The attractiveness of performance-based insurance

The attractiveness of performance-based insurance.

Demi Vollemans 10313680

August 14, 2018

Abstract

With the development of new innovative technologies, insurers are now able to offer insurances at personalised prices. The price is determined by behaviour and performance of the user. In this research, it is examined by the use of literature and experiments if consumers would prefer performance-based insurances compared to traditional insurance. This research shows that this new form of insurance has a number of characteristics that can evoke resistance by consumers to buy this product, the main factor is, giving up information about action and behaviour. However, this insurance has also its potential for certain target groups when consumers realise they can achieve an attractive monetary benefit.

Statement of Originality

This document is written by Student Demi Vollemans who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for supervision of completion of the work, not for the contents.

List of contents

1. Introduction……… 5

1.1 New innovative developments……… 5

1.2 Research topic……… 6

1.3 Contribution to existing literature……….... 7

2. Literature review………... 9

2.1The market………...9

The Consumer………. …….9

The Insurers………11

2.2 Decision-making involving privacy trade-offs……… 12

2.2.1 Uncertainty from privacy trade-offs………12

Incomplete information………...13 Unknown preferences……… 14 2.2.2 Privacy paradox……….14 2.2.3 Context dependence………...16 2.3 Insurance Behaviour………..17 2.3.1 Uncertainty……….17

2.3.2 Behavioural economics and insurance behaviour………...17

2.4 Findings applied to Performance-based insurance………...,..18

3. Experimental Design and procedure………...20

3.1 The insurance experiment……….20

3.1.1 Procedure………..… 20

3.1.2 The matrix task……… .… 21

3.1.3 The game………22

3.1.4 Policies……….. 22

3.1.5 Treatment……….. 23

3.2 The Maximum Differences experiment……….24

3.2.1 Experimental design………....24

3.3 Participants………...… 25

4. Expected results………. 25

4.1 Rational Agents and Expected payoffs.……….………...25

4.2 Prospect Theory………26

5. Results………29

5.1 summary statistics……… 29

5.2 Treatment effect ……….. 31

5.3 Self-selection……….32

5.3 Multinomial Logistic Regression………....33

5.4 Maxdiff analysis………35

6. Discussion and Conclusion……….39

References………... 43

1. Introduction.

Pay How You Drive (PHYD) is a recent auto insurance innovation that is based on driving behaviour of individual car drivers. PHYD records information about how a vehicle is used by devices such as black boxes, chips or mobile phone applications, that register and keep track of driving behaviour. These devices are known as telematics devices and use Machine-to-Machine communications (M2M) or Internet of Things (IoT) technologies. They are networking, intelligent, remote devices that gather and exchange information automatically without human interaction (Soleymanian, Weinberg, & Zhu, 2017). Telematics devices can monitor and report a considerable amount of data directly while the vehicle is used. The basic idea of PHYD is to collect driving data and driving performances of consumers enabling insurers to offer users a personalised (lower) premium. The telematics device sends collected data to a database where an algorithm determines personal driving scores. This enables insurers to offer consumers individually targeted premium discounts (Soleymanian et al., 2017). Data collected by the black box consists of various information variables that are relevant for underwriters to address personal risk of an accident such as hard cornering, acceleration and location.

Adopting PHYD can benefit insurers, consumers and society as a whole. However, there are some challenges and barriers to the growth of telematics in the insurance sector. The intelligent devices in cars use location facilities to precisely measure different attributes of someone’s driving behaviour. For utilizing this insurance, consumers need to agree to share this formerly private information. This attribute of the PHYD may invoke resistance and generate an inherent tension between the protection of personal data and innovations. The ‘big brother’ factor could prevent PHYD from going mainstream and ultimately revolutionising the motor insurance industry. Another source of resistance is the uncertainty this insurance entails. If a driver obtains an excellent driving score and is perceived to be a safe driver, he gets a discount on his premium (or gets rewarded in another way), for good driving behaviour. However, there is also a chance that users will not reach the threshold of a certain driving score. Resulting in no reward or discount. In this case, that person shares private information without getting any compensation, and would have probably been better off with traditional insurance.

In some European countries, insurances based on driving behaviour is becoming a popular alternative to traditional car insurance. Insurers worldwide foresee a switch from traditional- to usage-based insurances. This switch already seems to be occurring in some countries. For example,

driving safe. Further, in August 2017, a law that recommends telematics for all insurances was adopted in Italy, in order to prevent fraud (Hallauer, 2018). The market share of digitally enabled motor insurances is predicted to be 23% in the UK and 27% in Italy by 2020 (Deloitte, 2016). On the other hand, the adoption rate of PHYD and its predicted market share in the Netherlands is lower, as a market share of 13% in 2020 was predicted by Deloitte (2016).

The Dutch insurance sector would greatly benefit if insurers can assess risk precisely while remaining competitive. The Dutch insurance market is saturated, and the sector has been struggling with deficits the last couple of years. The Dutch Bank (DNB) urged insurers to stop offering insurance prices which do not cover the expected risk. Over the last year, premium prices have been increasing and are predicted to continue to rise, especially young drivers will pay more for insurance (De Nederlandsche Bank, 2016b). Due to structural losses, there is an urge to innovate insurance business models. Offering PHYD could provide a solution to decreasing profits because it enables insurers to reduce their risk exposure and to remain competitive.

The question is; are consumers willing to pay according to performance and give up personal information on action and behaviour. In this research, the attractiveness of performance-based insurances is tested relative to traditional insurance (TI). It is also researched if higher premium costs change insurance behaviour and would make TI less attractive.

1.2 Research topic

When picking insurance, many factors influence consumers’ decision. This research focusses on three aspects that affect consumer preference for PHYD; privacy concerns, insurance behaviour under risk and relative consumer preferences for policy attributes. These factors play an important role in the decision-making process and will now be further discussed.

Firstly, PHYD insurance requires personal information exposure from consumers which can evoke aversion. Privacy is the Right to keep something for yourself. Though nowadays, consumers are more and more obligated to release a piece of information about oneself if they want to obtain or use a product. This information about the consumer, enables firms to customise or personalise their products or services, to bind consumers this way. However, making that trade-off is not always easy. Privacy exposure is a complex decision obstacle wherein attitudes, behaviours and opinions differ substantially from one individual to another (Alfssandro Acquisti & Grossklags, 2005). Various studies have shown that asking for privacy concerns or protection preferences is not representative for behaviour. When designing a lab experiment for privacy behaviour, it becomes complicated to maintain control over attitudes and assigned probabilities. Therefore, this research conducts a literature review on the decision-making process when facing

privacy trade-offs. Findings from the behavioural economics field and several field experiments on the subject apply to consumer privacy preferences and cognitive biases when making decision trade-offs. These studies will be used to examine the characteristics of PHYD, and to explain which aspects evoke resistance.

Secondly, PHYD premium costs are uncertain because they are based on performance. This is because people have to estimate whether they are going to perform good enough to get rewarded. People can control their behaviour, but hazards can occur from uncontrollable external factors. Besides, people can be uncertain about their skill or ability to perform good. However, people buy insurance to no longer worry about uncertainties and to ‘buy peace of mind’, while performance-based insurances still allow for uncertainty. In this research, preferences for such insurances are tested in a controlled environment, with the aim of finding out to which extent people find it attractive to pay a premium based on performance. When using field data, variables are often affected by other events that influence outcomes. An experiment allows for testing behaviour and preference for different insurance policies within a controlled environment. The core idea of PHYD can be researched without provoking attitudes towards the subject.

Thirdly, other (competitive) insurance policies with other advantageous attributes can prevent the consumer from eventually choosing PHYD insurance, for example, the degree of transparent conditions and terms. These different advantageous attributes are compared with a maximum difference (maxdiff) analysis to eventually be able to assign weights to attributes of insurances and to give insights whether this new pricing mechanism could be successful.

1.3 Contribution to existing literature

This research offers insights into why the Dutch insurance sector experiences low PHYD adoption rates. The findings generate a valuable insight for Dutch insurers, as DNB recently has put pressure on its current business model and emphasises the need for change to reduce deficits (De Nederlandsche Bank, 2016a). Further, findings of this research could also apply to other industries, such as the health sector, as strategies are developed to reform the current health care system to a value-based system (Porter, 2009). Health care is getting more expensive and is over-used. IoT technologies make it possible to offer health care premiums based on the health condition of the patient. For example, a smartwatch could monitor the health of the insured and those that sport regularly get a discount on health insurance. This system would be ‘fairer’ for people that make less use of health care. Literature suggests that value-based insurances could improve the current health care system by paying more attention to preventive care and improve the quality of care

demanding new forms of consumption, tying prices of a product or service to usage. The next step is to pay according to (desirable) behaviour, (e.g. driving safely, living healthy).

Human action mainly influences the occurrence of hazard, people who stay away from risky situations have little chance of becoming in an accident. Although this is not guaranteed, the chance of getting in an accident is partly determined by a factor of luck or bad luck. Individuals want to insure themselves against these unknown probabilities. People tend to insure themselves when they face the chance of hazard occurrence. There is a variety of literature on insurance behaviour and why people insure themselves. However, the introduction of performance-based insurance is only recent and has different characteristics than traditional insurance and is therefore interesting to study. Therefore, a contribution is made towards the limited research into decision-making involving privacy trade-offs when attaining modern day goods and insurance behaviour.

Former research essentially studied insurance behaviour where people could choose from no insurance or traditional insurance. The focus was mainly on the effect of different hazard probabilities on insurance preferences and irrational insurance behaviour. This research contributes to existing literature on insurance behaviour by examining how preferences change by adding a new insurance option. Up to now, insurers were unable to monitor the behaviour of consumers. Technological developments, such as telematics enable this possibility. To date, no research has been conducted on whether people would choose insurance that determines the premium based on performance. Also, it is tested if these preferences differ when individuals face higher costs for insurance. In this area too, little research has been done, where probabilities of hazard and underlying losses remain the same, but only premium costs become higher. According to expected utility theory, higher costs for insurance should allow for more risk-taking behaviour, which can benefit the attractiveness of performance-based insurance. It is tested if people behave rational when facing a high premium environment.

Also, a source of uncertainty arises with performance-based insurance and comes from privacy concerns. Products or services increasingly require exposure of personal data in order to use or obtain it and uncertainty arises from unknown preferences for privacy exposure or protection. Furthermore, people lack information about the consequences of sharing data, and this results in irrational decision-making. Therefore, a literature review is added to this research to examine the effect of privacy concerns on resistance of performance-based insurance.

This research distinguishes itself because a new innovative pricing model is investigated with great adoption prospects in various industries. Also, data collected in the maximum differences (maxdiff) experiment is analysed with two different techniques, counting analysis and the more advanced latent class analysis. The maxdiff analysis is ideal for filtering preferences with

no opportunity for scale use biases and therefore discover attributes are most decisive when choosing insurance.

This research is proceeds as follows. First literature review is outlined which provides an overview of causes for irrational behaviour when consumers face decisions involving privacy trade-offs. At the end of the literature review the main effects will be applied to PHYD insurances. Section 3 firstly introduces the insurance experiment and the maxdiff experiment, then experimental design and procedures are presented. Section 4 explains anticipated outcomes of the experiment. Then, results and interpretations are outlined in section 5. Followed by a discussion and conclusions. The next section briefly explains which UBI forms there are. After that, the privacy concerns of people are discussed and its effect on consumer behaviour. In section 3, the experimental design and procedures are presented. Based on available research, section 4 explains anticipated outcomes and behaviour. Thereafter, the results of the experiment and the interpretations are presented, followed by a discussion and conclusion.

2. Literature review

Firstly, the current state of the Dutch motor insurance market is lined out. This will provide the necessary insights into why the consumers feel resistance for performance-based insurance, and why insurers would like to sell this product. Secondly, according to the literature, the influences of uncertainty and privacy concerns on consumer decision-making are explained. Finally, conclusions of previous research are applied to PHYD insurance characteristics, and a motivation for the insurance experiment is given.

2.1 The Market The Consumer

Analytics and data collection are a fast-growing multibillion-dollar business. It is questionable to what extent users understand how various organisations obtain their personal information. On a regular basis newspapers post about large data collectors dealing with data leaks, which means people’s personal information can be everywhere and unsecured. This kind of newspaper reports results in an increasing aversion to exposing personal information. A reason why privacy security issues arise and affects so many people is that, while there was an increase in the number of data collecting firms there has been no similar increase in efforts to secure it.

The usage of mobile devices is widely accepted, and everyone wants to be continuously ‘connected’ to friends and family. Nowadays almost everyone uses mobile applications to collect

information by email, phone, social media platforms or alternative channels. Individuals are generating more and more data, which exposes valuable information about consumers preferences, and habits (Alessandro Acquisti, Brandimarte, & Loewenstein, 2015). Corporations and authorities are trying to interpret data from individuals to prognosticate their changing demand and preferences and strive to create more value for consumers and themselves. Individuals, businesses and society as a whole can benefit from sharing information and from the applications of increasingly complicated analytics of large datasets (McAfee & Brynjolfsson, 2012).

New technologies enable businesses to innovate and make processes more efficient, convenient and cheaper. However, there are also concerns about exposing and sharing too much personal information. Consumers are forced to decide for themselves what they want to share and what information they want to keep private. While data-driven firms are continuously trying to tempt consumers to give up personal information in exchange for small discounts or other rewards. Resulting in complex trade-offs where consumers have to invoke their own ability to manage privacy (Solove, 2013).

Many consumers have concerns regarding the privacy of the data they share with insurance companies, and they question insurers’ ability to safeguard their data given the recent cases of major corporate security breaches. The consumer cannot foresee the consequences of giving up all this information. All available data about an individual gets cross-referenced, or leaks and one’s data ends up used for many purposes by different firms, known as at the ‘dossier effect’ (Goldberg, I., Wagner, D., & Brewer, 1997). Consumers fear that their information is traded to third parties and people can no longer keep track of who is in possession of their data. Individuals face uncertainties, incomplete information and bounded rationality when purchasing a product or service involving a privacy trade-off. Acquisti & Grossklags (2005) state that perfect rationality no longer captures the nuances of a consumers’ decision making and is disturbed by various factors. A term often referred to is the privacy paradox, consumers express their concerns that their ability to control their data is violated in the marketplace, however, despite the complaints, it appears that consumers freely provide personal information (Norberg, Horne, & Horne, 2007).

However, privacy concerns cannot be the only source of resistance to PHYD insurance in the Netherlands. After all, drivers in the UK and Italy face the same degree of privacy intrusion when choosing PHYD. Though, both countries had its reasons and incentives to adopt PHYD increasingly. The motor-insurances prices in the UK are high, in 2017, approximately 60 million pounds of insurance costs were saved by adopting the new policies. Mainly the younger generation drivers benefitted, young men between the ages of 17 and 20 faced the most expensive premiums. The market for PHYD is most mature in Italy, in 2016 the market share was 15% and is expected

to rise to 27% by 2020 (Deloitte, 2016). Telematics is widely used in the southern areas of Italy where tariffs and insurance fraud are higher and can, therefore, result in a more prominent decrease in premium prices for low-risk drivers (Reuters, 2017). Also, telematics is stimulated by the Italian government which benefits the adoption of PHYD.

Citizens of the Netherlands are one of the most insured people in the world if looking at the average premium turnover per citizen (Verbond van Verzekeraars, 2017). Besides, Dutch people are also satisfied with their insurers. Insurers get the highest grading compared to other sectors within the financial industry. Compared to the UK and Italy, the Dutch drivers have no obvious incentive to drastically change insurance behaviour. However, Dutch motor-insurers deficits still haven’t subsided enough in 2017, and some newspapers and articles predict premiums are likely to go up in the future (Business Insider, 2018; Volkskrant, 2018).

The Insurers

On the other side, firms across different industries are increasingly using technologies based on real-time consumer data to create more value and enhance their productivity. Examples of such innovation are telematics, self-driving cars, blockchain technologies and the use of big data and machine learning.

The current Dutch insurance sector is currently in a downward spiral. The business model of the insurance sector has been under pressure in recent years. DNB warned the industry in 2016 for their growing deficits, deteriorating financial position and decreasing solvency. The low-interest environment, declining premium volumes, technological developments and high competition in the market have been stimulating factors. Especially motor-insurances are suffering; the claim costs have been increasing for several years. Not caused by unforeseen developments, and insurers could, therefore, have responded by adjusting their pricing. However, the market is saturated and is in heavy price competition. Consumers are mainly guided by price when choosing insurance, stimulated by comparison sites. In this situation insurers often offer non-cost-covering premiums to capture market share, the underlying strategy is to win new customers and sell them other products (De Nederlandsche Bank, 2016a).

By investing their funds, Dutch insurers hope to make a reasonable return to cover some of their claim costs, but in 2014, most of them did not succeed. The number of car claims is decreasing. However, the claims get more expensive. The claim costs rise due to higher car repair costs, extreme weather and usage of mobile phones while driving. The Dutch Bank urges insurers to tackle the current situation and set up a healthy business model. They suggest that specific innovative and economic developments may bring potential opportunities. Namely, innovative

technology, changing customer behaviour and economic growth.

Even though there is currently no major incentive to switch insurance policy for the Dutch consumer, PHYD benefit both insurers and consumers. Still, there is little knowledge if deficits of the motor-insurance market will improve if PHYD will be adopted.

Benefits consumers and insurers

The first effect where both insurer and consumer can benefit from is change in behaviour. The most interesting advantage is that PHYD can increase profitability of insurers because they can assess risk more accurately, insurers benefit from PHYD because the technology enables them to minimise risk exposure (De Nederlandsche Bank, 2016b). If buying PHYD results in ‘safe’ driving behaviour then consumers benefit from the new insurance as well, they can enjoy a more accurate and fairer premium and no longer they have to pay for the risk of ‘unsafe’ drivers.

Change in behaviour is expected through smart use of sensors or chips that can help prevent damage by giving feedback to drivers and improve driver’s safety. PHYD can positively change driving behaviour because drivers know their behaviour is monitored and priced accordingly. Also, users get feedback on their driving behaviour resulting in fewer accidents (which also saves money) and reduce the severity of accidents (Soleymanian, Weinberg, & Zhu, 2017). This will have a direct effect on the overall welfare of the insurance sector, this benefit is divided between insurer and consumer.

The second benefit comes from a selection effect. Adopters of telematics could benefit from drivers’ self-selection, people who know that they are driving safe are more likely to take out PHYD insurance. Though, the net welfare of the insurance sector would not benefit from selection, because ‘bad’ drivers also must be insured.

Other benefits for consumers of PHYD are value-added services like emergency services, stolen vehicle tracking and vehicle diagnostics. Plus, PHYD has a potential to reduce fraud, the data that is retrieved provides a solid basis for assessing the validity of claims.

2.2 Decision-making involving privacy trade-offs

This section reviews the factors that affect the decision-making process when asked to expose personal information. Lessons from behavioural economics are applied. The first theme discusses individuals’ uncertainty about the nature of privacy trade-offs and their preferences over them. The second theme explains the existence of a privacy paradox and how it can explain current behaviour. Then the importance of context is discussed.

Privacy:

Traditionally, privacy has been conceptualised as a Right to control information about oneself. Privacy can be defined as an individual being able to determine for themselves how, when, and to what degree information is transmitted to others (Boguslaw & Westin, 1968). Privacy concerns can be affected by various activities, such as information collection, processing, distribution and intrusion (Solove, 2006). Privacy trade-offs involve uncertainty and are likely to play a significant role, but many other theories that can contribute to a final decision.

2.2.1 Uncertainty from privacy trade-offs Incomplete information:

The first difficulty when facing a privacy decision is incomplete information. This factor distorts choice when people face a decision which involves a privacy trade-off. The consequences and what outcomes may occur in different contexts are a major unknown when giving up personal information (Alessandro Acquisti & Grossklags, 2008). Consumers often know less than data collecting firms about the magnitude of retrieved data and the use of the acquired or shared information, which is extracted willingly and knowingly or not. This leads to asymmetric information. In addition, companies are often very vague for what other purposes the information is being used. Consumers rarely have explicit knowledge of what data companies, authorities or other people have about them, or how those data are used and with what consequences. Actions taken by firms are hardly predictable, leaving individuals unknowing whether past shared information is for example used to exploit price discrimination strategies. Because people lack such information or are not aware of their ignorance, they are likely to be uncertain about how many data to share (Alessandro Acquisti et al., 2015).

Gains and costs associated with privacy exposure are context-specific and complicated. They require trade-offs, are bundled with other products or services and are often observed after privacy invasions have taken place (Alfssandro Acquisti & Grossklags, 2005). The privacy ‘good’ is usually attached to other goods in complex bundles, meaning to get access to a good or reward, individuals have to make a privacy trade-off. For example, consumers share purchasing patterns by using a loyalty card in return for price reductions. Comparisons between these different ‘goods’ are hard because of their combinatorial character, and it becomes even more complicated if offers are uncertain or ambiguous (Alessandro Acquisti & Grossklags, 2008). Privacy harms and potential consequences are hard to ascertain, some harms are tangible some are not. Tangible harm could be a monetary cost of identity theft, an example of intangible cost is being spammed by numerous advertising emails.

Bounded rationality limits our ability to include all information available in our decision, and it makes us rely on a simplified heuristic or strategy. Individuals do not have access to complete information and are unable to process and remember which personal data is available to others. Moreover, even if people had access to all information and could calculate the optimal strategy for privacy-sensitive decisions, they still would deviate from rational theory (Alfssandro Acquisti & Grossklags, 2005).

Unknown preferences:

Another source of privacy uncertainty is related to our privacy preferences. Slovic, (1995) studied preference uncertainty and showed that individuals often have little sense how much they prefer goods over other goods, privacy seems to be no exception. Different studies tried to measure individual and group differences and found important variations in privacy preferences. Privacy perceptions vary significantly across populations and within specific demographic characteristics, such as education (Culnan & Armstrong, 1999). Lower educated people tend to express more concern about how firms use personal data compared to higher educated people (Phelps, Nowak, & Ferrell, 2000). Older people found to be more concerned than younger people, and higher income is positively related to privacy concerns (Zukowski & Brown, 2007).

Also, some specific personal details are perceived as more private than others. Individuals expressed to be more resistant to disclose financial, family or medical information than to share information about media usage or brand consumption (Horne, D. R., & Horne, 1997). Several studies used attitudinal scales tried to predict the dependent variable privacy concern. However, the prediction power of attitudinal scales for actual privacy behaviour was quickly questioned in the literature (Lubin & Harrison, 1964). However, more recent literature agree that concerns and attitudes are not a valid predictor for actual behaviour (Alessandro Acquisti & Grossklags, 2008; Beresford, Kübler, & Preibusch, 2012; Norberg et al., 2007).

2.2.2 Privacy paradox

The theory explaining the inconsistency between behaviour and attitudes towards privacy disclosure has become known as the ‘privacy paradox’. Various studies tried to explain the dichotomy between privacy concerns and actual behaviour (action). Spiekermann, et al. (2001) experimented in the context of e-commerce. During online shopping, they tried to compare self-reported privacy preferences with actual exposure behaviour. Subjects were first asked to complete a questionnaire on privacy attitudes and preferences, then, to visit an online store. During their shopping in the store, they were encountered in a sales dialogue with a shopping bot. Consumers

answered most questions, even if these were highly private. Indicating that even though internet users claim that privacy is a high priority, they do not behave accordingly. On the other hand, some studies challenged the hypothesis of the existence of a paradox, a survey conducted by Lutz and Strathoff (2014) and found a weak but statistically significant effect of privacy concerns on protection behaviour.

An explanation for the existence of the paradox might be, that the paradox is elusive, meaning that broadly defined intentions and action should not be closely related to attitudes. Even though an individual cares deeply about its privacy and is aware of the risks involved, individuals perform a ‘privacy calculus’, and depending on the costs and benefits prevailing in a specific situation decide to disclose personal information or seek privacy protection (Jiang, Heng, & Choi, 2013). Literature in psychology explains how individuals mispredict their future preferences or draw wrong conclusions from past choices. Consumers also suffer from self-control biases; people tend to trade benefits and costs in ways that can harm our future utility in favour of immediate gratification. When ‘calculating’ net utility, the immediate benefits outweigh the risks and costs in the long term, leading to a low valuation of privacy (Alessandro Acquisti et al., 2015).

Considering the attempt of Acquisti et al., (2013), which was inspired by literature on endowment effect, they attempted to estimate accurate valuations individuals assign to privacy. They gave half of the subjects a $10 ‘anonymous’ gift card and the other half a $12 trackable card, where their name would be linked to certain transactions. Then subjects were all given the opportunity to switch. The subjects with the ‘anonymous’ card were asked if they would accept an extra $2 to allow transactions to be linked to their name. Participants with the trackable card were asked if they would accept a card with $2 less value to make their card ‘anonymous.’ Most people with the anonymous card kept their card five times more than people who had the trackable card. Suggesting that there is an endowment effect, and people who have privacy, value it more than when they do not have it. This finding is compatible with the conclusion that older people are more reluctant to provide personal information. People probably attach a higher value to their privacy because they have not had to give up much information in the past and believe that much information about them is still private.

Another explanation for the paradox is that habitual intrusion ensures that less attention is paid the privacy trade-off. Integration of Social Networking Sites (Debatin, Lovejoy, Horn, & Hughes, 2009) and the fact that people engage in privacy trade-off decisions all the time, results in people getting used to the degree of intrusion. Revealed preference theory would then conclude that because technologies for information sharing have been enormously successful, whereas technologies for information protection have not, individuals hold overall low valuations of privacy

(Alessandro Acquisti et al., 2015). Stimulated by dual system theory would conclude that habits and repeated choices are made in system 1 (Kahneman, 2003). Choices in system 1 are described as quick choices that are not thought through, made very quickly, automatically, and governed by habit. These decisions are therefore difficult to control and change, which could be an explanation for low privacy values when confronted with privacy decisions on a daily basis. Decisions in system 1 are extremely context-dependent, which is implied by the concept of accessibility.

2.2.3 Context dependence.

An individual can be unaware of, but in other situations be extremely concerned about privacy issues. Kahneman & Tversky (1984) have shown that how a question or problem is framed affects how subjects respond to it. When people are uncertain or uninformed, they tend to search for guidance or signals in their environment. Since guidance is related to context, behaviour is as well. The guidance that individuals use to judge the importance of personal information may result in sensible behaviour. For example, the presence of authority regulation has been shown to lessen consumers concern, and people use it as a signal there is a certain degree of protection (Xu, Teo, Tan, & Agarwal, 2012). However, in other situations, disclosing behaviour is insensible, in an online-experiment participants were less likely to share information on a formal website than on an informal website where a banner stated: “How Bad R U”, yet the formal website was perceived as safer (John, Acquisti, & Loewenstein, 2011)

Another signal that influences perceptions of privacy is descriptive norms. When people got to know that others disclosed information, the chance that one will reveal information as well increases, known as imitation (Alessandro Acquisti, John, & Loewenstein, 2012). This makes our decisions malleable, and some entities have developed expertise in exploiting behavioural processes and framing biases to promote information exposure (Calo, 2014). For example, the default setting is a frequently used marketing trick to nudge people to disclose information. Extensive research on the default bias or status quo bias has shown its effectiveness (de Haan & Linde, 2018; Kahneman, Knetsch, & Thaler, 1991). Another variable studied which could contribute to consumers’ willingness to share information is; the level of trust in an organisation. Schoenbachler and Gordon, (2002) asked 5000 subjects to express their willingness to provide information as a depended variable upon the degree of trust in the firm, which is asking for information. They concluded, more trust stimulated willingness to provide information although this varied across industries.

Concluding, people are uncertain about their preferences and suffer from asymmetric information when facing a decision. People have the best intention to protect their personal

information however biases affect our risk-benefit analysis when deciding to disclose information or not. The rules people follow to manage their data vary by the situation are based on demographics, are learned over time and influenced by context. This results in inconsistent and irrational choices. Also, when exposing personal information leads to a direct gain then this will probably weigh more heavily in the judgment than when benefits are not immediate, and the valuation of personal data will be different. Once an individual has started to expose personal information, the threshold to share something next time will be lower. When encountered with the same decision on a regular basis people will make decisions with less attention and both risks and benefits will be valued less or even be ignored.

2.3 Insurance Behaviour 2.3.1 Uncertainty

One of the most influential theories about choices under uncertainty is the prospect theory (Kahneman & Tversky, 1979). It provides an interpretation of how people value and compare uncertain losses and gains. The theory implies that individuals’ valuations around losses and benefits can be described as an S-shaped value function starting in the centre which represents the reference point. The function is kinked at the reference point and is steeper for losses than for gains. In the domain of benefits, the function is concave, favouring risk-aversion, in the domain of losses the function is convex and favouring risk-seeking behaviour. Meaning when the same absolute change appears from the reference point, the change in value looms larger for losses than for gains, resulting in a decision where the expected utility theory no longer holds. Furthermore, Kahneman and Tversky (1979) proposed that very low probabilities be overweighed, which is a reason why people insure themselves or buy lottery tickets. Individuals either overweight or ignore extreme possibilities because of our limited ability to evaluate them. In addition, people prefer known risk over unknown risk, meaning that people are ambiguity averse (Ellsberg, 1961). 2.3.2 Behavioural economics and insurance behaviour

The reason why individuals purchase coverages against both small and large losses is studied extensively, and there is substantial evidence to believe that the utility function for money is concave. Why otherwise do people buy expensive insurance that exceeds the expected costs? However, there are some contradictions with risk aversion in some various forms of insurance, suggesting that the value function is not concave everywhere. Insurances that offer limited coverage with a low or no deductible are often preferred over complete coverage with higher deductibles with similar policies (Munch, 1976), which is contradicting with risk aversion.

A insurance program called Probabilistic Insurance was introduced to research the contradiction. With a Probabilistic Insurance an individual pays half of the regular premium, but in case of damage, the probability that your claim is reimbursed is a half (Kahneman & Tversky, 1979). They compared it with prematurely replacing your tires or stop smoking, these actions reduce the chance of unfortunate events, but do not eliminate risk altogether. Imagine Probabilistic Insurance as follows; if a hazard occurs on an odd day, then you will have to suffer the losses yourself. If it happens on an even day, the insurer will pay the costs of the claim. According to expected utility theory, assuming concavity, Probabilistic Insurance should outperform traditional insurance. However, this insurance seemed most unattractive, suggesting that reducing the probability of a loss from 100% to 50% is less valuable than lowering the probability of a loss from 50% to 0%.

Wakker, et al. (1997), elaborated on this insurance and found as well that people value elimination of uncertainty disproportionally more than the reduction of uncertainty. A one percent deviation from certainty results in a dramatic decrease in the attractiveness of a prospect. But here too, context can influence preferences. Slovic, Fischhoff and Lichtenstein, (1982) state that a hypothetical vaccine that decreases the chance of catching a disease from 20% to 10% is less engaging if it is described as ‘effective in half of the cases’ than if it is presented as ‘completely effective against one of two exclusive and equally probable virus strains’ (Kahneman & Tversky, 1984).

2.4 Findings applied to Performance-based insurance

Previous research argues that comparisons between personal information sharing and rewards are complicated for customers when facing uncertain offers. Deciding on insurance is not a decision consumers make on a daily basis. Therefore, insurance is chosen slow effortful, rule-governed and every attribute is consciously included in the decision. Pay How You Drive (PHYD) has several attributes that differ from traditional insurance (TI) and therefore contribute to the decision to switch or not. People do not know what the consequences are when sharing their real-time location and put a device in their vehicle which tracks every movement; additionally, the compensation (a good driving score) is also uncertain. Consumers suffer from bounded rationality because of incomplete information. Insurers who offer PHYD describe a ‘safe driver’ very vague and do not mention after how many ‘mistakes’ your score is inadequate and will not get any discount. Insurers do not give insights into the algorithms who determine driving scores However, giving insights into the complex algorithms would probably give not much clarity for consumers.

When making a ‘privacy calculus’, PHYD is at a disadvantage. According to the revealed preference theory, TI has already proven to work well, and most people take out this insurance, where PHYD still has to prove its potential. Also, there is no effect of immediate gratification when a consumer takes out PHYD, in fact, consumers have to pay a higher premium beforehand for PHYD compared to the cheapest traditional policy on the competitive Dutch market. At the end of a particular period drivers learn if they are considered as a safe driver and get a discount on their premium, but first, consumers have to pay for a higher premium and give up information on action, behaviour and location-based services. So, if drivers consider PHYD insurance they will put more weight on the privacy cost than the benefit of getting a potential discount.

Uncertainty also arises from consumers’ unclear preferences over sharing our location-based services, action and behaviour with insurers; it can be considered as reasonably sensitive and personal information, despite that most people plausibly already share this (ignorant) with various mobile applications. Besides privacy valuations differ across demographics. Derikx, de Reuver and Kroesen, (2016) found that consumers value information on action and behaviour more than the privacy of location. Findings on privacy involving the endowment effect suggest that people value information more when they believe it to be private. Data on behaviour and location-based services used to be private, so exposing this information will negatively influence the choice for PHYD. The PHYD insurance is still in its infancy in the Netherlands, and the adoption rate of both insurers and drivers is low in the Netherlands. As a result, the current pricing mechanism of PHYD cannot enjoy effects of descriptive norms, such as word-of-mouth advertising or any form of imitation. Often when innovative products or services are introduced, many consumers let young people or early adopters explore and test the products first before considering buying it themselves.

According to prospect theory and Probabilistic Insurance, individuals make irrational choices when uncertainty is involved. People prefer to eliminate risk rather than reducing risk by the same absolute value. People can influence their driving score and behaviour but cannot eliminate the risk entirely of not getting the discount. With PHYD, consumers also have the option for different coverages, but the monthly premium is uncertain and determined by personal driving score. The monthly premium of PHYD, without any discount, is usually higher than the cheapest traditional insurance premium. If one does not get the premium, this feels like a loss, because one could have been better off by choosing traditional insurance. Since, change in values loom larger for losses than for gains, people put more weight on the chance of not getting a good driving score. The probability distribution of not getting any discount may depend on external factors. The ‘unknown’ chance of not getting the discount may be overweighed; also, people tend to be ambiguity averse, preferring known risk over unknown risk.

3. Experimental Design and procedures

The easiest way to research preferences for performance-based insurance is to observe behaviour in a controlled environment. Designing an experiment is the easiest way to do this. The main advantage of doing an experiment is that a controlled environment can be created. When using field data, variables are often affected by other events that influence outcomes. An experiment allows for testing behaviour and preference for different insurance policies within a controlled environment. This insurance experiment tries to test preference for the core idea of PHYD compared to traditional insurance and excludes privacy concerns.

First, a situation needs to be created where participants can insure themselves. Then something is needed that represents good driving behaviour; a task has been devised for this. People can score points when answering tasks correctly. Scoring well on the task represents good driving behaviour. Therefore, it must be a task where the final-outcome is ambiguous and cannot be influenced by skill. Final-score in this experiment can be influenced by luck or bad luck, similar to the fact that an external factor can influence the probability a driver ends up in an accident or just escapes it. A random variable is, therefore, included which influences the final-score on the task.

People insure themselves against a ‘small’ probability of incurring a large loss. In this experiment, participants can insure themselves against losing their endowment in tokens. People lose their endowment when they do not reach a certain threshold of final-score. Participants can choose from three types of insurance including a performance-based. From the insurance choices people make in the experiment, it is tested if people would find it interesting to pay a premium based on performance.

A high premium treatment is included in this experiment to test whether the success of PHYD in the UK has been stimulated due to high premiums.

3.1.1 Procedure

The experiment was computerised in PHP and MySQL. The experiment consists of a programmed individual decision-making task with no interaction between participants. At the beginning of the experiment participants are endowed with 100 tokens. Earnings are in ‘tokens’, which are exchanged for euros, at an exchange rate of 30 tokens = 1 euro, at the end of the experiment. From the 121 participants, 1 out of 15 is randomly selected for payment. They earned 2.50-euro average. In this experiment, the focus is on the demand side of the market for insurance policies. In

read the programmed instructions and answer two control questions about their understanding of the task and game. After they correctly answer the control questions, the experiment begins. The experiment consisted of three phases, a decision phase, the 3-minute game, and a maxdiff experiment. In the decision phase participants have to decide on insurance, to be covered or not for uncertain outcomes during the game. In the next phase participants have 3 minutes to solve as many tasks as they can, then participants are directed to the results page where they learn how they performed on the tasks and how many points they earned. After that, the maxdiff experiment follows.

3.1.2 The Matrix task

The task is simple to understand. Two matrices are shown on the screen, the size of the matrices remain the same during the experiment, and both have a 5 X 5 format. The matrices consisted of 25 random integers ranging from 55 to 100. The purpose is to find the highest integer of the left matrix and find the highest integer of the right matrix, then add those numbers up. The sum is the answer to one task. The task remains the same over the 3 minute-period to solve tasks. During the game, there is no correction possibility, so once an answer is submitted, the next task is loaded immediately. The other attributes of the experiment are explained below. An example of a task is shown in Figure 1.

Figure 1: Example task of experiment, find the highest number of both matrices and add those up, this is the

3.1.3 The Game

Participants are endowed with tokens, representing euros. Participants are asked to solve tasks for 3 minutes. For every correctly answered task, they earn five points. The personal score is determined by the number of tasks solved times five points. However, the final-score is adjusted by a random variable, defined by means of a lottery. The values of the random variable range from minus seven to plus seven points [-7; +7]. The random variable represents the uncertain factor which affects final performance, similar to the chance of getting in an accident is partly determined by luck or misfortune. The goal of the game is to achieve a certain threshold in points. When this threshold is reached the participant can keep all tokens. The threshold for winning the game is to earn 30 points of the final score, including the random variable. For example, to keep all tokens, the participant has to answer at least six tasks correctly within 3 minutes, his personal-score is then 30 points. However, if the participant gets a negative random variable assigned, the threshold is not reached.

3.1.4 Policies

Insurance policies are presented in the rows in the table, denoted as insurance: A, B or C. Each policy has a different coverage for uncertain profits and is characterised by its premium. Payoffs are shown in Figure 2. When participants choose covering insurance, they pay a traditional premium for insurance B, denoted as the ‘fixed premium’, or a ‘variable premium’ for insurance C representing the performance-based premium. Neutral labelling for policy names is chosen, which is in line with traditional experimental economics, to avoid noise in the results. Because there can exist uncontrolled preferences by participants to words as; ‘WA insurance’ or ‘all-risk’,

Option A does not offer any coverage for uncertainty and means the participant chooses not to be insured. If the participant does not reach the threshold of 30 points of final-score, the participant loses all 100 tokens. If it does reach the threshold including the random variable, the participant can keep all tokens without paying for insurance.

Option B represents traditional insurance; this insurance covers all uncertainty. If the participant selects insurance B, this means that he or she is covered for any outcome. Doing tasks for three minutes is just a formality. The performance on the tasks and the random variable which affects the final-score in points will not influence the final payoff in tokens. Participants know with certainty how many tokens they can keep at the end of the experiment. Namely; 100 tokens minus ‘fixed premium’.

Option C represents the performance-based insurance, with this insurance the participants can lower its ‘variable premium’ with the final-score of the game. The final amount of tokens

participants can win is influenced by their performance on the task and the random variable they get assigned. See figure 3 for an overview of the payoff formulas.

Figure 2: Formula per insurance, presented to the participants.

3.1.5 Treatments

As soon a potential participant clicks the link that refers them to the online experiment, the participant gets a ‘participants number’ assigned. All even ‘participant numbers’ are placed in the high premium treatment, and all odd numbers are placed in the low premium treatment. The fixed premium in the low premium treatment is 56 tokens, the variable premium is 67 tokens. In the high premium environment, the fixed premium costs 78 tokens, the variable premium costs 89 tokens. Because some participants clicked on the link and then left the page, not all treatments have the same number of participants. There were no treatments in the maxdiff experiment, and all 121 participants completed the same questions.

It is believed that many of the results apply to a simplified insurance choice. When choosing insurance, a consumer cannot consider all policies. A bounded rational consumer is assumed, and therefore the number of policies are limited. For the sake of simplicity of the experiment, only three policies are included and represent simplified translations of Dutch motor-insurances. Option A is the risky choice and in this experiment zero-premium insurance and can be seen as a Dutch WA-insurance (which only covers damage and costs inflicted to others by the driver and is only encouraged to take out by advisory sites when a driver owns a car, whose market value is low). When a participant chooses policy B, or ‘traditional insurance’, he buys off any form of uncertainty, similar to drivers with all-risk insurance pay for all uncertainties even if complete coverage is not always needed. Insurance C is insurance where the premium is based on performance and action, representing PHYD insurance. Drivers are asked to take out an all-risk or WA insurance or

performance-based insurance. The performance-based insurance is added to test driver’s preference for this insurance.

3.2 Maximum Differences experiment:

A Maximum Differences (maxdiff) experiment is designed to examine relative preferences and attributes which are an important determent when buying insurance. Likert-scale questions would be problematic when researching preferences for insurance, asking people to rate traits of good insurance would not be insightful since they are all important when deciding. Leading to results that may not be discriminating. Ranking items only examine the order of importance and not the strength, also for a large set of items the middle rankings might be fuzzy and inaccurate. Maxdiff forces participants to make trade-offs between the alternatives in the set and enables better differentiation between the items. Maxdiff is a methodology developed by Louviere and Woodworth, (1983), that results in interval scale estimations that are based on relative judgments that can be answered quickly.

3.2.1 Experimental design

Each question presents a participant with a set of four alternatives. Each question asks to select the most important and least most important item in the set when choosing new insurance. This statement remains the same for all seven questions. The experiment consists of seven questions and seven different items. The design of the experiment is composed carefully. Every pair of items must be shown twice, and every participant must see each item four times. If the experimental design is not perfectly balanced, some items and pairs are displayed more often than others, this leads to a skewed distribution of observations. Table 1 presents the construction of experimental design, and shows which items are presented in which round. This technique obtains the relative degrees to which the items are seen as an essential decision factor. Every round, the question stated: “Which of the following items do you think is most and least important when deciding on insurance”. Depending on which question participants face they could choose from the displayed items.

The following items are included in the experiment: - "Preference for a well-known insurer." (Brand) - "The lowest premium." (Premium)

- "High service standards." (Service)

- "Transparent and clear policy terms." (Transparency) - "Positive reviews of others about the insurer." (Reviews) - "Preference for my current insurer." (Status Quo) Table 1: Maxdiff design; order of alternatives displayed

3.3 Participants

The 121 participants were predominantly highly educated, 99 participants had a bachelor’s degree on University level. 45% of the participants were female, and the average age was 33, varying between 19 and 79 years.

4. Expected results

4.1. Rational Agents and Expected payoffs.

Standard expected utility theory predicts that people use expected utility formulas to determine preferences regarding choices that have uncertain outcomes. Based on subjective probabilities participants have to decide for themselves how they want to be insured, if they want to be insured at all.

The subjective chance of reaching the threshold influences the insurance decision. If a person expects to score very well and solve at least six tasks in three minutes, then no insurance, option A, would be an appropriate choice when risk attitudes are disregarded. When a participant decides to insure its endowment, it can choose from an all-risk insurance, option B, and the performance-based insurance, option C. Option B, is 11 tokens cheaper than option C. When people expect to solve less than three tasks then option B is most obvious. If one expects to answer at least 3 tasks correctly but no more than six option C is the appropriate insurance.

One may expect to be very good at solving tasks but may dislike the uncertainty that comes from the random variable. Therefore, a participant who is highly risk-averse may decide to insure

Question Alteranative 1 Alteranative 2 Alteranative 3 Alteranative 4

1

Service

Transparency

Reviews

Status Quo

2

Brand

Discount

Reviews

Status Quo

3

Brand

Premium

Transparency

Status Quo

4

Brand

Premium

Service

Reviews

5

Premium

Service

Discount

Status Quo

6

Brand

Service

Discount

Transparency

himself even though he can reach the target easily. Because of different risk attitudes, insurance preference is influenced by both performance expectations and risk attitude.

At last, premium prices might influence the expected payoff and insurance preference as well. Assuming that the decision to insurance oneself depends on subjective chance, risk attitude and price of an insurance. Changing insurance costs influences the point where someone decides to choose to insure himself or not, while subjective chances to solve six tasks remain unchanged. Different insurance proportions are expected over the treatments where the difference between the two price environments is 22 tokens. If a participant who faces high premiums chooses option B, the all-risk insurance, he pays 78 tokens, so he will only have 22 tokens left at the end of the game. According to standard economic reasoning, demand of a good decreases when it becomes more expensive. Therefore, the relative attractiveness of option A increases compared to the insurance options B and C when the price of insurance is high. It is expected that a rational participant is more likely to opt for no insurance when facing high premiums. The difference between the premiums across the treatment remains 22 tokens, therefore the relative attractiveness between option B and C remains unchanged across treatments, so there is no reason to expect that people who decide to insure themselves will not change proportions for option B and C across the treatments.

Concluding, a risk-neutral participant who ignores the random variable, decides to choose the option with the highest expected value according to its estimated ability and subjective chance of reaching the threshold. When facing high premium prices participants that would prefer insurance may shift to no insurance since attractiveness and expected payoff of insurances decrease, and therefore tempting people to take on more risk. A rational participant who wants to be insured and chooses between B and C would not be influenced by high premiums because the difference between them remains constant over the treatment. The relative same proportions of option B and C are expected.

4.2 Prospect Theory

However, the prediction made according to expected payoffs and rational agents might not be realistic, people do not estimate their chances rationally, and not everyone who chooses option A will reach the threshold and keep all tokens. Research on decision-making under uncertainty has proven that this theory does not always hold when making decisions involving uncertainties and argue the non-linearity of the value function. However, theories on the subject do not straightforwardly give a precise prediction of behaviour in this experiment.

function to calculate expected values. The weighing function of the perceived losses and uncertainties may differ per individual. The weighing function is steeper for losses than for gains, and small probabilities are often overweighed or completely ignored, the case of overweighing explains the existence of insurances. People want to insure themselves against the ‘small’ probability of hazard occurrence. This suggests that during the insurance experiment participants would choose option B or C rather than option A. Which policy the subject chooses may depend on the risk attitude, treatment, beliefs about personal ability and overconfidence. Overconfidence can arise due to optimism, and people will overestimate their ability to solve tasks. People who estimate to solve at least six tasks there is still a change of getting a negative random variable assigned or answer some tasks incorrectly. Solving more tasks decreases the relative impact of the random variable, therefore it can be seen as a small factor influencing final-score. However, the theory falls short to fully explain the complex problem. In case people overweight the probability of not reaching the threshold then preference for insurance is expected. In case people ignore the random variable and mispredict probabilities, risk-taking behaviour is expected favouring option A.

Also, Kahneman and Tversky (1979) introduced Probabilistic Insurance, for this insurance a low premium is paid. However, the probability the claim will be paid is uncertain. This insurance would have the same (or higher) expected utility as the traditional insurance. Though, it is argued that Probabilistic Insurance appears not attractive for most people while an expected utility maximiser should prefer this insurance. This insinuates that people are not risk-seeking when facing losses. Wakker, Thaler, and Tversky, (1997) tested Probabilistic Insurance in a different context and found that a 1% default risk demanded a 20% reduction in premium cost, and the weighing function of prospect theory predicts the reluctance to take such insurance. This implies that people prefer a sure loss over an uncertain one. In the case of performance-based insurance, participants are not sure how high their final premium costs will be, due to the inclusion of the random variable and uncertainty from one’s subjective chances to solve enough tasks. Which indicates people might prefer certain premium costs over uncertain premium costs, and therefore, prefer traditional insurance over performance-based insurance.

Other literature on insurance behaviour found that insurance extents to the medium range probabilities and argues that small chances of disaster are sometimes entirely ignored (Slovic, Fischhoff, & Lichtenstein, 1977). Besides, they argue that people buy more insurance when there is a reasonable chance that inflicts a small loss than facing a low probability inflicting a high-loss event (Kunreuther & Pauly, 2004; Slovic et al., 1977).

insurance behaviour depends on subjective probabilities of reaching the threshold and that an increased proportion will choose no insurance in the high premium treatment. However, prospect theory states that expectations are not linear and instead a weighing function and therefore subjective expected payoffs may be affected by factors like optimism or overconfidence to solve tasks and biases from small probabilities. Overconfidence or optimism favour preferences for no insurance, also some researched found that people tend to be risk seeking when facing sure losses and small probabilities might ignored. If this is the case a large proportion will choose option A, no insurance. On the other hand, the overweighing of small probabilities and in case from losses might evoke risk-aversion and stimulate buying insurance. According to prospect theory participants should prefer insurance to cover for a 100 percent loss. Findings on Probabilistic Insurance and prospect theory suggest people who will choose insurance where uncertainties are minimized, namely option B. However, when people who face high premium costs, people might find it less attractive to choose option B or C.

4.3 Self Selection

In reality, it is assumed that PHYD insurance will provide self-selection. Insurers expect to take advantage of PHYD by reducing costs by accurately assessing risk profiles but also benefit from self-selection. This seems obvious, insurance that rewards consumers for safe driving behaviour is more attractive for people who know they are safe drivers. There can also be an effect of awareness that makes people pay attention to behaviour.

However, there is a difference between saying to be a good driver and driving good accordingly. Similar to people who tell people to be a good driver but back out when anyone asks them to verify it with measurement. Anyhow, due to competition and loss aversion of the endowment, it is expected that participants who opted for insurance A and C will score higher on average than people who choose insurance B. This is because the profit from both options A and C depends on performance on the tasks. Participants who choose option B, they participate in the game, but it is only a formality since no factor can influence their final payoffs anymore. Therefore, there is no incentive to perform well on the tasks. It would be logical to predict a difference between performance of participants who choose option B and participant who choose option A or C.

If agents were perfectly able to estimate their abilities to solve tasks, then people who expect to solve more than six tasks will choose option A, people who think they are not able to solve more than three tasks are best off choosing option B, everyone in between is best off choosing option C. If expectations are formed according to this reasoning, it is logical to expect that people who

opt for A, perform best. Then people who choose performance-based insurance are expected to solve at least three tasks on average.

4.4 Demographics.

As is mentioned in the literature review, privacy concerns differ from individual to individual. However, research has found some commonalities across demographic variables. For example, older people and lower educated people are more reluctant about sharing personal information on the internet, than young people and higher educated people. This may also hold for insurance preferences.

Demographic variables are also affected by optimism biases. DeJoy (1989) hinted that optimism regarding risk accident and driving competency increased with driving experience and marginally with age. The effect of age on preference for performance-based insurances is ambiguous. Due to optimism and more driving experience than young drivers, older people may find PHYD attractive. On the other hand, people who are older earn more on average than younger people and will not deter from increasing premiums costs when they are satisfied with the current policy. Or else, elderly may feel less urge to to experiment with innovative concepts as PHYS and therefore prefer the status quo. Plus, in general, older people have a family and want to insure all passengers, which is included with all-risk insurance.

Optimistic perceptions of driving competency and accident risk have often been linked to the excessive involvement of young males in traffic crashes. DeJoy (1992) argued that both sexes reported substantial optimism, but men tended to be more optimistic when judging their driving skill. Men also tend to be more overconfident than women. This could affect the results of this experiment. If men are more optimistic than women and overestimate their ability to make tasks, this could result in gender differences in insurance preferences. It might not only result in a difference in terms taking risks but also differences in preferences for policy attributes. Therefore, policy attribute preferences are examined per gender in the maxdiff experiment. If men would be more optimistic about their ability to solve tasks than women, it is expected to result in more men opting for option A and more women opting for option C. Therefore, men and women can also have different preferences for insurance policy characteristics, this can be tested in the insurance and maxdiff experiment.