The Endowment Effect in a Field Study with Risk-Reducing Instruments

Abstract

The endowment effect is a phenomenon whereby ownership increases the perceived value of goods, thereby reducing the willingness to trade. This paper presents the results of a field experiment, designed to assess the robustness of this effect when decisions concern unique, non-tangible instruments of a very high utility. Participating students were unaware of being involved in the experiment, as we monitored their decisions concerning the examination bonuses in a natural academic setting. In contrast to the majority of prior studies that focus on commodities of a known and certain value, this research examines the effect in the context of risky and ambiguous instruments, which are more difficult to evaluate. Our results strongly support the endowment effect hypothesis, proving that the effect is robust, and likely to also be observed in markets concerning high-utility, risky instruments, such as financial ones. Additionally, we find a weak level of support for the gender effect, with female respondents demonstrating a lack of bias when endowed with ambiguous instruments. The study also analyzes the efficiency of these instruments in terms of their utility in risk reduction, and the effort incentives created.

1. Introduction

The endowment effect is a cognitive bias in which individuals assign greater value to objects simply because they own them, often exceeding the objective market value. The phenomenon was first identified in Aristotle’s Nicomachean Ethics but became widely recognized and studied in economics as “the endowment effect” since the work of (Thaler 1980). According to the Coase theorem (Coase 1960), when the property rights are clearly defined and there are no transaction costs involved, the initial allocation of goods is irrelevant, as economic agents are expected to engage in transactions that lead to the Pareto efficient outcome.

Economists have tested this hypothesis by conducting laboratory or field experiments. A traditional experimental design involves randomly distributing two types of commodities among participants. In the absence of transaction costs, half of the goods are expected to be exchanged. However, in most experiments, the number of transactions was significantly lower. For instance, in (Kahneman et al. 1990), participants provided with mugs and chocolate bars were found to be reluctant to trade their initial items when given the opportunity. An alternative design involves endowing half of the participants with one type of commodity to measure their willingness to trade it. This treatment enables researchers to observe that owners typically require higher prices for their items compared to non-owners (Knetsch and Sinden 1984; Kahneman et al. 1991). This effect has often been explained as a result of loss aversion, but nevertheless, a few studies question this hypothesis, with (Morewedge et al. 2009) claiming that the ownership, rather than loss aversion, causes the endowment effect, and (Weaver and Frederick 2012) arguing that the bias is more accurately explained as an aversion to bad deals, and (Smitizsky et al. 2021) and (Achtypi et al. 2021) suggesting that it may be better explained as a consequence of the (adaptively) rational decision-making of buyers and sellers.

The endowment effect can significantly impact the functioning of various markets. Naturally, it seems that this bias would reduce the number of transactions in second-hand goods’ markets. However, while (Bu 2023) confirms that the effect remains valid in the case of the second-hand clothing industry, (Da Silva et al. 2015) show that it diminishes once transactions are conducted online. In one of the earliest field studies on that topic, (Heberlein and Bishop 1986) found that Wisconsin deer hunters valued their hunting permits more than those they did not possess. (Hossain and List 2012) demonstrated how the endowment effect can be leveraged to increase work productivity, showing experimentally that employees exerted more effort to retain bonuses perceived as already owned. Similarly, (Faulk et al. 2019) indicated that students endowed with extra credit points worked harder to retain them. A study on the housing market by (Cheung et al. 2023) provided evidence that the endowment effect is reduced when a good enhances its exchange nature, such as through the provision of an exchange right. Furthermore, (Mwanyepedza and Mishi 2024) studied the housing market and the role that the disclosure of information plays in reducing the gap between the willingness to pay (WTP) and the willingness to accept (WTA).

Given its significance, the endowment effect has been tested in various research, aiming at assessing its magnitude in relation to factors such as perceived utility and future value. Concerning the value of the object, (Ortona and Scacciati 1992) conducted experiments to demonstrate that the endowment effect is stronger when the payoffs are high and fictitious but diminishes when the payoffs are high and real. (Bordalo et al. 2016) followed by proposing an explanation based on the concept of salience, which affects the valuations of buyers and sellers. (Jaeger et al. 2020) examined an extensive list of items and showed that the evolutionary salience score can explain 52% of the valuation differences. The properties that increase the magnitude of the effect are the object’s relevance to thriving and survival, and reproductive opportunities. Moreover, tangibility also enhances the strength of this bias. (Knetsch and Wong 2009) explored the role of the reference state, testing the conditions that alleviate the valuation disparities.

Surprisingly, the number of papers directly aimed at studying the role of risk in the case of the endowment effect, and its significance for the functioning of financial markets, is relatively small. (Loomes and Weber 1997) conducted two laboratory experiments with abstract context to study the endowment effect in relation to risky assets (lotteries). As the effect is significant, when subjects are incentivized, it is not that strong in hypothetical decisions. Using laboratory experiments, (Sprenger 2015) demonstrated the “endowment effect for risk”, while (Holden and Tilahun 2022) reached the same result by conducting ‘‘lab-in-the-field’’ experiments in rural Ethiopia. In these studies, participants were requested to make choices in the certainty equivalent and probability equivalent tasks. The results showed that participants demonstrated a preference for certain amounts and were unwilling to exchange such endowments for risky ones. To investigate whether the certainty effect may have influenced previous findings, (Crockett and Crockett 2019) conducted an experiment in which all lotteries involved some degree of risk and demonstrated a preference for prospects with lower variance. (McGranaghan and Otto 2022), using a standard example of chocolate, found that the more uncertainty that was involved, the bigger the differences in valuations, and therefore the stronger the endowment effect. However, the type of uncertainty matters, with their study emphasizing the role of intrinsic (value) uncertainty: the features of the participants’ preferences for a good, given its attributes. (Peñón and Reyes 2018) stressed the role of the endowment effect in shaping the risk-taking behavior of entrepreneurs, and demonstrated how the potential of losing their firms leads them to take higher risks. Additionally, (Gine and Goldberg 2018) ran a field study to show the endowment effect in the context of savings accounts, but did not address the problem of risk distribution directly.

Studies examining the role of risk factors in the endowment effect typically rely on laboratory experiments with abstract lottery choices or involve commodities of relatively low value (chocolates). Therefore, it would be insightful to investigate whether the effect persists when decision-makers are highly motivated and are making decisions about prospects with potentially high utility, but also significant valuation challenges due to high uncertainty (ambiguity). This is the main research gap that inspired our study. To address this, we conducted a field experiment, observing students’ decisions within their natural, academic environment. The participants were provided with two types of examination bonuses, which were unique, disposable and intangible risk-reducing instruments of high utility. We then monitored their willingness to exchange them. Although students were involved, it was not a laboratory experiment but a natural field study, as they were unaware of their participation. The examination bonuses differed by the properties, with one (PointBonus) being a more standard risk instrument, and the other (TimeBonus) being more ambiguous, with the value depending on unknown factors (the exam’s difficulty).

The novelty of our study results from three aspects. Firstly, we study the role of the endowment effect in the field of highly motivated participants. Secondly, instead of using lotteries and probabilities, we presented the participants with unique endowments that were much harder to evaluate. Each of them could be considered a risk-reducing instrument, but their actual utility was uncertain. Thirdly, some studies demonstrated that the endowment effect intensifies with time (Morrison and Oxoby 2013; Yamamoto and Navarro-Martinez 2022); however, this factor is rarely considered in experiments, where participants are typically asked to evaluate their endowments immediately upon receiving them. Yet, real-life endowments usually last much longer. The natural field experiment that we report allowed us to introduce the one-week time interval between the endowment and exchange stages, which increases the external validity of this study.

Thus, the endowments we chose were intangible, risky goods of a relatively certain intrinsic value but of an extrinsic uncertainty, as participants were unsure of their exact future role (McGranaghan and Otto 2022). We anticipated that this design could reduce or even eliminate the endowment effect, as the instruments had distinctly different properties and, potentially, a very high value for the participants. Additionally, we expected the participants to exhibit a preference for PointBonus, given its lower risk characteristics.

Nevertheless, the experiment provided support for the endowment bias hypothesis, as the unwillingness to exchange was observed for both instruments, deeming differences in their riskiness insignificant on the general level. At the same time, though, we found that there might exist a gender effect in the case of the endowment effect for risk, with female participants showing a higher preference for the PointBonus instrument. We believe that our results can have significant implications for studies on decision-making under risk, and for the functioning of the financial markets. Apart from testing the endowment effect, we use the examination results to conduct an ex-post analysis of the rationality of participants’ choices and the role of bonuses as incentive mechanisms.

The remainder of the paper is organized as follows. In Section 2 we provide detailed information on the experimental design and the hypotheses tested. Section 3 reports the results. Section 4 discusses the main findings and their implications, as well as the limitations of the research.

2. Methods

2.1. Participants and Type of Experiment

This paper presents the findings of an economic experiment. The experimental method follows a well-established tradition in economics, offering a robust framework for analyzing decision-making processes. This approach enables researchers to investigate complex phenomena that are challenging to address through standard empirical observations, such as those involving uncertain outcomes or other-regarding preferences (Kagel and Roth 2015). As one of recent applications of experiments for the areas, hardly covered by any other empirical evidence, one can mention the research on credence goods, in particular the financial advice services (Mugerman et al. 2020; de Bruin et al. 2024).

Even though our study was conducted with students as participants, it is a field experiment regarding their decision-making in a natural environment (university). The choice for the field experiment rather than a laboratory one resulted from the fact that we wanted to observe real behavior, i.e., study participants’ decisions that have real-life consequences. One of the major advantages of natural field experiments is that they guarantee incentive compatibility, as decision-makers act in their natural habitat (Harrison and List 2004). A potential downside of this approach is that it precludes the informed consent of participants, who are not supposed to know that they are a subject of any experiment. As recommended in this situation, we made sure that the experimental design causes no potential harm to participants (Phillips 2021). We will discuss this problem further in the Section 4.2.

In total, 41 first-semester finance students from Wroclaw University of Economics and Business participated in the experiment. As one of the authors was having Microeconomics classes with the students, and was their examiner, it was possible to introduce the experiment smoothly. The choice of this sample resulted from the fact that we wanted to observe the decisions made by participants who had not been familiar with the endowment effect, which is explained to students much later in the educational process.

2.2. Procedures

2.2.1. Stage 1

The experiment was conducted in three stages. First, several weeks before the final exam in the microeconomics course, students were informed that the lecturer was considering granting them “examination bonuses” but decided to conduct an anonymous questionnaire first to determine their preferences. It was revealed to the students that the exam would last 60 min, and involve 20 questions of various difficulty levels, each granting a maximum of 2 points. Therefore, the exam was for a total of 40 points, and students needed to gain at least 20 to pass it. The Microeconomics course is regarded as challenging for the students and is infamous for its relatively high failure rate. It is important to note that there was no income effect involved, as the grade on the exam is independent of any workshop activity. Moreover, the economic good being a subject of the experiment was a disposable one: it could be utilized only by its owners, disregarding the question of potential future consumption or exchanges. The examination bonuses were designed as risk-reducing instruments, increasing the students’ chances of achieving a high grade on the exam. Given the students’ strong motivation to pass, we believe that the designed goods were of remarkably high utility to the participants. There existed no substitute for the designed “examination bonuses”.

The questionnaire was started on 19 January 2023, on Microsoft Teams, which was used as the communication platform. Students were asked to anonymously reveal their preferences concerning two potential “examination bonuses”: extra points and extra time to write the exam. They were provided with the following information:

I consider granting each student participating in the Microeconomics exam an examination bonus. You will be allowed to use the bonus only one time: the first time you attempt the examination test. There are two types of examination bonuses considered, and the random mechanism will be used to allocate them to students:

• bonus A: additional 2 points added to your overall examination result,
• bonus B: additional X minutes that you can use to write the exam.

I would like to learn some of your preferences, using this anonymous questionnaire. Please answer the following question.

If you were given a choice, you would rather receive:

• additional 2 points (bonus A);
• additional 10 min of time (bonus B);
• it’s hard to say.

After making the choice, students were asked an additional question:

What is the value of X (additional number of minutes) that makes you indifferent between bonus A (2 points), and bonus B (X minutes)? Or to put it otherwise: (at least) how many minutes of additional time would you have to be granted to (slightly) prefer it over 2 additional points?

In the case of the second question, students typed in a number. The goal of this stage was to determine the objective indifference rate between the extra points and the extra minutes bonuses.

2.2.2. Stage 2

Stage 2 began one week later. Students were informed that the lecturer decided to randomly allocate two types of bonuses among them, designed on the basis of Stage 1’s results. Therefore, half of the group was informed that they were endowed with PointBonus, meaning that their examination result would be automatically augmented by two points. The other half was informed that they were endowed with TimeBonus, which meant they were allowed to take the final exam for additional time (the value determined in Stage 1). No further information was revealed to the participants.

2.2.3. Stage 3

The final stage commenced one week after Stage 2 and one week before the final exam. The time interval before Stages 2 and 3 was implemented to enable the students to form an attachment to their initial examination bonuses. On the other hand, there remained a substantial amount of time before the final exam. The lecturer communicated with the students, informing them of his decision to introduce an exchange system, allowing participants to submit an offer to trade their bonuses with others. Students were notified that an exchange would occur only if matching offers were identified in the system. The system was active for 4 days, and after its closure (before the exam) the students were informed of the completed exchanges.

2.2.4. Examination Process

The endowment effect experiment was conducted in three stages, that we have described above. But apart from studying the bias, we also wanted to analyze the efficiency of both bonuses as risk-reduction instruments, and for that it is important to clarify certain nuances of the examination process. Students were required to complete the workshops before attempting the exam. Failure to do so necessitated a retake of the workshop test and precluded participation in the first exam. Moreover, the finance students, who participated in the experiment (conditional on passing the workshops), were afforded up to three opportunities to take the exam: a zero-term exam (E0), a first-term exam (E1), and the second-term exam (E2). All exams were of the same difficulty level. The E0 was an additional examination option offered to students. Since it was not an official term, failure did not result in a negative grade being documented in the student’s academic records. As a result, E0 offered lower incentives for the participants (no risk of an F grade in records), which may have led some students to attempt the exam unprepared, relying on chance. To mitigate this effect a rule concerning the bonuses was introduced: they could be utilized just once, during the student’s first attempt at the exam. Thus, students who attended E0 unprepared risked forfeiting their bonus1.

2.3. Model and Hypotheses

The design of the experiment allowed us to form the following null hypotheses.

H1a: 

Half of the participants endowed with PointBonus make an offer to exchange them for TimeBonus.

H1b: 

Half of the participants endowed with TimeBonus make an offer to exchange them for PointBonus.

H1c: 

The proportion of participants making swap offers is the same in both bonus types’ groups.

The hypotheses H1a, H1b result from the Coase theorem, and H1c, if supported, would demonstrate that the bonuses were well calibrated in Stage 1.

Apart from testing the endowment effect hypothesis, the experiment can be utilized to study the rationality of participants’ decisions, concerning the instruments under study, with respect to their utility for participants as risk-reducing instruments and in terms of the incentives they generated. To do so, it is necessary to introduce a model.

Let $x\in \left[\mathrm{0,1}\right]$ denote the agent’s effort level and let $s\in \left[\mathrm{0,1}\right]$ be the examination score. These numbers can be interpreted in percentage terms, indicating the amount of the material that the agents mastered ($x$), and the percentage score they reached ($s$). The score depends stochastically on the effort level. In order to pass, an agent needs to reach $s\ge 0.5$. Let $F\left(s\right|x)$ denote the cumulative distribution function of the final score and let the $f\left(s\right|x)$ denote the corresponding density function. Using $P\left(s\right|x)=1-F(s\left|x\right)$ to denote the probability of reaching at least a score $s$ under effort $x$, we assume that ${\displaystyle \frac{\mathrm{d}P\left(s|x\right)}{\mathrm{d}x}}\ge 0$, i.e., an increase in the effort level increases the probability of reaching a higher outcome.

The agent fails the exam when $s<0.5$. Since it makes no difference how many points the agent failed the exam with, we assume that the agent’s benefit is 0 for any $s<0.5$. Then, the expected benefits of the decision-maker are given by:

$$B\left(x\right)={\int }_{0.5}^{1}u\left(s\right)·f\left(s|x\right)\mathrm{d}s.$$

(1)

Increasing the effort level increases the probability of getting a higher score and the expected benefits, but it also increases the costs $c\left(x\right)$ of the decision-maker. Deciding how much effort to take to prepare for the examination is a decision made under risk. Combining benefits and costs into the expected utility function of an individual we get:

$$U\left(x\right)=B\left(x\right)-c\left(x\right)={\int }_{0.5}^{1}u\left(s\right)·f\left(s|x\right)\mathrm{d}s-c\left(x\right).$$

(2)

We will use $x^{\ast }$ to denote the agent’s optimal effort level, i.e.,:

$$x^{\ast }=\arg \underset{x}{\max }U\left(x\right).$$

(3)

By applying the marginal analysis, we find that the necessary condition for expected utility maximization is the equalization of marginal benefit ($B\prime \left(x\right)$) and marginal cost ($c\prime \left(x\right)$):

$$B\prime \left(x^{\ast }\right)=c\prime \left(x^{\ast }\right).$$

(4)

We assume that $x^{\ast }$ is well defined, as the marginal benefit is decreasing, and the marginal cost is increasing with $x$. The optimal effort is individual, as it depends both on the agents’ risk attitude (reflected by function $u$), and their cost function (some agents learn fast, in the case of the others it is more time- and effort-consuming). We assume that for most agents:

$$P\left(s=1|x^{\ast }\right)<1,$$

(5)

i.e., it would not be optimal to “learn everything” and get the highest mark with p. 1. So, each participant taking the exam is facing some risks.

The examination bonuses are risk-reducing instruments. Various bonuses have different properties, though. The PointBonus increases the final score by a fixed value of 2 points, i.e., five percentage points. For any effort level, it increases the probability of reaching an aspiration level, and so is clearly a risk-reducing instrument. The expected utility of the PointBonus holders is given by (6):

$$U\left(x\right)={\int }_{0.45}^{0.95}u\left(s+0.05\right)·f\left(s|x\right)\mathrm{d}s+u\left(1\right)·P\left(0.95\right)-c\left(x\right),$$

(6)

Compared to the lack of any bonuses, this means a parallel shift of the marginal benefit curve to the left, and therefore a decrease in the agents’ optimal effort. Yet, the change in the optimal effort is less than the shift in $B\prime \left(x\right)$. We put forward the following null hypotheses:

H2a: 

Participants endowed with PointBonus exert the same effort as participants without any bonuses.

H2b: 

Participants endowed with PointBonus end up with the same score as participants without any bonuses.

We expect the data to contradict these hypotheses, as we predict the effort level of PointBonus holders to be lower, and the final score to be higher than that of the no bonus participants.

The TimeBonus increases the efficiency of the score-effort function, i.e., increases the expected value of the score for each effort level by affecting the probability function $P\left(s\right|x)$. Nevertheless, compared to PointBonus, it is harder to estimate the value of this instrument in terms of its risk-reduction properties. Since additional time is unlikely to have a negative impact, it introduces an element of ambiguity. If the exam is extremely easy, the additional time might be of little value, whereas in the case of more difficult exams, it could be of more value. We put forward the following null hypotheses:

H2c: 

Participants endowed with TimeBonus end up with the same score as participants without any bonuses.

H2d: 

Participants endowed with TimeBonus end up with the same score as participants with PointBonus.

We expect the data to contradict hypothesis H2c, as we predict that additional time will have a positive impact on the agents. Hypothesis H2d results from the rationality of the participants. As they showed no inclination to exchange the bonuses, we have no basis to assume ex ante that any of the bonuses is more efficient than the other.

3. Results

3.1. Endowment Effect

As many as 33 students filled in the survey in Stage 1. Concerning the first question, all participants stated that they would rather receive an additional two-point bonus than an additional 10-min bonus. For the second question, the median of answers was used to determine the indifference point, which turned out to be 30 min for two points.

It is worth noting that the first question may have served as an anchor for some students when answering the second one. If this effect was in place, then the indifference rate might have been underestimated, and later in the experiment the PointBonus (additional 2 points) could have been slightly preferred to the TimeBonus (additional 30 min) one.

Since the experiment involved barely any transaction costs, under the Coase theorem, it was expected that half of the students would opt for trading their bonuses.

Table 1. Participants’ willingness to exchange the endowment depending on the bonus’ type.

Variable	TimeBonus	PointBonus
Number of participants	20	21
willing to keep	15	19
willing to exchange	5	2
Expected no. of swap offers	10	10.5
Binomial test	$p=0.041$	$p=0.0002$
Chi-squared test	$p=0.188$

Note: TimeBonus denotes participants endowed with additional 30 min, and PointBonus denotes participants endowed with two additional points. The $p$ values for the binomial test show the results of the independent two-sided binomial tests, with the null hypotheses stating that the numbers of participant willing to exchange and to keep the original endowments in each bonus group are equal. The $p$ value for the Chi-squared test shows the results of the 2 × 2 test with null hypothesis stating that the proportion of exchange offers is independent of the bonus’ type.

presents the actual results2.

The hypotheses H1a and H1b stated that, in each group, we would observe an equal number of students willing to retain and exchange their endowments. Yet, the number of students making a swap offer in the case of each group was significantly lower than anticipated, which compels us to reject the null hypotheses at 5% significance level. Therefore, the findings support the endowment effect hypothesis. Although a preference for PointBonus was observed (which could be due to underestimation of the indifference point) a 2 × 2 Chi-squared test shows no basis to reject the hypothesis H1c, stating that the proportion of exchange offers is independent of the bonus’ type ($p=0.188$). We believe that the few days following the allocation of bonuses were sufficient to foster object attachment among the students. It is important to note, however, that participants were unable to gain additional information about the properties of the goods, as these were not physical objects, and their actual utility would only be revealed during or after the exam. A decision to give away the endowed bonus was always accompanied by a risk. By giving away the PointBonus students risked that they might fail the exam by a small margin. Similarly, by giving away the TimeBonus students risked that they might fail the exam because they lacked time to complete it. These potential threats outweighed the prospective benefits of acquiring a different risk-reducing instrument. Such conclusions are in line with the findings of (Nayakankuppam and Mishra 2005), which indicated that sellers tend to emphasize positive features, while buyers are more inclined to focus on negative features while evaluating the object.

Conducting a natural field experiment restricted our access to data beyond the participants’ gender3. As the sample is relatively small, one has to be very careful when drawing far-reaching conclusions based on it, yet the analysis of the endowment decisions with respect to participants’ gender shows interesting patterns.

Table 2. Participants’ willingness to exchange the endowment depending on gender.

Participants	TimeBonus	PointBonus
Females	10	9
Willing to keep	6	8
Willing to exchange	4	1
Expected no. of swap offers	5	4.5
Binomial test	$p=0.754$	$p=0.039$
Males	10	12
Willing to keep	9	11
Willing to exchange	1	1
Expected no. of swap offers	5	6
Binomial test	$p=0.022$	$p=0.006$

Note: TimeBonus denotes participants endowed with additional 30 min, and PointBonus denotes participants endowed with two additional points. The $p$ values for binomial tests show the results of the independent two-sided binomial tests, with the null hypotheses stating that the numbers of participants willing to exchange and to keep the original endowments in each bonus group equal.

shows a distinct behavior of female participants with regard to TimeBonus, as their number of swap offers is close to the one predicted by the Coase theorem. We can speculate that in the case of female respondents the risk aversion (resulting with preference for PointBonus) overcame the endowment effect. This result suggests a potential gender effect, but a bigger sample would be needed to make conclusions more robust.

The subsequent section provides a deeper analysis of the differences between both instruments to gain more insight into the rationale of the students’ decisions concerning them.

3.2. Risk Reduction and Incentives

Table 3. Participation in the exams.

Participants	E0	E1	E2
Overall	25	24	12
With bonuses	25	14	2
Passed	15	14	3
Failed	10	10	9
Success ratio	60%	58.3%	25%

Note: E0, E1, E2 denote the examination term (zero term, first term and second term, respectively).

shows the number of participating students in all three examination terms, as well as the examination results.

In what follows, we will dedicate the analyses to E0 and E1, as E2 has a low number of participants overall, as well as an extremely small number of participants endowed with any bonus.

Table 4. Results of the examination and the role of bonuses.

Statistic	E0	E1	E0 and E1
All students
Av. test score	21.16 (5.84)	19.83 (7.09)	20.79 (6.44)
Av. no. of points with bonus	22.04 (5.91)	20.42 (6.81)	21.72 (6.27)
% passed	60%	58.3%	59.2%
PointBonus students
Av. test score	21.09 (5.96)	15.86 (4.79)	19.06 (6.09)
Av. no. of points with bonus	23.09 (5.96)	17.86 (4.79)	21.06 (6.09)
% passed	72.7%	57.1%	66.7%
TimeBonus students
Av. test score	21.21 (5.74)	24.43 (6.95)	22.29 (6.36)
% passed	50%	71.4%	57.1%
No bonus students
Av. test score	-	19.4 (6.68)	19.4 (6.68)
% passed	-	50%	50%

Note: E0, E1, E2 denote the examination term (zero term, first term and second term, respectively). TimeBonus denotes participants endowed with additional 30 min, and PointBonus denotes participants endowed with two additional points. The variable av. test score shows the average test score, excluding the two-point bonus, and av. no. of points with bonus shows the average test score when these points are included. Values in brackets show standard deviations.

shows the results of the zero-term exam, first-term exam, and both terms combined, depending on the bonus type.

An initial analysis of combined E0 and E1 results suggests that PointBonus benefited the students more than TimeBonus, as the rate of success within the former (66.7%) is higher than the latter (57.1%). However, this result is partially affected by mere luck, as many students from the PointBonus group narrowly exceeded the 20-points threshold. It seems to be much more insightful to investigate the average number of examination points in both groups. The average number of points of the TimeBonus students (22.29) exceeds the PointBonus group’s score (19.06), as well as their final outcome, once we add the bonus points (21.06). But the difference is (weakly) significant only once we exclude the bonus points ($p=0.062$) and is not significant when the bonus points are included ($p=0.276$).

The analysis of the first-term results provides the greatest insights, as it allows for a comparison between the scores of the two bonus groups and the no bonus group. The TimeBonus group achieved the highest average score of 24.43 points. Following this, students without any bonuses (which were utilized during E0), attained an average score of 19.40 points. The average score in the PointBonus group is significantly lower, at 15.86, even when augmented by two additional points. When comparing the averages, by applying a one-sided T-test, we find the difference between the TimeBonus and the PointBonus groups’ scores to be significant ($p=0.015$ when excluding the bonus points, and $p=0.042$ when including the bonus points). There is also a (weakly) significant difference between the TimeBonus students’ and no bonus students’ scores ($p=0.094$). Therefore, the results provide a basis to reject H2c and H2d, demonstrating that TimeBonus inclined the participants to exert significantly higher effort than the other groups’ students. Although the average test score of PointBonus students is visibly lower than that of the no bonus students (15.86 vs. 19.4 or 17.86 vs. 19.4 when the bonus is included), these differences are statistically insignificant, and so there are no grounds to reject H2a ($p=0.125$) or H2b ($p=0.305$).

The PointBonus appears to have negatively impacted students’ exam preparation, generating disincentives for exerting effort. Conversely, additional time may have created a positive incentive, as students who were aware they would have enough time during the exam, allocated more effort toward acquiring additional knowledge, confident they would have sufficient time during the exam to demonstrate it4. Both bonuses differed significantly in terms of their risk-reduction potential. The PointBonus’ utility was relatively straightforward to estimate as additional points increased the probability of reaching the aspiration level. However, the TimeBonus utility was more challenging to assess, and it introduced some ambiguity to the decision-making problem. As the risk-reducing instruments under study have different properties, one would expect the participants to form strong preferences concerning them, which could lead to the overturn of the endowment effect. Yet, our study confirmed the robustness of the bias.

Concerning the robustness of our own research, we made all the data available in public domain, and we provided detailed information on the experiment’s protocols, facilitating the reproduction and replication of our study. We did not exclude any data from the analysis. No strong assumptions were required to apply either the binomial or the Chi-squared test, ensuring that the conclusions (hypotheses: H1a–H1c) are robust. Nevertheless, the results suggest a potential role of gender in shaping the endowment effect for risk, which could be more directly addressed in a separate study. The hypotheses H2a–H2d were tested using T-test which might be questionable under a small sample size. Therefore, we tested these hypotheses additionally with the Mann–Whitney U test. Again, we found no grounds to reject H2a ($p=0.26$), and H2b ($p=0.46$), and we did find statistical grounds to reject H2d ($p=0.05$). The only difference in conclusions concerns H2c, where we now find no grounds for rejection of null hypothesis ($p=0.14$). As there are apparent differences in the average ranks of TimeBonus and no bonus students’ scores in the Mann–Whitney test (11.14 vs. 7.5), the sample is simply too small to make this difference significant.

4. Discussion

4.1. Main Findings

While the endowment effect has already been extensively demonstrated in numerous studies, a substantial body of recent research continues to explore its nuances. This paper presents the findings of the economic experiment, designed to examine whether this bias persists when decision-makers are endowed with unique, intangible goods of high value. To achieve this, we analyzed the decisions of students preparing for their final exam, observing their behavior in a natural academic environment where their choices carried potentially significant consequences for their near-term futures.

The examination bonuses provided to the participants functioned as risk-reduction tools unavailable in any market. These endowments had high expected utility for the students, being exclusively usable during the examination process, and becoming obsolete afterward. Given the natural decision-making environment and the involvement of high-utility risk-reduction tools, we anticipated that students would make rational decisions under risk, potentially mitigating or eliminating the endowment effect. As neither endowment guaranteed a specific final payoff, the TimeBonus appeared more ambiguous, and therefore more likely to be susceptible to the endowment effect. Nevertheless, our data strongly supported the endowment bias hypothesis, with only a small proportion of students in any group submitting offers to exchange their instruments. Contrary to earlier findings, this implies that the effect remains robust even for risky instruments of high and real value. Furthermore, our results also suggest a potential gender effect, as the endowment effect for the TimeBonus was not observed among female participants.

The collected data enabled a thorough analysis of the rationality of students’ decisions and the incentives generated by the risk-reduction instruments. We found that the TimeBonus, which provided additional examination time, created positive incentives. Students endowed with this instrument achieved significantly better results than their peers. In contrast, rational students’ responses to the PointBonus (reduction in effort) called into question the effectiveness of this tool: at best, it had no effect on the overall result; at worst, its impact was negative.

4.2. Ethical Concerns and Study Limitations

Although conducting a field experiment enhances participants’ motivation and increases the study’s validity, it also entails certain trade-offs. The implied lack of informed consent from participants raises ethical concerns. To address this, it is essential to minimize potential harm, thoroughly assess the benefits and risks of the experimental design, and provide participants with a complete debriefing afterward (Ifcher and Zarghamee 2016; Phillips 2021). The experiment did not involve any manipulation that goes beyond what is understood as a natural examination process; additional points or additional time are standard academic instruments that are commonly used in such contexts. One of the co-authors obtained general written permission from the rector’s office several years prior, authorizing the application of such experiments for scientific purposes. No personal information about the students was collected, except for what was necessary to evaluate their exams, and this information was erased afterward. The intervention at the core of the experiment led to all participants achieving higher expected examination scores. Among the 41 students involved, only three submitted offers that were not realized, which may have caused minor dissatisfaction. However, all of them passed the exam in the first term and raised no complaints. At the conclusion of the experiment, all students were debriefed and informed about the experiment’s goals, outcomes, and potential financial applications. Their feedback on the experiment was positive.

Regarding the study’s limitations, concerns may arise about the sample’s numeracy and homogeneity. As the experiment was conducted in a natural university setting, participants shared similar age and educational backgrounds, which may have partially influenced the results. Although a sample size of 41 participants may appear relatively small, it is typical for one-treatment economic experiments, enabling efficient testing of the Coase theorem. However, the limited sample size prevents advanced subsample analyses, as they result in reduced statistical power and inconclusiveness in certain tests.

4.3. Practical Implications

To analyze the potential implications of our research, we begin with the economic interpretation of the instruments studied. Although these tools had specific context within the natural experiment’s environment, their properties can be speculated upon at a more general level. While both instruments had uncertain utilities, they differed in their implications for the probabilities of achieving the aspiration levels. The PointBonus guaranteed an increase in the probability of reaching a success and so could be considered a standard risk-reducing instrument. The role of TimeBonus, on the other hand, was more ambiguous. Using financial analogies, the TimeBonus can be considered an option, as agents might, but do not have to, make use of it. This instrument could have been less preferred by participants, due to a known phenomenon of ambiguity aversion (Becker and Brownson 1964). While PointBonus could be considered as a tool offering an objective and certain utility increase (increased probability of a higher outcome), the TimeBonus offered a subjective, uncertain utility increase. Moreover, while the utility of the PointBonus was fixed and could not be influenced by its owners, the value of TimeBonus was partially dependent on the owners’ actions, making its value more owner related.

When making economic decisions, risk-averse individuals are expected to prefer certain prospects and, when confronted with two risky options, to select the one perceived as less ambiguous. It can be speculated that, given a choice, most participants would have opted for the PointBonus, as suggested by some during the discussion. However, when eventually endowed with TimeBonus, they demonstrated reluctance to trade it. This finding carries significant implications for financial markets, suggesting that investors might stick to some of their investments solely due to endowment bias. Although the demand for ambiguous instruments is generally lower than for risky ones, this effect can diminish during trading, yielding to higher valuations, than those assumed by simple demand analysis.

Potential government policy recommendations depend on the recognized problem, expected outcomes, and the specific characteristics of the products in question. If the government aimed at mitigating the endowment effect in the markets of risky financial instruments, the policy should focus on stressing the negative (risky) aspects of the instruments, to make investors more inclined to trade them. However, the reduced transaction volume caused by the bias is not inherently negative, particularly in highly volatile markets. Conversely, if the objective is to amplify the endowment effect, policies should highlight the positive attributes (opportunities) associated with the instruments.

4.4. Final Conclusions and Further Research

We demonstrated that the endowment effect is a strong bias, which affects even important decisions made in the natural environment. It is not limited to standard consumption goods (like mugs or clothes) where forming an emotional connection is relatively easy, but it also extends to risk-reducing instruments: intangible goods, for which, emotional attachment is significantly more difficult to establish, suggesting a more rational approach might be expected. The design of the bonuses made it impossible for the owners to learn any new properties of the objects. Similarly, participants were unable to exhibit strategic behavior during the exchange stage, as it did not involve any pricing mechanisms. Our design excluded the possibility of any consumption or exchange in the future, and so the decision-makers had to evaluate the utility of goods they were endowed with. Almost all of them decided that what they had was better than what they did not have.

The experimental environment possessed certain specific characteristics which do not allow us to easily generalize all conclusions. As an example, consider the endowment funds. These funds are built on the concept of endowment and, similarly to the instruments studied in our paper, could be considered a risk-reducing instrument. Nevertheless, contrary to our setup, due to their nature, the endowment funds can hardly be traded or exchanged in any markets, which makes it difficult to draw any conclusions concerning them based on our study. Dedicated experimental research could assess the role of the endowment bias for the functioning of these funds and suggest some implications concerning potential government policy.

Our research implied that there might be differences in the way the risky, financial endowments are utilized and evaluated by males and females. A potential explanation lies in the well-recognized tendency for women to exhibit higher levels of risk-aversion compared to men (Croson and Gneezy 2009; Eckel and Grossman 2008). This suggests that individuals endowed with a risky instrument are subject to two opposite incentives: as the endowment effect mitigates the willingness to trade, the risk aversion reinforces it. A dedicated study with a larger sample size and a design focused on risk attitudes and gender differences could provide deeper insights into this intriguing problem.