet

The reason for the computation of our dependent variable is to factor out the influence of the price on the absolute usage of electricity. In this way, our study is capable to compare the absolute usage between our treatment and control group in a more sophisticated manner. Hence, changes in consumption behavior that have occurred due to the introduction of dynamic prices, such as load shifting behavior, can be controlled for. A central assumption of our dependent variable is that the entire load of a household is elastic, which is not the case in reality. In fact, only a fraction of a household’s load is elastic to prices, making it necessary to collect more data about the household’s appliances, in order to understand the percentage of elastic load each household has. We have elaborated on this issue in our limitations in section 5.2.


28

The hypotheses one and two are covered by this equation. This study derives the following equation for the above-mentioned analytical approach.

Equation 1: Investigating the Influence of Hourly Dynamic Prices on the Electricity Usage of Households During Different Times of the Day

Y WTUi,t = αi + β1*occupants + β2*buildingAge + β3*buildingSize + β4*buildingType + β5*RoofInsulation + β6*solarInfluxi,t + εi,t

where the dependent variable Y is the willingness of a household i to use electrical energy for a specific hour during day t. The αi reflects household-fixed effects that record potential time-invariant, household-level heterogeneity related to the willingness of a household to use electrical energy (Y). Moreover, occupant is the number of people permanently occupying the household with the main effect β1. The coefficient estimate of β1 indicates the main effect of the number of occupants on the household willingness to use energy. Building age represents the age of a household’s building and β2 is the main effect of the variable on the dependent variable. Additionally, the same effect is investigated for the size of a household’s building, the type of the building, and the type of the roof insulation with the effects of the variables on the dependent variable represented as β3, β4 and β5. Lastly, β6 reflects the main effect of the control variable solar influx as a proxy for PV panel production for a household i at time t on the household’s net electricity usage. Lastly, εi,t is an error term.

Moreover, the panel data regressions for the treatment group and the control group are compared and an overall regression analysis including interaction terms for each independent variable with sample membership (treatment/control) is used to determine whether the household attributes have a significantly different relationship with the dependent variable in the dynamic pricing setting compared to the TOU pricing setting.

Proving Statistical differences in the regression coefficients between treatment and control group

A dummy variable called ‘dynamic’ indicates whether a household was part of the treatment or the control group. Additionally, five interaction variables were created in combination with the ‘dynamic’ variable. The interaction terms ‘Dynamic*Persons’, ‘Dynamic*BAge’, ‘Dynamic*BSize’, ‘Dynamic*BType’, and ‘Dynamic*RInsulation’ measure whether the influence of one of the predictor variables is significantly different in the two groups, and to what amount the regression coefficient, and hence the direction of the relationship, is different. Logically, the above-mentioned variables test the hypothesis that

H0: βx treatment = βx control.

3.2.3 Research Question 1 – Analysis

Descriptive Statistics

Table five displays the descriptive statistics for the treatment group and the control group of this study. A total of 47,275 observations can be made from the treatment group. Moreover, each of the 75 participating households has 720 observations made over one month. The average usage per price unit


29

is 2.115 kWh in the treatment group, while it is 3.524 kWh in the control group. This gives reason to believe that the average price sensitivity of the treatment group can be considered higher than the average price sensitivity of the control group.

Summary Statistics


usage/price - treatment group 47,827 2.115 1 -34 37 usage/price – control group 24,480 3.524 2 -30 91.74

Table 5. Summary Statistics of Willingness To Use Electricity

Moreover, correlations between the variables are described in Table six. By observing the correlations of the variables we are able to preliminarily test for cases of multicollinearity, hence, to test whether certain predictor variables are influencing each other in a way that would bias the regression results. It is worth pointing out that the variable House Type and Persons have a positive correlation of 0.26. This can be attributed to the fact that certain household compositions, such as families, are likely to appear in specific types of houses, then in, for instance, apartments. However, none of the predictor variables showed correlations above the threshold of 0.5. We can thus infer that no signs of multicollinearity exist.

1 2 3 4 5 6 7

1. Usage / price 1

2. Persons 0,24 1

3. Building Age -0,17 0,14 1

4. Building Size 0,2 0 -0,13 1

5. Building Type 0,13 0,26 -0,04 0,06 1

6. Roof Insulation -0,11 0,15 0,01 0,04 0,08 1

7. Solar Influx -0,29 0 0 0 -0,01 0 1 Table 6. Correlation Matrix

Results

A panel data regression analysis was performed in order to clarify the hypothesized relationships of H1, H1a, H1b, H1c, H1d, H1e and H2. Before the initial analysis, the regression methods were subject to a Hausman test, in order to determine the necessity to include fixed effects in the panel data regression or not. For all 24 data sets, the p-value showed to be above the threshold of .05. Therefore, it is recommended to use random effects for the panel data regression analysis (Green, 2008).

The Influence of Household Attributes on the Willingness of Households to Use Electricity

The first part of our analysis aims to assess how household attributes influence the willingness of households to use electrical energy in a dynamic pricing setting. The following elaboration of this section will clarify the influence of each household variable individually. Table seven displays the overall panel data regression results over the day.


30

First, the number of occupants of a household has a constant significant positive relationship with a household’s willingness to use electrical energy. The regression coefficient is 0.51 (0.14) and is significant at the 99% level. This means that an increase of one household occupant results in an increased willingness to use electrical energy of approximately 0.51kWh per €. Hence, it is reasonable to claim that H1a is true and can be supported. The number of household occupants does have a significant positive relationship with the willingness of that household to use electrical energy.

Second, the categorical variable building age has a significant positive relationship with a household’s willingness to use electrical energy. Hence, the results indicate that the older a household’s building, the higher the willingness of the household to use electrical energy. The regression coefficient is 0.25 (0.13) and is significant at the 90% level. This means that the building age is significantly increasing a household’s willingness to use electrical energy. Hence, we can regard H1b as true.

Third, the size of a household’s building has a significant positive relationship with a household’s willingness to use electrical energy. The regression coefficient is 0.01 (0.002) and is significant at the 95% level. This means that the household willingness to use electricity increases by 0.01kWh per € per one square meter increase of the building size in a given hour. Hence, we can regard H1c as true.


31

Panel Data Regression Results

Dependent variable:

Usage/Price

Persons 0.51***

(0.14)

Building Age 0.25*

(0.13)

Building Size 0.01**

(0.002)

Building Type -0.05

(0.14)

Roof Insulation -0.25

(0.44)

Solar Influx -0.01***

(0.0003)

Constant 2.13**

(1.01)

Observations 47,827

R2 0.05

Adjusted R2 0.05

F Statistic 436.36*** (df = 6; 47820)

Note: *p<0.1; **p<0.05; ***p<0.01

Table 7. Panel Data Regression Results RQ1

Hypotheses H1d and H1e cannot be supported according to our panel data regression results. However, a more fine-grained analysis of 24 different panel data regressions has been conducted, in order to examine during which hours of the day the household attributes are influencing the willingness to use electricity.

The Difference of the Influence of Household Attributes Between Dynamic Pricing and TOU Pricing Groups.

In an attempt to examine the differences of the previously investigated relationship, this study will proceed to compare the average willingness to use electrical energy over the course of the day. Subsequently, this study will compare the results of the 24 panel data regressions with the results of the control group. Lastly, a test is conducted, in order to statistically prove that the influence of the household attributes is significantly different between the two groups; control and treatment group.

Comparing the Average Willingness to Use Electrical Energy

It is reasonable to assume that dynamic electricity prices change a household’s willingness to use electricity, as the constant reminder of the electricity price can increase the price sensitivity. Figure three is visualizing the average willingness of the two samples to use electricity over the course of the day. It can be generally stated that the willingness of those households exposed to dynamic prices, to use electrical energy, is lower than the willingness of the control group. It is especially remarkable to see


32

that the willingness of the households of the treatment group is even decreasing between 10AM and 3PM, while it stays generally constant for the control group. Additionally, it is observable that the slope is steeper when households start to increase the willingness to consume electrical energy between 5PM and 6PM in the treatment group compared to the control group. Hence, from 5PM onwards, households are slowly going back to their old willingness to use electrical energy, but never fully reach it until 11PM.

Figure 3. Comparing the Willingness to Use Electricity between Treatment and Control Group

When comparing the regression results of the control group to the regression results of the treatment group, it becomes evident that the influence of the household attributes on the willingness to consume electrical energy is different. Figure four is visualizing the differences in the regression results between both groups.

When examining the 24 different panel data regression coefficients between treatment and control group in tables eight and nine in the Appendix, it becomes evident that the household attributes have a very different influence on both groups. The observed differences are either time-differentiated or show an opposite relationship with the willingness to use electrical energy. Figure four provides a rough overview of the outcomes of the 24 panel data regressions. Additionally, we can see that some variables, such as the building size, have a distinct gap in which the influence of the variable is not relevant for household willingness to use electricity. Moreover, the variables building type and roof insulation showed to be insignificant in the overall panel data regression, but have specific times of the day where they indeed are significant.

0

1

2

3

4

5

6

1 3 5 7 9 11 13 15 17 19 21 23Willingn

esstouseelectricity

(usage/pric

e)

Time(h)

Willingnesstouseelectricityoveraday

TreatmentGroup

ControlGroup


33

Figure 4. Comparison of Regression Results between Treatment and Control Group

Thus, we conclude that there is an observable difference in the regression results between the treatment and the control group. In order to statistically prove that this difference exists, and whether the strength of the relationship between the household variables and household willingness to use energy is weaker in a dynamic pricing setting compared to a TOU pricing setting, an additional statistical test was conducted.

Proving Statistical Differences in the Regression Coefficients between Treatment and Control Group

When interpreting the results in table eight, it is vital to recall that the variable dynamic takes the value 1 for households of the treatment group and the value 0 for households of the control group. Consequently, the households of the control group are the omitted group for the results.

The variable ‘Dynamic*Persons’ corresponds to the difference of the slopes for the variable Persons between the treatment and the control group (slope treatment group – slope control group). The regression coefficient is 0.33 (0.03) and is significant at the 99% level. Hence, the number of occupants is increasing household willingness to use energy by 0.33kWh per € more in the treatment group than in the control group. This is a surprising finding, as this study has assumed that the influence of household attributes will be weaker in a dynamic pricing setting.

Moreover, the regression coefficient of ‘Dynamic*BAge’ is -0.22 (0.03) and is significant at the 99% level. Hence, the building age is having a significantly different influence on household willingness to use energy in the treatment group than in the control group. More specific, the slope is -0.22 weaker in the treatment group than in the control group.

The regression coefficient of ‘Dynamic*BSize’ is -0.02 (0.0004) significant at the 99% level. Hence, the size of a household’s building is positively influencing a household’s willingness to use energy in a dynamic pricing environment, but to a lower extent than in a TOU pricing environment.

Furthermore, the regression coefficient of ‘Dynamic*BType’ is -0.29 (0.02) and significant at the 99% level. Consequently, it is safe to say that the building type of a household is influencing household willingness to use energy at a significantly lower degree in the treatment group than in the control group.


34

Lastly, the influence of the variable Roof Insulation has not been found to be significantly different between the treatment and the control group.

Consequently, this study concludes that Hypothesis 2 can be partly supported by our findings. The relationship between the attributes of a household and the willingness to use energy is weaker in a dynamic pricing setting compared to a TOU pricing setting with one exception. The number of household occupants has a significantly higher influence on the dependent variable in the dynamic pricing setting. Explanations why this variable in specific has shown a different relationship can be found in the Discussion section of this study.

Table 8. Testing Influence of Household Attributes between Treatment and Control Group

Coefficient Test Results

Willingness to use electrical energy (Usage/Price)

Y

Dynamic -0.40** (0.19)

Persons 0.18*** (0.02)

Building Age 0.06*** (0.02)

Building Size 0.02*** (0.0003)

Building Type 0.20*** (0.02)

Roof Insulation -0.47 (0.06)

Dynamic*Persons 0.33*** (0.03)

Dynamic*BAge -0.22*** (0.03)

Dynamic*BSize -0.02*** (0.0004)

Dynamic*BType -0.25*** (0.02)

Dynamic*RInsulation 0.29 (0.08)

Solar Influx -0.01*** (0.0002)

Constant 2.08*** (0.25)

Observations 72,273 N R2

108 0.13

Adjusted R2 0.13 F Statistic 803.61*** (df = 12; 72259) Note: *p<0.1; **p<0.05; ***p<0.01


35

Robustness of the Models and Explanatory Power

The adjusted R2 of the 24 panel data regression models ranged between 2% and 15%, meaning that the model used for the regression analysis only explained between 2% and 15% of the changes in the households’ willingness to use energy. Moreover, the explanatory power of our model is, compared to the other times of the day, higher during the evening times. Although such low levels of explanatory power indicate a weakness of the model used in this study, it can be argued that comparable studies have found similar levels of accuracy. Filippini (2011) found R2 values between 25.2% and .6%, depending on the investigation of peak and off-peak energy consumption. Hence, we can consider the explanatory power of our model as weak but acceptable. Another possible explanation for the low explanatory power is the fact that the dependent variable household net electricity usage depends on several factors, such as sunshine hours and PV panel production, and is therefore harder to determine than pure electricity consumption.

The Influence of the Control Variable Solar Influx on the Willingness to Use Electricity

The control variable solar influx could only be implemented in the models for 8AM until 8PM. The reason for this is that the variable Solar Influx was constant of value 0 during times where the sun was not shining and could therefore only be included during the times where the sun was shining. Furthermore, Solar Influx in J/cm3 is significantly influencing net electricity usage between 8AM and 8PM. Since net electricity usage is computed as electricity consumption subtracted by electricity production, it is self-explanatory why this relationship exists.

3.3 Analysis of Relative Electricity Usage (RQ2)

3.3.1 Method

In order to investigate RQ2, another panel data regression has been conducted. Further details about the panel data regression analysis can be read in section 3.2.

Data Preparation

In order to answer the second research question, the collected hourly consumption data were analyzed in several steps. First, two separate panel data regressions were conducted in order to examine the different results between the dynamic pricing and the TOU pricing group. Subsequently, a test determines whether dynamic electricity prices and TOU electricity prices have a significantly different influence on the relative electricity usage of households. Lastly, the data set is a split into 24 different data sets, one data set for each hour of the day. By investigating the relative daily electricity usage during each hour of the day, this study aims to investigate during what hours of the day households are changing their behavioral usage patterns due to varying price levels.

Equation 2

The electricity usage behavior is represented by the relative electricity usage at hour t, as a share of the daily total electricity usage.

Relative daily electricity usage = Usagei,t / Daily sum of usagei


36

We have aimed to measure the effect of prices on the electricity usage behavior of households. The electricity usage behavior is computed as the relative electricity usage of a specific day, in percent. Hence, the total sum of the relative daily electricity usage should always add up to 1. This variable will enable us to understand if households are able to change their consumption patterns and participate in load shifting behavior.

Investigating Differences in the Influence of Dynamic and TOU Prices on Behavioral Usage Patterns

In a first step, we are going to compare the differences in the regression models of the dynamic pricing and the TOU pricing group. Subsequently, we will conduct an overall regression analysis including an interaction term of prices and treatment group membership, in order to determine whether the influence of the two different types of prices is statistically significant. More specific, a regression analysis including a dummy variable called ‘dynamic’ indicates whether a household was part of the treatment or the control group. Additionally, an interaction variable was included in combination with the ‘dynamic’ variable. The interaction term ‘Dynamic*Price’ measures whether the influence of the predictor variable is significantly different in the two groups and to what amount the regression coefficient, and hence the direction of the relationship is different. Logically, the above-mentioned variables test the hypothesis that

H0: βx treatment = βx control.

Investigating Changes in Behavioral Patterns during each Hour of the Day

Subsequently, we will proceed to investigate the relationship between dynamic electricity prices and the relative electricity usage of households during each hour of the day. By looking at each hour independently, we hope to identify specific times during the day, in which the households of our sample are adapting their behavior more rigorously in order to take advantage of varying price levels.

Additionally, if the relative daily share during a specific hour decreases with increasing dynamic prices, it is reasonable to assume that a load shift has occurred, as the absolute consumption at the moment in time must have decreased and the absolute consumption during other times of the day must have increased. In the setting of our study, this means we will investigate whether the dynamic electricity prices of ‘Qurrent Energie’ enable the participating households to shift part of their daily consumption to a different time with low electric charges.

The Hypothesis three is covered by this equation. We derive the following equation for the above-mentioned analytical approach.

Equation 2: Investigating the Influence of Dynamic Prices on the Daily Share of Electricity Usage of Households

YUsage%i,t = αi + β1*pricet + β2*SolarInfluxi,t + εi,t

where the dependent variable Y is the relative daily electricity usage of a household i during a specific moment of the day t. The αi reflects household-fixed effects that record potential time-invariant, household-level heterogeneity related to the net electricity usage (Y). Moreover, pricet is the price displayed to the households during the time t, with the main effect β1. The coefficient estimate of β1


37

indicates the main effect of the price on the relative daily electricity usage of a household i. Moreover, the control variables AvSolarInflux is included with the main effect β2. Lastly, εi,t is an error term.

3.3.2 Research Question 2 – Analysis

Descriptive Statistics

Figure five and six display the dynamic and the TOU pricing scheme charged to the households during the month of March of the pilot project. The TOU pricing scheme consisted of a peak price (08:00 – 20:00) of €0.1868/kWh and an off-peak price (21:00 – 07:00) of €0.1744/kWh. The dynamic pricing scheme was coupled to the APX day-ahead market prices. Naturally, the dynamic pricing scheme of this study is more complex than the TOU pricing scheme. Figure five and table 11 in the Appendix provide a better overview of different prices charged to the treatment group. The highest recorded price in March was charged at 7PM. Moreover, the highest average price was charged at 8PM during the course of the experiment. Lastly, the lowest recorded price was charged during 9AM, while the lowest average price was charged at 5AM.

Figure 5. Dynamic Pricing Scheme with Min and Max.

0,14

0,15

0,16

0,17

0,18

0,19

0,2

00:00:00

01:00:00

02:00:00

03:00:00

04:00:00

05:00:00

06:00:00

07:00:00

08:00:00

09:00:00

10:00:00

11:00:00

12:00:00

13:00:00

14:00:00

15:00:00

16:00:00

17:00:00

18:00:00

19:00:00

20:00:00

21:00:00

22:00:00

23:00:00

Price

(€)

Time(h)

AverageDynamicPricewithMinandMax

DynamicPrice

Series2

Series3

Min Max


38

Figure 6. TOU Pricing Scheme.

Results

The Influence of Dynamic Electricity Prices on the Relative Electricity Usage

In order to understand whether dynamic electricity prices are influencing the electricity usage behavior of the participating households of the pilot project, it is essential to examine the relative usage profile. Figure seven visualizes the relative load profiles of the treatment and the control group. Looking at Figure seven, it becomes evident that the electricity usage behavior of households is different in a dynamic pricing setting compared to a TOU pricing setting. More specific, when comparing the two graphs, it is observable that part of the relative electricity usage during the afternoon has been shifted to the late evening hours. Moreover, another, smaller shift occurs between 10AM and 2PM in which households use less electricity in the morning in order to consume it immediately after noon.

Figure 7. The Average Relative Load Profiles of Treatment and Control Group.

0,173

0,178

0,183

00:00:00

01:00:00

02:00:00

03:00:00

04:00:00

05:00:00

06:00:00

07:00:00

08:00:00

09:00:00

10:00:00

11:00:00

12:00:00

13:00:00

14:00:00

15:00:00

16:00:00

17:00:00

18:00:00

19:00:00

20:00:00

21:00:00

22:00:00

23:00:00

Price

(€)

Time(h)

TOUPricingScheme

Series1

0

0,01

0,02

0,03

0,04

0,05

0,06

0,07

0,08

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24

relativ

eelectricityusage

Time(h)

AverageRelativeLoadProfileTreatmentGroup

ControlGroup

TOU Price


39

This finding gives reason to believe that dynamic electricity prices have a different relationship to the relative electricity usage of households than TOU prices. The following elaborations are aimed to clarify whether the influence of TOU prices is significantly different from the influence of dynamic prices on the relative electricity usage of households.

The Difference of the Influence of Electricity Prices Between Dynamic Pricing and TOU Pricing Groups.

In an attempt to examine the differences of the previously investigated relationship, we will proceed to compare the overall panel data regression models of the treatment and the control group. Subsequently, we are going to statistically prove that the influence of the dynamic prices is significantly different from the influence of TOU electricity prices.

Table nine shows the overall panel data regressions comparison between the treatment and the control group. The results show that the dynamic electricity prices have a positive significant relationship with a regression coefficient of 0.78 (0.28), while the TOU electricity prices have a positive significant relationship with a regression coefficient of 0.95 (0.04). The relationships are significant at the 95% and 99% level respectively. Consequently, we can see that the dynamic electricity prices have a less positive relationship with the relative electricity usage of households compared to the TOU electricity prices.


Relative Electricity Usage

Treatment Control

Dynamic Price 0.78*** (0.28)

TOU Price 0.95*** (0.04) SolarInflux -0.0002*** -0.0000***

(0.0000) (0.0000) Constant -0.10** -0.13*** (0.05) (0.01)

Observations 47,275 103,515

N

R2

73

0.34

150

0.48

Adjusted R2 0.34 0.48

Note: *p<0.1; **p<0.05; ***p<0.01 Table 9. Comparison of Regression Results


40

Proving Statistical Differences in the Regression Coefficients Between Treatment and Control Group

When interpreting the results in table ten, it is vital to recall that the variable dynamic takes the value 1 for households of the treatment group and the value 0 for households of the control group. Consequently, the households of the control group are the omitted group for the results.

Moreover, the variable ‘Dynamic*Price’ corresponds to the difference of the slopes for the variable Price between the treatment and the control group (slope treatment group – slope control group). The regression coefficient is -0.16 (0.31) and is significant at the 95% level. Hence, the dynamic electricity price has a significantly lower positive influence in the relative electricity usage of households than the TOU electricity price.

Therefore, we can confirm that the influence of dynamic electricity prices throughout the entire day is significantly less positive than the influence of the TOU prices. Hence, Hypothesis 3 can be supported; dynamic electricity prices have a significantly different influence on the relative electricity usage of households compared to TOU electricity prices.

Coefficient Test Results

Relative Electricity Usage

Dynamic 0.03** (0.05) Price 0.95*** (0.27)

Dynamic*Price -0.16** (0.31) Solar Influx -0.00 (0.0000) Constant -0.13*** (0.05)

Observations 155,676 N R2

220 0.02

Adjusted R2 0.02 F Statistic 45.92*** (df = 4; 155452)

*p<0.1; **p<0.05; ***p<0.01 Table 10. Statistical Significance of Differences in Regression Coefficients between Treatment and Control Group


41

Investigating the Influence of Dynamic Prices during Every Hour of the Day

In order to statistically prove that the load shifts can be attributed to the dynamic prices, we will proceed to perform 24 different panel data regressions, one for each hour of the day. By examining the influence of the prices on the relative electricity usage, this study aims to clarify changes in behavioral patterns of the households.

Additionally, the panel data regression methods were subject to a Hausman test, in order to determine the necessity to include fixed effects in the panel data regression or not. The p-value showed to be above the threshold of .05. Therefore, it is safe to use random effects for the panel data regression analyses. The results of the panel data regressions can be found in table eleven.

Looking at the results, it becomes evident that the dynamic electricity prices have a significant negative relationship with the relative electricity usage of households between 8AM and 5PM. The strongest relationship occurs at 1PM, with a regression coefficient of -0.08 (0.02) and is significant at the 99% level. Hence, this study can claim that the households of the ‘Qurrent Energie’ pilot project successfully manage to increase their relative electricity usage when the dynamic electricity price is low and decrease their relative electricity consumption when the dynamic electricity price is high, between 8AM and 5PM. Consequently, the findings of this study cannot support Hypothesis 3a. The dynamic electricity prices between 8AM and 5PM do show a significant relationship with the relative electricity usage, a negative one.

Moreover, the findings of this study cannot support Hypothesis 3b as well. The dynamic electricity prices between 5PM and 7PM only show a significant negative relationship with the relative electricity usage of households at 5PM but not at 6PM or 7PM.

Lastly, the deviations in the relative electricity usage visible in Figure seven occur exactly during the times in which the panel data regressions of our study prove that households are capable of appropriately adapting their behavior to increasing and decreasing electricity prices. Hence, there is sufficient evidence to claim that the households of the ‘Qurrent Energie’ pilot are shifting part of their electricity usage to different times of the day. As our results have shown, it is likely that these shifts are encouraged by the dynamic prices. Hence, Hypothesis 3c can be supported. The dynamic electricity prices encourage the households to shift loads.


42


Dependent variable: Relative Electricity Usage (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) (21) (22) (23) (24)

price1 -0.03*** (0.01)

price2 0.01 (0.01)

price3 -0.005 (0.01)

price4 -0.01** (0.01)

price5 -0.01 (0.01)

price6 0.02*** (0.005)

price7 -0.0003 (0.01)

price8 -0.01* (0.004)

price9 -0.02***

(0.01) price10 -0.05***

(0.01) price11 -0.06***

(0.01) price12 -0.05***

(0.02) price13 -0.08***

(0.02) price14 -0.07***

(0.01) price15 -0.07***

(0.01) price16 -0.03***

(0.01) price17 -0.02*

(0.01) price18 -0.01

(0.01) price19 0.01

(0.01) price20 -0.01

(0.01) price21 0.003

(0.01) price22 -0.03**

(0.01)

price23 0.17

(0.22)

price24 -0.97**

(0.39) SolarInflux -0.000 -0.00*** -0.00*** -0.00*** -0.00*** -0.00*** -0.00*** -0.00*** -0.00*** -0.00*** -0.01*** 0.001**

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.004) Constant 0.04*** 0.04*** 0.04*** 0.04*** 0.04*** 0.04*** 0.04*** 0.04*** 0.05*** 0.06*** 0.06*** 0.06*** 0.06*** 0.06*** 0.06*** 0.05*** 0.05*** 0.04*** 0.04*** 0.04*** 0.04*** 0.04*** 0.01 0.43***

(0.002) (0.002) (0.002) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002) (0.002) (0.003) (0.003) (0.003) (0.002) (0.003) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.4) (0.10)

Observations 1,990 1,923 1,987 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,999 2,100 R2 0.05 0.06 0.07 0.09 0.09 0.04 0.03 0.04 0.07 0.10 0.11 0.12 0.11 0.12 0.09 0.05 0.03 0.06 0.04 0.03 0.03 0.03 0.01 0.01 Adjusted R2 0.05 0.06 0.07 0.09 0.09 0.04 0.03 0.04 0.07 0.10 0.11 0.12 0.11 0.12 0.09 0.05 0.03 0.06 0.04 0.03 0.03 0.03 0.01 0.01

F Statistic 25.75***(df = 4; 1985)

31.98***(df = 4; 1918)

36.54***(df = 4; 1982)

50.71***(df = 4; 1986)

50.54***(df = 4; 1986)

22.46***(df = 4; 1986)

17.70***(df = 4; 1986)

15.81***(df = 5; 1985)

28.56***(df = 5; 1985)

42.55***(df = 5; 1985)

50.98***(df = 5; 1985)

52.26***(df = 5; 1985)

49.79***(df = 5; 1985)

53.20***(df = 5; 1985)

38.66***(df = 5; 1985)

20.77***(df = 5; 1985)

13.66***(df = 5; 1985)

24.25***(df = 5; 1985)

15.13***(df = 5; 1985)

17.83***(df = 4; 1986)

14.54***(df = 4; 1986)

13.58***(df = 4; 1986)

-43.86 (df = 4; 1994)

2.90**(df = 4;

2095) Table 11. 24h Panel Data Regression Results of Dynamic Prices


43

Robustness of the Models and Explanatory Power

The adjusted R2 of the 24 panel data regression models has ranged between 1% and 12%, meaning that the model used for the regression analysis has only explained between 1% and 12% of the changes in the households’ willingness to use energy.

The Influence of the Control Variable Solar Influx on the Relative Electricity Usage

The solar influx does show a significant negative relationship with the dependent variable. Therefore, it is safe to assume that the solar power production of a household, and the weather conditions on a day significantly influence a household’s relative usage patterns throughout the day, in the ‘Qurrent Energie’ pilot project.

3.4 Analysis of Household Segmentation (RQ3)

3.4.1 Method

Principal Component Analysis

A Principal Component Analysis (PCA) analyzes a data set, with the goal to extract important information from the data and to express this information as a set of new orthogonal variables called principal components (Abdi & Williams, 2010). A PCA has three general goals. First, a principal component analysis aims to extract the most important information from the data set. Second, a PCA aims to compress the size of the data set by keeping only this important information. Third, a PCA aims to simplify the description of the data set and to analyze the structure of the observations and variables (Abdi & Williams, 2010). Subsequently, PCA computes new orthogonal variables out of the variables of the data set called principal components. The first principal component is supposed to have the largest possible variance and thus explains the largest part of the inertia of the data set. The second and all following principal components are computed under the constraint of being orthogonal to the previous component, in order to have the largest possible inertia.

This study will conduct a principal component analysis (PCA) and plot the first two principal axes of the PCA with the outlined clusters. If the outlined clusters are strong at the selected levels, the clustering plot should not display substantial overlaps.

Cluster Analysis

Following the investigations on household attributes and dynamic prices for electricity during different times of the day, this study aims to segment the participating households into objectively distinguishable household archetypes. The underlying goal of this segmentation strategy is to prove the assumption that different groups of energy consumers have varying capabilities to react to dynamic electricity prices.

A clustering analysis is a technique that is used to segment elements in data sets in a way that similar elements are assigned to the same cluster, while elements with different attributes are assigned to a different cluster. In this way, clustering is particularly useful to efficiently search for groups of elements in a multi-dimensional data set (Bhatia, 2004). By following an unsupervised clustering approach, we


44

aim to identify groups of households that are different in their household composition and their buildings conditions.

Many different types of clustering methods are discussed in the academic literature and can be generally categorized into hierarchical and non-hierarchical clustering methods. Hierarchical clustering methods aim to cluster a data set in a bottom-up approach, starting with the same number of clusters as instances available in the data set and gradually grouping similar instances together. Non-hierarchical clustering methods aim to determine the number of clusters prior to clustering and assign instances to the cluster centroids that are closest to them. We will use a k-means clustering approach in order to identify clearly distinguishable household segments.

K-means clustering

The k-means clustering technique is one of the earliest clustering techniques in the literature (Anderberg, 1973). In k-means clustering, clustering is based on the identification of K elements in the data set that are used to establish an initial representation of clusters. Moreover, these K elements form the cluster seeds, to which other elements are then assigned to form clusters (Bhatia, 2004). The ultimate goal of the k-means algorithm is to assign each element of a data set to a cluster, with the aim to maximize the between-cluster variation while minimizing the within-cluster variation.

Choosing the Correct Cluster Solution

This study will follow a k-means clustering analysis, in order to come up with meaningful household groups. In a first step, a set of tests will be conducted with the aim to identify the correct cluster solution for the data set of this study. More specific, we will proceed to compare the sum of squared error (SSE) for a number of cluster solutions. SSE is defined as the sum of the squared distance between a cluster element and the centroid of its cluster (Peeples, 2011). Hence, the SSE can be regarded as a global measure of error that decreases with an increasing number of clusters. Additionally, we will compare the SSE over cluster solutions with the SSE of 250 randomized versions of the original input data, in order to understand if the original data set has an underlying structure that allows it to be appropriately clustered. Lastly, we will examine the absolute differences between the actual and the random SSE against the tested cluster solutions. Ideally, the cluster solution with the highest difference between the SSE of the original data set and the randomized dataset should be chosen for the k-means analysis.

Subsequently, we will perform the aforementioned k-means clustering analysis. The k-means approach aims to assign each household to the cluster with the closest cluster center. The k-means analysis was run with a variety of different cluster sizes, ranging from 2 clusters up to 15 clusters. The Hypothesis four is addressed through the proposed clustering analysis. The results can be seen in the following section.

3.4.2 Research Question 3 - Analysis

Before our initial analysis, a PCA was conducted with the aim to identify the most powerful variables for the cluster analysis of our households. The PCA results can be seen in Table 16 in the Appendix. We have proceeded to select the four strongest variables of the first two components, as these are already capable of explaining 50% of the point variability of the entire data set.


45

The four variables used for the cluster analysis were as follows:

Persons – The number of people permanently occupying a household.

Building Size – The size of a household’s building measured in square meters.

Building Type – One of five different types of buildings a household is living in.

Terrain Type – A categorization indicating whether the building of a household is located in a rural, suburban or urban area.

Choosing the Correct Cluster Solution

In a first step, we have compared the within group SSE and the log of the within group SSE between the actual data set and 250 different randomized versions of the data set. In order to successfully prove that the original data set possesses an underlying structure, the within group SSE and the log of the within group SSE need to be decreasing faster than the within group SSE of the respective randomized samples. An overview of this test can be found in figure eight and figure nine.

Figure 8. and 9. Within Group SSE of Actual and 250 Randomized Data Sets against 15 Cluster Solutions

Looking at the two figures above, it becomes evident that the within group SSE and the log of the within group SSE is lower in the actual data set. Consequently, we can claim that the chosen data set possesses an underlying structure and clusters are present. However, it is hard to determine the ‘elbow’ in the scree plot, which normally indicates the ideal cluster solution for this data set.

Another way to evaluate the appropriate cluster solution is to examine the absolute differences between the original and random SSE against the chosen cluster solutions (Peeples, 2011). The appropriate cluster solution will be the solution at which the actual SSE differs the most from the mean of the random SSE. In order to visualize this comparison, figure ten and figure eleven have been developed.


46

Figure 10. and 11. The difference of Within Group SSE of Actual and 250 randomized Data sets against 15 Cluster Solutions.

Inspecting both figures, it becomes evident that the highest recorded differences between the actual and the random SSE is in both cases the five cluster solution. Therefore, we will follow a five cluster solution for the remainder of the analysis.

Principal Component Analysis

Subsequently, a principal component analysis has been conducted, in order to better visualize the cluster solutions along the two strongest principal components. The results of the PCA can be seen in Table 12.

Table 12. PCA Results

Consequently, the five cluster solution is visualized in Figure 12.

PC1 PC2 PC3 PC4

Standard deviation 1,4419 1,0237 0,9343 0,0000

Proportion of Variance 0,5198 0,2620 0,2182 0,0000

Cumulative Proportion 0,5198 0,7817 1,0000 1,0000

Persons 0,4427 -0,4998 0,6156 0,4187

Building Size -0,6935 -0,0138 -0,0024 0,7204

Building Type 0,5546 0,1914 -0,6076 0,5355

Terrain Type 0,1250 0,8446 0,5019 0,1381


47

Figure 12. Visualization of Five-Cluster Solution of K-Means Analysis Along the Two Strongest Principal Components

As visible from figure 12, the data set was segmented into clearly distinguishable clusters. We can thus confirm that the chosen clustering solution is strong. The two strongest principal components of the PCA explain 78.17% of the point variability within our cluster.

The following elaborations are going to deepen the analysis of the identified clusters. More specific, we are going to explain and define the different clusters. Lastly, we are going to examine whether the cluster analysis based on household attributes was able to segment groups that are also distinguishing themselves in the way they are capable of reacting to dynamic electricity prices.


48

Interpreting and Profiling the Clusters

Each of the five identified clusters of this study should be seen as an archetype that categorizes the households across the sample of the ‘Qurrent Energie’ pilot program. The following section is going to provide a general overview of the attributes of each cluster and will compare how these attributes compare between the archetypes. An overview of the distribution of the 75 clustered households of this study across the identified clusters is provided in Figure 13.

Since Cluster 1 is only comprised of 2 households, this cluster will be ignored for the following elaborations.

Furthermore, ‘Cluster 2’ is characterized by a small group composition of 2 persons on average, and a wide variety of different house types, that also differentiate themselves substantially in their size. Additionally, this group primarily contains urban households.

‘Cluster 3’ represents 21% of the participating households. Similar to ‘Cluster 2’, this cluster primarily contains two-person households, which either live in detached houses or corner houses, that are approximately 150m2 big. Moreover, this cluster is comprised of a mix of urban, rural and suburban households.

‘Cluster 4’ represents 17% of the participating households. Other than the previous clusters, this group of households consists of four-person groups on average, that are only residing in urban or suburban areas. Moreover, the majority of this group resides in semi-detached and row-houses.

Lastly, ‘Cluster 5’ represents the largest group of households with 36%. This group represents households of two to four persons that primarily live in row houses, detached houses, and semi-detached houses. The terrain type is distributed across all three types.

Household Occupancy

The average household occupancy across all clusters was 2.63 people per household. As visible from Figure 14. The biggest household groups can be found in Cluster 4, with 4.15 occupants on average. Disregarding Cluster 1, the smallest group of households can be found in Cluster 2 with exactly 2 occupants on average.

Cluster13%

Cluster223%

Cluster321%Cluster4

17%

Cluster536%

ClusterOverview

Figure 13. Overview of the Distribution of Clusters


49

Figure 14. Average Household Occupancy for Each Cluster

Building Type

The distribution of building types across the five identified clusters can be found in Figure 15. As visible, the biggest proportion of detached and corner houses can be found in Cluster three. This is an interesting finding, as the cluster with the highest amount of household occupants does not possess the largest type of building. Moreover, it can be observed that the majority of semi-detached houses can be found in Cluster four, while the Building Type composition in Cluster two and three are mixed.

Figure 15. The Distribution of Building Types for Each Cluster

Building Size

The average building size across all clusters can be observed in Figure 16. The average building size across all clusters was 158.2m2 per household. The biggest houses are owned by Cluster 2 with an average building size of about 200m2. Moreover, the smallest average building size can be found in Cluster 4, with an average size of about 130m2. This is a surprising finding, since Cluster 2 has an average household occupancy of 2 persons, while Cluster 4 has an average household occupancy of 4.15 persons.

0

1

2

3

4

5

Cluster1 Cluster2 Cluster3 Cluster4 Cluster5

NumberofOccupants

0

0,2

0,4

0,6

0,8

1

1 2 3 4 5

DistributionofBuildingTypepercluster

Semi-detachedHouse

Apartment

RowHouse

DetachedHouse

CornerHouse


50

Figure 16. Average Building Size for Each Cluster

Terrain Type

The distribution of terrain type across the five identified clusters can be found in Figure 17. It is worth noting that Cluster 2, despite having the biggest building sizes, has the highest proportion of urban households. Especially when taking into consideration that Cluster 2 has an average household occupancy rate of 2 persons, it becomes evident that the income per person of this household might be unproportionally higher than in the rest of the sample. Furthermore, it is worth noting that Cluster 3 has the highest proportion of households based in rural areas, which might explain why this Cluster also has the highest proportion of detached houses. Lastly, Cluster 4 has the highest share of households based in Suburban areas.

Figure 17. Distribution of Terrain Type per Cluster

Assessing Differences in Behavioral Patterns of the Identified Clusters during the Qurrent Project

The relative electricity usage across clusters can be seen in Figure 18. Looking at the graph, it is recognizable that Cluster 3 is showing extreme behavioral deviations in their relative electricity usage. Comparing it to the relative electricity usage of our control group, it becomes evident that Cluster 3 is exchanging part of its daily energy consumption to early noon hours, and saving high amounts during afternoon hours, when the electricity prices are starting to rise again. In addition, Cluster 2 is showing a similarly favorable relative load profile for the dynamic prices. More specific, Cluster 2 has an above

0

50

100

150

200

250


BuildingSize

0

0,2

0,4

0,6

0,8

1


DistributionofTerrainTypeperCluster

Rural%

Suburban%

Urban%


51

average relative electricity usage during early morning and late evening times, which are usually the times in which the prices are comparably low. Lastly, Cluster 4 and 5 have very similar relative load profiles compared to the control group. We can thus assume that the dynamic electricity prices had little influence on the behavioral patterns of these households.

Figure 18. Differences in the Relative Electricity Usage Across the Identified Clusters


52

4. Discussion

4.1 The Influence of Household Attributes on Willingness to Use Electricity

We have investigated the overall influence of our household attributes on the willingness of households to use electricity, in order to understand what aspects of a household explain variations on electricity usage. Our results have shown that the number of household occupants, the building age, and the building size positively influence the household willingness to use electricity. Hence, the more people living in a household, the higher the willingness to use electricity; the older a building, the higher the willingness to use electricity; and the bigger a building, the higher the willingness to use electricity. However, previous studies have found the influence of building age to be negatively related with electricity usage, due to the presence of more electrical appliances in modern houses (Statistik Austria, 2011). One very obvious explanation for these results is that the absolute electricity usage is simply higher in these type of households (McLoughlin et al., 2012; Yohannis et al., 2008). However, it can also be argued that individuals living in households with these attributes perceive their individual impact as irrelevant, due to fixed factors such as a building’s energy inefficiency due to its age, or the unfavorable behavior of other household occupants (Ingham et al., 1974). As a result, households with high values for these variables can be considered less likely to respond to dynamic electricity prices. For the same reasons, we can assume that the households are less responsive to the delivery of new electricity-related information and base their consumption patterns on habits rather than rational decisions (Fischer, 2008). Therefore, a more fine-grained delivery of electricity usage information should counter this effect and decrease the strengths of the relationship between household attributes and electricity usage (Matthies, 2005). One possible way to increase the level of delivered information is the introduction of hourly changing prices (dynamic prices) for TOU prices, as dynamic price signals constantly remind households about their electricity consumption patterns and reflect the current electricity market situation.

Therefore, we have examined whether the influence of household attributes on the willingness to use electricity is significantly different in a dynamic pricing environment compared to a TOU pricing environment. According to our findings, this is indeed the case, as all household characteristics other than the availability of roof insulation have a significantly different influence on the willingness to use electricity. More specific, the variables building age and building size have a weaker positive relationship with willingness to use electricity in the dynamic pricing setting than in the TOU pricing setting. This supports our earlier argument that a more fine-grained delivery of electricity information decreases the strength of the relationship between household attributes and willingness to use electricity, as we can regard the dynamic electricity price as an enhanced way of communicating the electricity market situation to residential households. Based on these findings, we can also make the assumption that the investigated household attributes reflect lifestyle patterns of households (Hayn et al., 2014). With the relationship of the household attributes on willingness to use electricity decreasing, it is possible that these lifestyle patterns have changed or have become less relevant predictors of electricity usage. However, one exception contrasting this assumption are the number of household occupants, whose influence on willingness to use electricity has actually increased. A possible reason for this could be the excitement of individuals to be part of the dynamic pricing sample, and thereby consume more electricity than usual, as described by the Hawthorne-effect (McCarney et al., 2007).


53

Lastly, we have conducted an hourly analysis to investigate the effect of household attributes during different hours of the day. Our results show that the influence of household attributes is not constantly significant throughout the day. Hence, consumer targeting initiatives in dynamic pricing settings should follow a dynamic segmentation model that is capable of taking into consideration a different set of variables for different times of the day. Another remarkable result is the influence of the building size on willingness to use electricity. According to our results, building size positively influences the willingness to use electricity from 1AM to 7AM and from 4PM to 12PM. Hence, the size of a household’s building has not sufficiently explained variations in the price sensitivity of residential households from 8AM to 3PM. The influence of a household’s building size can therefore be taken into consideration in the morning and evening times when determining household willingness to use electricity, but not during day times. Based on our previous assumption that our household attributes reflect lifestyle patterns, we can assume that the building size between 8AM and 3PM is not sufficiently explaining variations in the willingness to use electricity, because the households of our treatment group have started to engage in more rational electricity usage behavior, based on prices rather than habits.

In order to gain a better understanding of the dynamic electricity prices, another analysis has been conducted, examining the influence of electricity prices on electricity usage behavior.

4.2 The Influence of Dynamic Prices on Usage Behavior

The two main goals for the analysis of the second section were to understand whether dynamic prices and TOU prices influence consumption behavior in a significantly different way and during which times of the day consumers are actually capable to change their behavioral consumption patterns as a reaction to electricity prices.

Our findings have shown that the relative load profiles of the treatment and control group are different from each other. More specific, the share of electricity usage is lower and higher at a few specific points of the day in the treatment group. This gives reason to believe that the households exposed to dynamic prices are shifting their loads by changing their behavioral patterns as a response to the electricity prices. Additionally, we have found that the influence of the dynamic electricity prices is significantly different from the influence of the TOU prices on the relative electricity usage of the households of our study. More precisely, the influence of the dynamic electricity prices is lower than the influence of the TOU prices. Consequently, we can assume that the households of our treatment group have started to change their electricity usage behavioral patterns due to the introduction of the dynamic prices.

In order to determine during which times of the day the dynamic price is influencing electricity usage behavior, we have run 24 different panel data regressions to get a better overview. Our results have shown that the dynamic prices are negatively influencing relative usage behavior between 8AM and 5PM. We can therefore imply that the households of our sample are capable of, or willing to, change their electricity usage behavior based on price fluctuations during these times but not during other times of the day. As the time window in which households are capable to change usage behavior is overlapping with the time window in which we were able to observe deviations in behavioral patterns between the treatment and the control group, we can imply that the dynamic prices have encouraged load shifting behavior in our sample.


54

Several unexpected findings have been made during this analysis. First, despite almost all other studies suggesting peak times to be the most crucial times for households to save energy as a response to price signals, there is only a partially significant influence of the dynamic prices during peak times (5PM-7PM) at 5PM (Schleich & Klobasa, 2013; Allcott, 2011; Bartush et al., 2011). However, this study has observed changes in the relative usage rather than the absolute usage, which enables us to make statements about behavioral patterns but not total consumption. Hence, we have found that residential households are not changing behavioral electricity usage patterns after 5PM. Additionally, we expected an insignificant relationship between the usage behavior during day times and electricity prices due to the absence of individuals in many households. However, our results are opposite. The participating households were especially capable of changing their electricity usage behavior during these times. One possible explanation is that individuals staying at home during the day, while other members are leaving the household, might feel a higher sense of control over their electricity usage and the overall impact of their behavior, and have therefore properly reacted to price signals during these times.

We have seen that the influence of household attributes on the willingness to use electricity is significant, but varying throughout the day. Additionally, households are only capable of changing their electricity usage patterns during specific moments of the day. Hence, it reasonable to assume that the willingness to use electricity and behavioral change based on changes in electricity prices are closely related to each other. Consequently, in order to sufficiently target residential households for energy services in dynamic pricing settings, utility providers need to find a way to properly segment these households by objective variables such as household attributes in a way that reflect the households’ capability to change electricity usage patterns. Additionally, this would provide evidence for our assumption that household attributes reflect lifestyle patterns of electricity consumers. The third part of our study has aimed to do so.

4.3 Clustering Usage Behavior Based on Household Attributes

Our PCA analysis has found that out of all available variables of our sample, the building size, and type, the number of occupants and the terrain type are the most significant when properly segmenting the households of our study. Five clearly distinguishable clusters have been found, out of which the first cluster is disregarded due to its size of two households. The second cluster is characterized by having the biggest building size despite the lowest number of occupants. Additionally, this cluster has the highest share of urban household locations. It is likely, though not provable, that the combination of highest building size with lowest number of occupants in urban areas is an indication for a high income group. Moreover, the third cluster is characterized by having the highest number of ‘detached’ houses, which are actual stand-alone houses. Additionally, this cluster has the highest amount of households living in rural areas. Hence, this cluster can be characterized by its rural locations and independent housing. The fourth cluster has the highest amount of household occupants despite having the smallest household sizes, which are primarily semi-detached houses. The share of urban and suburban households is also high for this cluster. We can therefore assume that low- and mid-income families from urban and sub-urban areas are represented by this group. Lastly, cluster five cannot be concretely defined, as all variables are showing evenly distributed results.

In the last step, we have investigated how capable these households are in changing their behavioral patterns. Our results show that clusters two and three are showing clearly distinguishable behavioral electricity usage patterns from the rest of our sample. Two conclusions can be made from these findings.


55

First, by segmenting households based on their household attributes, we were able to isolate two groups that are potentially engaging with the introduced dynamic prices by changing their behavioral patterns. We can claim that cluster 2 reduced their load during the day, while cluster 3 engaged in load shifting behavior. Second, we can claim that our cluster analysis confirms our assumption that household attributes reflect electricity usage lifestyle patterns.

4.4 Bringing the Findings Together

Our results have shown that changes in behavioral electricity usage patterns, based on the introduction of hourly changing electricity prices, have occurred between 8AM and 5PM. As we have pointed out in section 4.1, the willingness to use electricity increases strongly during evening times. This suggests that households during the evening are less price sensitive, and therefore less likely to change behavioral patterns as a response to dynamic electricity prices. One reason for the decreased price sensitivity could be that individuals perceive their impact on the households’ total electricity usage as weak in households with higher numbers of occupants, or bigger buildings and are therefore less responsive to the prices. Another valid explanation would be that electricity usage becomes more preferable during evening times, and therefore increases the willingness to consume electricity at higher prices. Additionally, the observation that household attributes have a lower explanatory power when dynamic prices are introduced was made. Especially during the specific times in which the dynamic prices have shown a significant influence on electricity usage patterns, the explanatory power of our household attributes has either decreased or was diminished. Sticking to an earlier assumption that the household attributes of our study reflect lifestyle patterns and habits, we can suggest that the introduction of hourly changing prices has encouraged households to change or break their behavioral patterns. Hence, moving towards a more rational decision-making process. The outcome of our household attributes-based cluster analysis can partly support the assumption that our household attributes reflect customer patterns, as two clusters have been found that substantially differentiate themselves in their behavioral electricity usage patterns. To conclude, we have observed that households have changed their behavioral patterns as a response to dynamic prices. It is possible to claim that the increased level of communicated information in the form of hourly prices that reflect real-world electricity market situations has enabled the households to do so. Perhaps, the additional information has increased energy usage awareness and a sense of higher individual impact. As a result, future energy services should provide more fine-grained information about energy use, ideally on an individual level rather than a household level, in order to unlock higher customer responsiveness. This applies to the general provision of information, as well as dynamic electricity prices (e.g. personalized prices). A possible consequence would be that customers will change their patterns also during other times of the day, after 5PM. Additionally, dynamic price variations could be increased in order for households to deem electricity prices as relevant. Essentially, all of these recommendations for energy services are addressing the need for utility providers to communicate a more comprehensive and complete picture of the energy sector and energy usage to residential households, thereby encouraging households to move towards a more rational decision making process. However, in the case that all residential households will receive dynamic electricity prices and everyone starts to change its consumption behavior based on these prices, the price variations on the APX electricity market will become flatter, as demand during low-price times increases and demand during high-prices times decreases. Consequently, this would reverse the effect of dynamic electricity prices


56

based on electricity market prices at least during the time of 8AM to 5PM, and households are likely to go back to previous habitual patterns. Therefore, future energy services should also focus on non-monetary incentives and an increased level of information provided to households. Additionally, future Demand Response Programs should explore the options to use personalized or artificially varying electricity prices to engage their customers in a dialogue.

4.5 Limitations and Recommendations for Future Research

For several reasons, the exact magnitude of household attributes, dynamic prices and household electricity usage might not generalize to other settings and time periods.

The participants of the ‘Qurrent Energie’ project were a selected group of people who are potentially more price elastic than the general population. Moreover, the variance in hourly electricity prices during the course of the project is lower than the variance of electricity prices used in other experiments. This is especially important because a higher price variation has been proven to increase potential gains from dynamic prices and because higher price uncertainties could encourage consumers to devote more attention to prices, which would have an impact on electricity usage (Allcott, 2011; Faruqui & Palmer, 2012; Schleich & Klobasa, 2013). It is likely that in reality households will only accept to change their usage patterns based on prices if the effect on the electricity bill can provide meaningful savings.

Additionally, our study was subject to several limitations that should be addressed in future research. The absolute electricity usage of the households during the ‘Qurrent Energie’ project was recorded as consumption subtracted by PV production. It is essential for more accurate examinations to investigate the influence on pure consumption, in order to fully understand consumption behavior and price sensitivity.

Moreover, we cannot claim with absolute confidence that the behavioral patterns of our households have changed due to the introduction of dynamic pricing signals, as our study was missing pre-treatment data for the analysis. Future studies should therefore include pre-treatment data in order to prove changes in behavioral patterns with absolute confidence. Other studies on this topic have suggested to follow a difference-in-differences analysis, in order to arrive at robust findings (Di Cosmo et al., 2014).

Furthermore, we have found that the importance of household attributes as determinants of a household’s energy usage patterns is decreasing. Hence, future research needs to explore what other types of variables have become more relevant for future dynamic pricing environments. Previous research has suggested a broad set of different variable types that can be explored, such as socio-economic information (e.g. education, income, occupation, richness of neighborhood, social grade), appliance ownership (e.g. electric vehicles, electric heating, other electric appliances) and belief systems (e.g. environmental awareness, motivation to change). Additionally, other than recommended, our segmentation is not dynamic but static. Future research needs to develop a dynamic segmentation model based on various variable types like socio-demographic information and appliance ownership variables in order to determine the most effective segmentation approach for each hour separately.

Lastly, in the computation of our dependent variable ‘Willingness to Use Electricity’ we are assuming that households are capable to change their entire load based on electricity prices. In reality, only a minor part of the consumption load is elastic. One way to solve this issue is to calculate how much of a household’s load is elastic by gathering more detailed information about the electrical appliance


57

ownership of the households. One example of such an approach is the MIT RED database, which is releasing how much percentage of elastic load certain types of households have.


58

5. Conclusion

5.1 General Conclusion

The main aim of our study was to examine customer patterns of residential households in a dynamic price setting. With the development of new technologies such as smart meters, utility providers have the opportunity to inherently change the way they are communicating with their customers. One way the utility providers are aiming to improve the interaction with their customers is by introducing dynamically changing electricity prices, based on real electricity market prices. However, less is currently known about consumer preferences and response to constantly changing electricity prices. Given that different people can have a diverse set of values and preferences, it is assumed that their reaction to changing electricity prices is quite diverse. Therefore, we have asked ourselves how household characteristics and dynamic electricity prices influence household electricity consumption in dynamic price settings. We believe that clarifying the relationship of these two components with electricity consumption behavior can deliver valuable insights for future energy services.

Our findings have shown that the price sensitivity of households exposed to technology-enabled dynamic price settings is higher as compared to household exposed to the usual Time-of-Use pricing applied to many households today. Moreover, this price sensitivity is significantly influenced by household attributes, such as the number of occupants, the building size, age and type, and the availability of roof insulation. This finding suggests that household attributes can be a useful tool to determine how likely households are to properly respond to electricity prices in general. In addition, we have found that the significance of these household attributes is varying during different hours of the day. Hence, future efforts to determine household price sensitivity have to evaluate price sensitivity on an hourly basis, based on different sets of variables. Lastly, we have found that, other than the availability of roof insulation, all household attributes have shown a weaker relationship with a household’s willingness to use electricity compared to our control group. This gives reason to believe that households exposed to dynamic prices have generally become more price sensitive and follow fever habitual patterns. A closer look at the influence of the dynamic prices was therefore essential.

Our findings have shown that our treatment group had a different relative load profile than our control group, which suggests that the dynamic prices have influenced electricity usage patterns in a way that the TOU prices have not. The analysis of this study has confirmed that dynamic prices have a significantly different, in our case less positive, relationship with electricity usage behavior compared to TOU prices. Although the relationship of electricity prices with relative usage is positive in our analysis, we believe that the excitement of the participants of our study is responsible for this. Hence, the inclusion of a data set with a longer time-period would probably show negative correlations. Therefore, the influence dynamic prices should normally be stronger, or more negative, than the influence of TOU prices. Moreover, our results are able to prove that the capability of households to change their consumption behavior based on changes in electricity prices only exists between 8AM to 5PM. Lastly, we have found that the time window in which household are capable to change their behavior based on electricity prices is overlapping with the time-window in which the relative usage between treatment and control group is deviating from each other. We have interpreted this observation as instances of dynamic pricing encouraged load shifting behavior.


59

Based on the previous two analyses, we have come to the conclusion that efforts to properly segment households into groups that reflect a household’s capability to change electricity usage patterns can be based on objective variables, such as the household attributes used in our study. Our findings show that this approach was successful in identifying two specific groups of households that have changed their electricity usage behavior, and are engaging in load shifting. Although we have proposed to follow a segmentation strategy that takes a set of different variables into account for every hour of the day, we have followed a static segmentation for practical reasons.

5.2 Managerial Implications

Utility providers are starting to introduce dynamic prices to their energy contracts with residential households, with the aim to accurately match fluctuating energy supply with increasingly uncertain demand patterns. Due to the introduction of smart metering devices, this has become realizable. However, in order to properly take advantage of this improved way of communicating, it has to be understood how different types of households react to prices in a dynamic pricing setting.

First, utility providers need to understand how much a household will respond to price variations. Hence, it is crucial to grasp what factors influence the price sensitivity of households. As the consumer choice theory correctly explains, households will make a purchasing decision based on budget constraints (electricity price), time constraints (their daily routine) and preferences. In order to create a complete picture of the price sensitivity of households, it is therefore crucial to examine the influence of household attributes, as these will reflect many of the preference of households and the development of the price sensitivity throughout the day. Our results show that utility providers have to examine price sensitivity indeed for every hour of the day and base their evaluation on a constantly changing set of variables. Hence, future customer targeting initiatives should segment households into dynamic groups based on a set of variables that proves to be significant on that exact hour of the day. In order to encourage a better interaction after 5PM, non-monetary incentives should be developed. In specific, utility providers should examine the influence of different types of informational feedback to specifically target household groups. This can be done in the form of gamification, consumption comparisons and other concepts suggested in previous studies (Hermsen et al., 2016; Darby, 2006; Fischer, 2008). Based on our findings, we suggest that non-monetary incentives such as informational feedback should also be examined on an hourly level, to properly evaluate the performance throughout the day. Lastly, our study has shown that utility providers would be well advised to find out as much information about their consumers as possible, in order to improve and extend our current line of variables. For example, socio-demographic information, such as education, income level or appliance ownership information have been proven to significantly influence price responsiveness (Hayn et al., 2014).

5.3 Academic Implications

Our study has attempted to stress that electricity consumers are neither rational decision makers, nor basing their consumption behavior on entirely irrational decisions. We have successfully proven that a mix of situational circumstances, budget constraints, time constraints and preferences play an important role while households make consumption decisions. Moreover, we have stressed the fact that irrational behavior comes into play when examining energy usage behavior in the form of habits.


60

Advanced metering techniques have the potential to encourage rational decision making based on prices and the change of behavioral patterns based on disruptive information, which can contribute to success of future energy services.

We have successfully proven that household attributes are significant determinants of household price sensitivity, as they are able to reflect a household daily electricity usage patterns. Furthermore, we have proven that dynamic prices significantly influence the relative usage profile of households and thus encourage households to break behavioral patterns in favor of cheaper electricity.

Moreover, another significant contribution of our study is the notion that future studies examining electricity consumption and behavior, have to fine-grain their analysis at least to an hourly level. Almost all scientific analyses have disregarded this fact and aggregated their results to a daily or peak/off-peak level, which does not do justice to the complex constraints residential household are facing throughout the day.

Another significant contribution is that we have successfully proven that dynamic prices influence electricity usage behavior in a significantly different way compared to TOU prices, and actually encourage load shifting behavior, which could not be proven by various previous studies investigating TOU prices.


61

6. References European Parliament, Council of the European Union. (2009). Directive 2009/28/EC of the european

parliament and of the council of 23 April 2009 on the promotion of the use of energy from renewable sources and amending and subsequently repealing Directives 2001/77/EC and 2003/30/EC. Official Journal of the European Union European Union.

Abdi, H., & Williams, L. (2010). Principle Component Analysis. WIREs Comp Stat, 2(4), 433-459. Accenture. (2015). The New Energy Consumer Balancing Strategic and Operational Imperatives.

Accenture. Albadi, M., & El-Saadany, E. (2007). Demand Response in Electricity Markets: An Overview. Power

Engineering Society General Meeting, 1-5; 24-28. Albadi, M., & El-Saadany, E. (2008). A summary of demand response in electricity markets. Electric

Power Systems Research, 78(1), 1989-1996. Alberini, A., Gans, W., & Velez-Lopez, D. (2011). Residential Consumption of gas and electricity in

the U.S: The role of prices and income. Energy Economics, 33(1), 870-881. Allcott, H. (2011). Rethinking real-time electricity pricing. Resource and Energy Economics, 33(1),

820-842. Anderberg, M. (1973). Cluster Analysis for Applications (1 ed.). New York: Academic Press. APX Group. (2016). APX power spot exchange. Retrieved July 08, 2016, from www.apxgroup.com/apx-

power-nl/ Baltagi, B. (2013). Econometric Analysis of Panel Data (5 ed.). United States: John Wiley & Sons. Bartush, C., Wallin, F., Odlare, M., Vassileva, I., & Wester, L. (2011). Introducing a demand-based

electricity distribution tariff in the residential sector: Demand response and customer perception. Energy Policy, 39, 5008-5025.

Bhatia, S. (2004). Adaptive K-Means Clustering. University of Missouri - St.Louis. American Association for Artificial Intelligence.

Bhattacharya, K., Bollen, M., & Daalder, J. (2001). Operation of Restructured Power System. London: Kluwer Academics Publishers.

Bloustein, E. (2005). Assessment of Customer Response to Real Time Pricing (1 ed.). New Jersey: Rutgers - The State University of New Jersey.

Borenstein, S., Jaske, M., & Rosenfeld, A. (2002). Dynamic Pricing, Advanced Metering, and Demand Response in Electricity Markets. Los Angeles, CA: The Energy Foundation.

Brandon, G., & Lewis, A. (1999). Reducing Household Energy Consumption: A Qualitative and Quantitative Field Study. Journal of Environmental Psychology, 19(1), 75-85.

Capgemini Consulting. (2015). European Energy Markets Observation. Paris: Capgemini. Contreras, J., Asensio, M., Quevedo, P., Munoz-Delgado, G., & Montoya-Bueno, S. (2016). Joint RES

and Distribution Network Expansion Planning under a Demand Response Framework (1 ed.). London: Elsevier Academic Press.

Darby, S. (2006). The Effectiveness of Feedback on Energy Consumption: A Review for DEFRA of the Literature on Metering, Billing and Direct Displays. University of Oxford. Oxford: Environmental Change Institute, University of Oxford.

Darby, S. (2012). Literature Review for the Energy Demand Research Project. Oxford: Environmental Change Institute - University of Oxford.


62

Darby, S. (2012). Literature Review for the Energy Demand Research Project. Oxford: Environmental Change Institute - University of Oxford.

Deaton, A., & Muellbauer, J. (1960). An almost ideal demand system. The American Economic Review, 70(3), 312-326.

DeMeo, E., Jordan, G., & Kalich, C. (2007). Accommodating wind’s natural behavior. IEEE Power and Energy Magazine, 5(6), 59-67.

Di Cosmo, V., Lyons, S., & Nolan, A. (2014). Estimating the impact of time-of-use pricing on Irish electricity demand. The Energy Journal, 35(2), 1-26.

Ehrhardt-Martinez, K., Laitner, J. A., & Donnelly, K. A. (2011). Chapter 10—Beyond the Meter: Enabling Better Home Energy Management. In S. Fereidoon Perry, Energy, Sustainability and the Environment (pp. 273-303). Boston: Butterworth- Heinemann.

Elliot, A. J., & Covington, M. V. (2001). Approach and avoidance motivation. Educational Psychology Review, 13(2), 73-92.

European Union. (2016). Cop21.gouv.fr. Retrieved January 22, 2016, from http://www.cop21.gouv.fr/en/learn/what-is-cop21/

Eurostat. (2015). europe.eu. Retrieved Januar 22, 2016, from http://ec.europa.eu/eurostat/statistics-explained/index.php/Consumption_of_energy

Farhangi, H. (2010). The path of the smart grid. Power and Energy Magazine, IEEE, 8(1), 18-28. Farhar, B. (1999). Willingness to Pay for Electricity from Renewable Resources: A Review of Utility

Market Research. Office of Scientific and Technical Information (OSTI). Faruqui, A., & Malko, J. (1983). The Residential Demand for Electricity By Time-Of-Use: A Survey of

Twelve Experiments With Peak Load Pricing. Energy, 8(10), 781-795. Faruqui, A., & Palmer, J. (2012). The Discovery of Price Responsiveness - A Survey of Experiments

involving Dynamic Pricing of Electricity. SSRN Electronic Journal, 4(1), 1-14. Faruqui, A., Sergici, S., & Sharif, A. (2010). evidence, The impact of informational feedback on nergy

consumption - A survey of the experimental. Energy, 35(4), 1598-1608. Faruqui, A., Sergici, S., & Sharif, A. (2010). The impact of informational feedback on nergy

consumption - A survey of the experimental evidence. Energy, 35(4), 1598-1608. Filippini, M. (2011). Short- and long-run time-of-use price elasticities in Swiss residential electricity

demand. Energy Policy, 39(1), 5811-5817. Fischer, C. (2008). Feedback on Household Electricity Con- sumption: A Tool for Saving Energy?

Energy Efficiency, 1(1), 79-104. Fisher, C. (2007). Discussion Paper 8: Influencing Electricity Consumption Via Consumer Feedback:

A Review of Ex- perience. Berlin: ECEEE 2007 Summer Study. Gardner, B. (2014). A review and analysis of the use of ‘Habit’ in understanding, predicting and

influencing health-related behavior. Health Psychology Review, 8(1), 1-19. Ghose, A., Goldfarb, A., & Han, S. (2013). How is the Mobile Internet Different? Search Costs and

Local Activities. Information Systems Research, 1-19. Greene, W. (2008). Econometric Analysis (5 ed.). New Jersey: Pearson Education. Grijalva, S., & Tariq, M. (2011). Prosumer-based smart grid architecture enables a flat, sustainable

electricity industry. Innovative Smart Grid Technologies (ISGT), 2011 IEEE PES, 1-6; 17-19. Hayn, M., Bertsch, V., & Fichtner, W. (2014). Electricity load profiles in Europe: The importance of

household segmentation. Energy Research & Social Science, 3(1), 30-45.


63

Hermsen, S., Frost, J., Renes, R., & Kerkhof, P. (2016). Using feedback through digital technology to disrupt and change habitual behavior: A critical review of current literature. Computers in Human Behavior, 57, 61-74.

Higgins, E. T. (1997). Beyond pleasure and pain. American psychologist, 52(12), 1280. Hino, H., Shen, H., Murata, N., Wakao, S., & Hayashi, Y. (2013). A versatile clustering method for

electricity consumption pattern analysis in households. IEEE Transactions on Smart Grid, 4(2), 1048-1057.

Hutton, R. B., Mauser, G. A., Filiatrault, P., & Ah- tola, O. T. (1986). Effects of Cost-Related Feedback on Consumer Knowledge and Consumption Behavior: A Field Experimental Approach. Journal of Consumer Research, 13(3), 327-336.

Ingham, A., Levinger, G., Graves, J., & Peckham, V. (1974). The Ringelmann effect: Studies of group size and group performance. Journal of Experimental Social Psychology, 10, 371-384.

International Business Publications. (2016). Netherlands Investment and Business Guide Volume 1 Strategic and Practical Information. IBP, Inc.

Jacobsson, S., & Bergek, A. (2004). Transforming the energy sector: the evolution of technological systems in renewable energy technology. Industrial and Corporate Change, 13(5), 815-49.

Jager, W. (2003). Breaking ‘bad habits’: a dynamical perspective on habit formation and change. In H. L., W. Jager, & L. Steg, Human decision-making and environmental Percep- tioneUnderstanding and assisting human decision-making in real life settings. Groningen: University of Groningen.

Jain, R. K., Taylor, J. E., & Peschiera, G. (2012). Assessing eco-feedback interface usage and design to drive energy efficiency in buildings. Energy and Buildings, 48, 8-17.

Kavousian, A., Rajagopal, R., & Fischer, M. (2013). Determinants of residential electricity consumption: using smart meter data to examine the effect of climate, building characteristics, appliance stock, and occupants behavior. Energy, 55, 184-194.

Kwac, J., Flora, J., & Rajagopal, R. (2014). Household Energy Consumption Segmentation Using Hourly Data. IEEE Transactions on Smart Grid, 5(1), 420-430.

Lampropoulos, Vanalme, & Kling. (2010). A methodology for modeling the behavior of electricity prosumers within the smart grid. Innovative Smart Grid Technologies Conference Europe (ISGT Europe), 1-8.

Matsukawa, I. (2001). Household Response to Optional Peak-Load Pricing of Electricity. Journal of Regulatory Economics, 20(3), 249-267.

Matthies, E. (2005). Wie können PsychologInnen ihr Wissen besser an die PraktikerIn bringen? Vorschlag eines neuen, integrativen Einflussschemas umweltgerechten Alltagshandelns. Umweltpsychologie, 9(1), 62-81.

McCalley, L. T., & Midden, C. J. (2002). Energy Conservation through Product-Integrated Feedback: The Roles of Goal-Setting and Social Orientation. Journal of Economic Psychology, 23(5), 589-603.

McCarney, R., Warner, J., Iliffe, S., van Haselen, R., Griffin, M., & Fisher, P. (2007). The Hawthorne Effect: a randomised, controlled trial. BMC Medical Research Methodology, 7(30), 1-8.

McLoughlin, F., Duffy, A., & Conlon, M. (2012). Characterising Domestic electricity consumption patterns by dwelling and occupant socio-economic variables: an Irish case study. Energy Build, 48(2), 240-248.


64

Midden, C. J., Meter, J. E., Weenig, M. H., & Zieverink, H. J. (1983). Using Feedback, Reinforcement and Information to Reduce Energy Consumption in Households: A Field-Experiment. Journal of Economic Psychology, 3(1), 65-86.

Murthy Balijepalli, V. S., Vedanta Pradhan, S. A., & Shereef, R. M. (2011). Review of Demand Response under Smart Grid Paradigm. India: 2011 IEEE PES International Conference on Innovative Smart Grid Technologies-India.

Nababan, T. (2015). Analysis of Household Characteristics Affecting the Demand of PLN’s Electricity. An Observation on Small Households in City of Medan, Indonesia. Academic Journal of Economic Studies, 1(2), 79–92.

National Research Council of the National Academies. (2006). Surface Temperature Reconstructions for the last 2000 years. Washington D.C.: The National Academies Press.

Northcraft, G. B., Schmidt, A. M., & Ashford, S. J. (2011). Feedback and the rationing of time and effort among competing tasks. Journal of Applied Psychology, 96(5), 1076.

Palensky, P., & Dietrich, D. (2011, August). Demand Side Management: Demand Response, Intelligent Energy Systems, and Smart Loads. IEEE Transactions on industrial informatics, 7(3), 381-388.

Peeples, M. (2011). R Script for K-Means Cluster Analysis. Retrieved July 10, 2016, from http://www.mattpeeples.net/kmeans.html

Salvatore. (2008). Consumer Preferences and Choice. In Microeconomics: Theory and Applications (pp. 57-86). Oxford: Oxford University Press.

Santer, Taylor, Wigley, Johns, Jones, Karoly, . . . Stouffer. (1996). A search for human influences on the thermal structure of the atmosphere. Nature(382), 39-46.

Schleich, J., & Klobasa, M. (2013). How much shift in demand? Findings from a field experiment in Germany. ECEEE 2013 SUMMER STUDY – RETHINK, RENEW, RESTART, 1(1), 1919-1925.

Seligman, C., Becker, L. J., & Darley, J. M. (1981). Encouraging Residential Energy Conservation through Feedback. Advances in Environmental Psychology, 3, 93-113.

Sexton, R., Johnson, N., & Konakayama, A. (1987). Consumer response to continuous-display electricity-use monitors in a time-of-use pricing experiment. Journal of Consumer Research, 14(1), 55-62.

Spiegel. (2014). Spiegel.de. Retrieved January 22, 2016, from http://www.spiegel.de/wirtschaft/unternehmen/rwe-meldet-2-8-milliarden-euro-verlust-fuer-2013-a-956749.html

Statistik Austria. (2011). Modellierung des Stromverbrauchs in den privaten Haushalten Österreichs nach unterschiedlichen Verwendungszwecken. Vienna: Statistik Austria.

Sundramoorthy, V., Cooper, G., Linge, N., & Qi, L. (2011). Domesticating Energy-Monitoring Systems: Challenges and Design Concerns. EEE on Pervasive Computing, 10(1), 20-27.

The Economist. (2013). Economist.com. Retrieved January 22, 2016, from http://www.economist.com/news/briefing/21587782-europes-electricity-providers-face-existential-threat-how-lose-half-trillion-euros

Torriti, J., Hassen, M., & Leach, M. (2010). Demand response experience in Europe: Policies, prgrammes and implementation. Energy, 35, 1575-1583.

U.S. Energy Information Administration. (2016). World energy demand and economic outlook. International Energy Outlook 2016.

US Department of Energy. (2006). Benefits of Demand Response in Electricity Markets and Recommendations for Achieving them. Washington D.C.: Report to the United States Congress.


65

Van Werven, M., & Scheepers, M. (2005). The Changing Role of Energy Suppliers and Distribution System Operators in the Deployment of Distributed Generation in Liberalised Electricity Markets. Amsterdam: Energy Research Center of the Netherlands.

Vine, D., Buys, L., & P., M. (2013). The Effectiveness of Energy Feedback for Conservation and Peak Demand: A Literature Review. Open Journal of Energy Efficiency, 2, 7-15.

Wallenborn, G., Orsini, M., & Vanhaverbeke, J. (2011). House- hold Appropriation of Electricity Monitors. International Journal of Consumer Studies, 35(2), 146-152.

Westerman, G., Bonnet, D., & A., M. (2014). Leading Digital: Turning Technology into Business Transformation. New York: Harvard Business Review Press.

Westerman, G., Bonnet, D., & McAfee, A. (2014). Leading Digital: Turning Technology into Business Transformation. New York: Harvard Business Review Press.

Wilhite, H., & Ling, R. (1995). Measured Energy Savings from a More Informative Energy Bill. Energy and Buildings, 22(2), 145-155.

Winett, R., Kagel, J., & Battalio, R. (1978). Effects of Monetary Rebates, Feedback, and Infor- mation on Residential Electricity Conservation. Journal of Applied Psychology, 63(1), 73-80.

Wood, G., & Newborough, M. (2003). Dynamic Energy-Con- sumption Indicators for Domestic Appliances: Environ- ment, Behaviour and Design. Energy and Buildings, 35(8), 821-841.

Wood, W., & Neal, D. T. (2009). The habitual consumer. Journal of Consumer Psychology, 19(4), 579-592.

Yohannis, Y., Mondol, J., Wright, A., & Norton, B. (2008). Real-Life energy use in the UK: how occupancy and dwelling characteristics affect domestic electricity use. Energy Build, 40(6), 1053-1059.


66

7. Appendix

Summary Statistics – Treatment Group


relative consumption 47275 0,042 0,0343 -0,00412 1

price1 720 0,1676 0,1682 0,1515 0,1765

price2 720 0,1664 0,167 0,1504 0,1725

price3 720 0,1655 0,1657 0,1493 0,1711

price4 720 0,1652 0,1662 0,1478 0,1696

price5 720 0,1664 0,1679 0,149 0,1711

price6 720 0,172 0,1732 0,1485 0,1825

price7 720 0,1773 0,1798 0,1487 0,1876

price8 720 0,1777 0,1803 0,1463 0,1896

price9 720 0,1781 0,179 0,1627 0,1913

price10 720 0,1762 0,1763 0,1631 0,1891

price11 720 0,1749 0,1759 0,1631 0,1851

price12 720 0,1734 0,1733 0,1611 0,1844

price13 720 0,1718 0,1718 0,1585 0,1827

price14 720 0,1702 0,1709 0,1536 0,1804

price15 720 0,1697 0,1698 0,1596 0,1789

price16 720 0,1703 0,1707 0,1537 0,1789

price17 720 0,1744 0,1751 0,1622 0,1827

price18 720 0,1812 0,1807 0,162 0,1999

price19 720 0,185 0,1862 0,1653 0,1994

price20 720 0,1797 0,1792 0,1666 0,1915

price21 720 0,1747 0,1744 0,1647 0,1836

price22 720 0,1736 0,1741 0,1638 0,1795

price23 720 0,1708 0,1714 0,1625 0,1807

price24 720 0,1696 0,1692 0,1503 0,1925

Solar Influx 720 37,74 1 0 259 Table 13. Summary Statistics of Prices and Relative Usage

Summary Statistics - Control Group


relative consumption 103515 0,04005 0,03279 -0,004 1

price1 1 0,1744 0,1744 0,1744 0,1744

price2 1 0,1868 0,1868 0,1868 0,1868

SolarInflux 720 37,74 1 0 259 Table 14. Summary Statistics of Prices and Relative Usage


67

Table 15. 24h Panel Data Regression Results Equation 1 – Treatment Group

Panel Data Regression Results – Treatment Group

Dependent variable: Willingness to use electrical energy (usage/price)

Y (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) (21) (22) (23) (24)

Persons 0.33*** 0.19*** 0.16*** 0.14*** 0.16*** 0.28*** 0.56*** 0.27*** 0.16** 0.21*** 0.39*** 0.57*** 0.63*** 0.79*** 0.70*** 0.69*** 0.76*** 0.81*** 0.81*** 0.82*** 0.71*** 0.66*** 0.61*** 0.39***

(0.05) (0.04) (0.04) (0.03) (0.03) (0.04) (0.05) (0.05) (0.07) (0.08) (0.09) (0.10) (0.10) (0.09) (0.08) (0.07) (0.07) (0.06) (0.07) (0.06) (0.06) (0.06) (0.06) (0.05)

Building Age -0.15*** -0.10** -0.09** -0.15*** -0.15*** -0.07* 0.14*** -0.12** -0.48*** -0.59*** -0.76*** -0.92*** -0.94*** -0.86*** -0.69*** -0.54*** -0.27*** -0.05 0.10 0.18*** 0.15*** 0.08 0.01 -0.03

(0.05) (0.04) (0.04) (0.03) (0.03) (0.04) (0.05) (0.05) (0.07) (0.08) (0.09) (0.10) (0.10) (0.09) (0.08) (0.07) (0.07) (0.07) (0.07) (0.06) (0.06) (0.06) (0.06) (0.05)

Building Size 0.01*** 0.01*** 0.02*** 0.003*** 0.003*** 0.003*** 0.002*** 0.001 0.001 -0.001 -0.001 -0.003* -0.003* 0.0001 0.002 0.003** 0.004*** 0.01*** 0.01*** 0.01*** 0.01*** 0.01*** 0.01*** 0.01***

(0.001) (0.001) (0.001) (0.001) (0.005) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002) (0.002) (0.002) (0.002) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001)

Building Type -0.06 -0.06 -0.14** -0.20*** -0.16*** -0.15*** -0.04 -0.005 0.02 0.01 0.04 -0.05 -0.02 -0.19** -0.25*** -0.31*** -0.07 -0.03 -0.07 -0.08 -0.09 -0.02 0.09 -0.002

(0.05) (0.04) (0.04) (0.03) (0.03) (0.04) (0.05) (0.05) (0.07) (0.08) (0.09) (0.10) (0.10) (0.09) (0.08) (0.07) (0.07) (0.07) (0.07) (0.07) (0.06) (0.06) (0.06) (0.05)

Roof Insulation -0.18 -0.13 0.01 0.16 0.19** 0.24* -0.09 -0.38** -0.53** -0.79*** -1.23*** -1.20*** -1.47*** -0.99*** -0.95*** -0.38* 0.16 0.40* 0.42* 0.42** 0.55*** 0.27 -0.31* -0.25

(0.16) (0.14) (0.12) (0.10) (0.09) (0.13) (0.17) (0.17) (0.23) (0.26) (0.30) (0.32) (0.32) (0.30) (0.25) (0.22) (0.22) (0.21) (0.22) (0.21) (0.20) (0.20) (0.19) (0.16)

Solar Influx -0.1*** -0.01*** -0.02*** -0.02*** -0.02*** -0.02*** -0.02*** -0.02*** -0.02*** -0.02*** -0.05** 0.36

(0.003) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) (0.003) (0.005) (0.02) (0.52)

Constant 0.44 0.53 1.51*** 1.39*** 0.75** 0.68 -0.001 3.10*** 5.38*** 7.39*** 7.51*** 7.89*** 8.65*** 6.49*** 6.06*** 3.52*** 2.08** 0.32 -0.15 -0.72 -0.02 0.51 -0.12 0.18

(0.51) (0.46) (0.39) (0.34) (0.31) (0.42) (0.58) (0.60) (0.81) (0.94) (1.04) (1.09) (1.13) (1.08) (0.92) (0.79) (0.81) (0.84) (0.73) (0.69) (0.65) (0.65) (0.60) (0.56)

Observations 1,990 1,923 1,987 1,991 1,991 1,991 1,990 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,991 1,999 2,100

R2 0.12 0.07 0.02 0.05 0.06 0.04 0.07 0.03 0.05 0.07 0.08 0.09 0.09 0.10 0.10 0.09 0.08 0.10 0.10 0.13 0.11 0.10 0.15 0.13

Adjusted R2 0.12 0.07 0.02 0.05 0.06 0.04 0.07 0.03 0.05 0.06 0.08 0.09 0.09 0.10 0.10 0.09 0.08 0.10 0.10 0.13 0.11 0.10 0.15 0.13

Hausman

(p>chi2) 0.16 0.15 0.22 0.08 0.09 0.06 0.09 0.10 0.10 0.11 0.12 0.11 0.11 0.15 0.14 0.15 0.11 0.11 0.12 0.14 0.06 0.06 0.07 0.08

F Statistic 44.94***(df = 6; 1983)

23.63***(df = 6; 1916)

7.86***(df = 6; 1980)

17.64***(df = 6; 1984)

21.00***(df = 6; 1984)

15.22***(df = 6; 1984)

25.33***(df = 6; 1983)

7.96***(df = 7; 1983)

14.13***(df = 7; 1983)

19.76***(df = 7; 1983)

25.28***(df = 7; 1983)

27.88***(df = 7; 1983)

28.67***(df = 7; 1983)

29.98***(df = 7; 1983)

31.26***(df = 7; 1983)

29.59***(df = 7; 1983)

25.23***(df = 7; 1983)

32.48***(df = 7; 1983)

30.67***(df = 7; 1983)

49.79***(df = 6; 1984)

42.30***(df = 6; 1984)

35.68***(df = 6; 1984)

60.02***(df = 6; 1992)

51.44***(df = 6; 2093)

Note: *p<0.1; **p<0.05; ***p<0.01


68

Table 16. 24h Panel Data Regression Results Equation 1 – Control Group

Panel Data Regression Results – Control Group

Dependent variable: Willingness to use electrical energy (usage/price)

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) (21) (22) (23) (24) Persons -0.15 -0.08 -0.14 -0.19* -0.15 -0.10 0.14** 0.13 0.21** 0.23* 0.41*** 0.37*** 0.39*** 0.26** 0.20 0.41*** 0.54*** 0.52*** 0.54*** 0.59*** 0.27** 0.15 -0.09 -0.19

(0.11) (0.09) (0.10) (0.11) (0.10) (0.07) (0.07) (0.08) (0.10) (0.12) (0.11) (0.11) (0.11) (0.12) (0.12) (0.11) (0.12) (0.11) (0.15) (0.15) (0.11) (0.09) (0.12) (0.13)

Building Age 0.13 0.01 0.01 0.06 0.06 0.05 -0.09 -0.08 -0.10 -0.06 -0.17 -0.10 -0.11 0.04 0.02 -0.02 -0.22* -0.36*** -0.15 -0.06 -0.15 -0.22** -0.03 0.14

(0.10) (0.08) (0.09) (0.10) (0.09) (0.07) (0.06) (0.08) (0.10) (0.11) (0.11) (0.10) (0.10) (0.11) (0.11) (0.10) (0.11) (0.10) (0.14) (0.14) (0.11) (0.09) (0.11) (0.12)

Building Size 0.02*** 0.02*** 0.02*** 0.02*** 0.02*** 0.02*** 0.02*** 0.02*** 0.02*** 0.03*** 0.02*** 0.02*** 0.02*** 0.02*** 0.02*** 0.02*** 0.02*** 0.02*** 0.01*** 0.02*** 0.02*** 0.02*** 0.02*** 0.02***

(0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) (0.002) (0.002) (0.001) (0.001) (0.001) (0.001)

Building Type 0.18** 0.07 0.08 0.11 0.16** 0.12** 0.14** 0.23*** 0.16* 0.17* 0.19** 0.15* 0.19** 0.27*** 0.24** 0.23*** 0.17* 0.28*** 0.44*** 0.38*** 0.33*** 0.18** 0.22** 0.22**

(0.09) (0.07) (0.08) (0.09) (0.08) (0.06) (0.05) (0.07) (0.09) (0.09) (0.09) (0.09) (0.09) (0.10) (0.10) (0.09) (0.10) (0.09) (0.12) (0.12) (0.09) (0.08) (0.10) (0.10)

Roof Insulation -0.16 -0.18 -0.16 -0.14 -0.35 -0.60*** -0.45** -0.75*** -0.98*** -1.02*** -1.09*** -0.96*** -0.99*** -1.22*** -1.15*** -1.32*** -0.29 0.28 0.12 0.22 0.06 0.19 -0.08 -0.12

(0.29) (0.24) (0.26) (0.30) (0.26) (0.19) (0.18) (0.23) (0.28) (0.32) (0.31) (0.29) (0.30) (0.33) (0.33) (0.30) (0.32) (0.30) (0.40) (0.39) (0.31) (0.25) (0.32) (0.34)

Solar Influx -0.001** -0.001** -0.001* -0.002** -0.002* -0.001** -0.002** 0.0000 -0.003** -0.01** -0.03** 0.09

(0.004) (0.002) (0.002) (0.002) (0.001) (0.002) (0.002) (0.003) (0.004) (0.01) (0.02) (0.34)

Constant -1.16 -1.33* -1.71** -2.08** -1.40** -0.96* -0.70 -0.46 1.04 0.58 0.74 1.11 1.15 1.93 1.91 1.01 0.13 -1.06 -2.82*** -2.55** 0.66 0.31 -0.43 -0.85

(0.87) (0.76) (0.79) (0.90) (0.71) (0.55) (0.64) (0.75) (0.87) (0.90) (0.93) (0.86) (0.95) (1.19) (1.24) (1.09) (1.00) (1.08) (1.04) (1.10) (0.97) (0.84) (1.06) (1.03)

Observations 1,020 986 1,020 1,020 1,020 1,020 1,020 1,020 1,020 1,020 1,020 1,020 1,020 1,020 1,020 1,020 1,020 1,020 1,020 1,020 1,020 1,020 1,020 1,020

R2 0.18 0.26 0.24 0.19 0.20 0.37 0.46 0.41 0.32 0.31 0.31 0.30 0.33 0.28 0.26 0.27 0.25 0.22 0.14 0.16 0.20 0.22 0.16 0.14

Adjusted R2 0.18 0.25 0.24 0.19 0.20 0.37 0.46 0.41 0.32 0.31 0.31 0.30 0.33 0.28 0.26 0.26 0.25 0.22 0.13 0.16 0.19 0.21 0.16 0.14

Hausman (p>chi2) 0.1 0.1 0.2 0.2 0.1 0.1 0.2 0.2 0.2 0.1 0.1 0.1 0.2 0.2 0.2 0.1 0.1 0.2 0.1 0.1 0.1 0.1 0.1 0.1

F Statistic 38.18***(df = 6; 1013)

56.04***(df = 6; 979)

53.64***(df = 6; 1013)

39.62***(df = 6; 1013)

42.25***(df = 6; 1013)

99.79***(df = 6; 1013)

143.90***(df = 6; 1013)

99.79***(df = 7; 1012)

69.06***(df = 7; 1012)

64.72***(df = 7; 1012)

65.55***(df = 7; 1012)

61.76***(df = 7; 1012)

70.83***(df = 7; 1012)

55.69***(df = 7; 1012)

51.40***(df = 7; 1012)

52.34***(df = 7; 1012)

49.13***(df = 7; 1012)

41.90***(df = 7; 1012)

22.66***(df = 7; 1012)

31.83***(df = 6; 1013)

41.12***(df = 6; 1013)

46.45***(df = 6; 1013)

32.78***(df = 6; 1013)

28.13***(df = 6; 1013)

Note: *p<0.1; **p<0.05; ***p<0.01


69

Table 17. PCA Analysis of All Available Household Variables

PC1 PC2 PC3 PC4 PC5 PC6 PC7 PC8 PC9

Standard deviation 1,74 1,22 1,10 1,00 0,96 0,84 0,63 0,51 0,00 Proportion of Variance 0,33 0,17 0,13 0,11 0,10 0,08 0,04 0,03 0,00 Cumulative Proportion 0,33 0,50 0,64 0,75 0,85 0,93 0,97 1,00 1,00 Building Age 0,29 -0,19 -0,43 -0,17 -0,47 -0,53 0,27 -0,04 0,29 Building Size -0,56 0,15 0,05 0,09 -0,07 0,01 -0,12 -0,06 0,79 Building Type 0,28 -0,53 0,11 0,25 0,43 0,23 0,39 -0,21 0,38 Solar Panels -0,03 -0,24 0,37 -0,80 0,25 -0,24 -0,18 -0,09 0,10 Terrain Type 0,45 -0,04 -0,04 0,29 0,16 -0,22 -0,77 -0,04 0,20 Persons 0,42 0,39 0,04 -0,22 0,15 0,19 -0,22 -0,68 0,24 Solar Heating -0,04 -0,05 -0,73 -0,35 0,17 0,49 -0,21 -0,14 0,05 Ventilation Type 0,27 0,15 0,09 -0,06 0,08 -0,01 0,12 -0,67 0,14 Roof Insulation 0,25 -0,16 0,33 -0,11 -0,67 0,54 -0,18 -0,10 0,11


70

7.1 Qurrent Energie Dashboard Screenshots

Documents

Identifying customer patterns for energy services in a ...utility providers (Capgemini Consulting, 2015). First, customers are looking for more flexible and responsive suppliers of