Ambiguity aversion: bibliometric analysis and literature review of the last 60 years

The number of publications on ambiguity remained relatively low until the beginning of the 2000s (see Fig. 3). After that, it rose sharply until the end of our observation period. The highest number of annual publications is recorded in 2018 with 43. In the years 1951–1999, an average of 0.76 papers are published. In the years 2000–2020, it is 20.2. This dynamic is in line with the development of behavioral economics analyzed by Costa et al. (2019) showing an exponential increase of publications between 2000 and 2015. Using the search terms “behavioral economics, “behavioral finance”, and “behavioral accounting”, Costa et al. (2019) observe 1–3 publications per year between 1967 and 1990, less than 30 publications per year between 1991 and 2001, over 100 in 2008, 250 in 2012, and 346 publications in 2015. Similarly, in the behavioral finance literature, Pleßner (2017) identifies two papers on the disposition effect in 2000 but 26 in 2014 and Jain et al. (2021) merely count one paper on behavioral biases in 1995 but 28 in 2019. In our data, the increasing trend of publications between 2000 and 2015 continues in the years 2016 and 2018. However, in 2017, the number of publications on ambiguity aversion decreases compared to the previous year. This could be interpreted as the beginning of a period of stagnation in the number of publications. The lower publication numbers in 2019 and 2020 compared to 2018 support this interpretation. Figure 3 also shows the distribution of our five clusters over time. In nearly every year, most of the papers are model applications (Gao and Driouchi 2018; Lo 1998; Turocy 2008). This does not correspond to the review by Goyal and Kumar (2021), who observe 86% empirical studies, 10% conceptual studies, 3% reviews, and 1% meta-analyses on financial literacy from 2000 to 2019. But also in the field of ambiguity aversion, empirical and experimental studies have been gaining ground from the 2000s onwards (Dimmock et al. 2016b, 2016a; Koudstaal et al. 2016; Muthukrishnan et al. 2009; Sutter et al. 2013). The low proportion of qualitative studies is striking. The proportion of comments is also low.

Table 1 analyzes the average number of authors per publication. In the early development of the research field – from 1961 until 1990 – more than 40% of the papers were single-authored (1.71 authors on average), in the recent development – from 2011 until 2020 – around 20% (2.35 authors on average). This is in line with the development of co-authorship in the literature of economics in general. Analyzing the RePEc archive, e.g., Rath and Wohlrabe (2016) observe an increase from 1.56 authors per paper in 1991 to 2.23 authors in 2013. This reflects the general trend and increased importance of intra- and interdisciplinary collaboration in behavioral economics among theorists, experimentalists, and empirics.

In addition to the number of publications, we examine the journal rankings of the papers based on JOURQUAL3 (Henning-Thurau et al. 2004). The ranking assesses the quality of the journals from the best category “A + ” to the worst “D”. The overall ranking includes several sub-rankings, e.g., “General Business Studies” or “Banking/Finance”. It should be noted that a large number of publications (321 of 456) are not part of the ranking. They are either listed in other rankings (e.g., SJR), or they do not have any ranking. These 321 publications are published in 125 different journals and 10 discussion paper series. The five journals in which these papers are published most frequently (Journal of Risk and Uncertainty, Journal of Economic Theory, Journal of Economic Behavior and Organization, Journal of Mathematical Economics; 21.8 papers on average per journal) have an average SJR ranking of 0.129. The 20 journals with the most frequent publications on ambiguity aversion outside JOURQUAL (8.4 papers on average per journal) have an average ranking of 0.148.

Most of the 135 papers analyzed in Fig. 4 are published in journals ranked “A + ” or “A” in JOURQUAL. For decision-theoretical models and comments, the proportion of publications rated “C” or worse is zero. 74.8% of the experimental studies, empirical studies, and model applications (without data) are published in journals rated “A + ”, “A”, or “B”. The share of model applications published in A + journals (34.2%) is lower than that of experimental studies (41.4%).

According to Google Scholar,¹ the clusters also differ in their citation frequency: The average number of citations in the clusters decision-theoretic model (113.5) and experimental study (100.0) is much higher than in model applications (without data) (59.5) and comments (8.1). Empirical studies are cited 82.5 times on average. Web of Science also reports the highest citation numbers for decision-theoretic models (47.1) and experimental studies (43.6), followed by empirical studies (26.5), model applications (without data) (22.2), and comments (2.4). The Business Source Premier database (via EBSCOhost) indicates a lower number of citations per cluster. According to this database, the citation frequencies of decision-theoretic models (15.3) and experimental studies (11.2) are higher than that of model applications (without data) (5.3) and comments (0.7). Empirical studies are cited nearly as much as experimental studies, 10.4 times on average.

Figure 5 shows the JOURQUAL journal rankings of papers on ambiguity aversion over time. The absolute scale of the ordinal axis refers to the number of publications in the respective year, while the relative scale shows the proportion of A + publications. Considering the whole period – including years in which there are no publications in ranked journals – the average proportion of A + papers is 23.9%. Excluding years with no ranked publications, the average share of A + journals is 52.6%. Thus, ambiguity aversion is very relevant and of general interest. The share of top publications decreases after 2006 because of the increasing number of publications on ambiguity aversion. The absolute number of A + publications is highest in 2017 with seven papers.

Citation frequencies in Google Scholar also differ between journal rankings: While publications in A + -ranked journals are cited 229.9 times on average, papers are cited 135.2 times on average if they are published in an A-ranked journal. In B-ranked journals, they are cited 17.2 times, and in C-ranked ones 6.5 times. Not-ranked journals have a citation rate of 35.1 on average. Web of Science reports lower levels of citation frequencies: A + papers are cited 85.9 times, A papers 53.6 times, B publications 6.8 times, and C publications 2.8 times on average. Not-ranked journals have an average citation frequency of 15.5 in Web of Science. Using the Business Source Premier database, EBSCOhost finds 26.9 citations for papers in journals ranked A + and 16.4 in those ranked A. B-ranked articles are cited 1.5 times and C-ranked ones 1.2 times. Papers that are not ranked in JOURQUAL are cited 3.9 times on average according to EBSCOhost.

Table 2 shows that the most prominent publication in the field of ambiguity aversion is the paper that started this line of research. According to Google Scholar, Ellsberg (1961) has already been cited by 9411 papers, EBSCOhost finds 831 citations from the Business Source Premier database. Moreover, the theoretical models of Gilboa and Schmeidler (1989), Schmeidler (1989), and Klibanoff et al. (2005) (see Sects. 3.2‐3.4) are cited very frequently. Heath and Tversky (1991), Judge et al. (1999), and Chen and Epstein (2002) are connected to these models. Table 2 and Fig. 3 reveal that theoretical papers set the basis of the research field of ambiguity aversion and are most cited up to now. Since 2000, a lot of experiments and empirical papers have been testing these models (Eichberger et al. 2012; Halevy 2007; Hey et al. 2010, see also Sect. 4). At least by now, however, these publications have been cited less often. Table 3 complements this finding by showing the five most-cited authors in the field. Besides the top-cited author Ellsberg, some prominent theorists co-author many theoretical applications or experimental tests of their models.

Furthermore, we conduct a co-occurrence analysis of the papers’ keywords. Figure 6 presents the mapping of the most frequently used keywords. We set the minimum threshold for an appearance on the map at 12. This criterion is met by 77 keywords. The size of the nodes in the figure indicates the frequency of the keywords. We see at least three topic areas. The left one shows an application-related focus on financial risk management, portfolio decision problems, and investments in general. Especially the literature we discuss in Sects. 3.5 and 4.4 is represented in this field, e.g., Anderson (2019; keyword: economic models), Vardas and Xepapadeas (2015; keyword: expected utility), or Dicks and Fulghieri (2019; keyword: risk aversion). The area on the right shows a decision-oriented focus (ambiguity tolerance, choice behavior, risk-taking). The literature located in this topic area can be found in particular in Sects. 3.2, 3.3, and 3.4 (e.g., Trojani and Vanini 2004 with the keyword decision making). The category in the middle bottom of Fig. 6 represents behavioral and experimental economics in general (e.g., Peysakhovich and Naecker 2017 use the keyword behavioral economics, see Sect. 6.1). The keyword Ellsberg paradox (in the middle on top) combines these three categories. It is for example used by Klibanoff et al. (2005), who present a new decision-theoretic model of ambiguity aversion. The most used keywords and the keywords with the most connections to others are (of course) ambiguity aversion, decision making, uncertainty, aversion, risk aversion, and probability theory. We can see from the interconnection of the categories that the keywords are centered around decision theory (building or applying models to describe human decision processes under uncertainty and ambiguity theoretically) and behavioral applications (theoretical or experimental and empirical analyses of how ambiguity aversion affects economic decision making). In Appendix A–E, we split the data set into five periods showing the development of the keywords’ co-occurrence. This analysis indicates a trend of the research field from theoretical contributions to experimental and empirical applications.

Our bibliometric analysis shows a significant rise in the number of publications during the last six decades, especially a boom of papers from the 2010s onwards. Simultaneously, the average number of authors per publication rises, which speaks for the increasing importance of collaboration. The development of the research field is in line with the results of other bibliometric research in economics and especially behavioral economics (Rath and Wohlrabe 2016; Pleßner 2017; Costa et al. 2019; Jain et al. 2021). The topics of the publications in the field of ambiguity aversion cluster around decision theory (Schmeidler 1989; Klibanoff et al. 2005; Etner et al. 2012) and behavioral applications (Dimmock et al. 2016a; Stahl 2014; Ellsberg 1961). Most of the papers are either theoretical model applications or empirical and experimental studies that test the theoretical propositions. Few publications are dedicated to the development of theoretical models. We observe a high but decreasing proportion of A + and A publications and an increase of publications with a lower ranking. Contributions on new theoretical models have the highest share of A + publications. Ellsberg’s (1961) seminal paper on his paradox and further behavioral theories build the basis of research on ambiguity aversion and have been cited most up to now, followed by model applications (Gao and Driouchi 2018), experimental papers (Stahl 2014), and empirical research (Anderson 2019). In recent years, the majority of papers have been data-driven (Altug et al. 2020; Anderson 2019; Brenner and Izhakian 2018).

Researchers may find our bibliometric analysis useful to get an overview of the important developments and trends in this stream of literature and to discover the potential for future research publishable in high-quality journals. Nevertheless, our approach of extracting and screening the papers for our literature review (see Fig. 1) might lead to biases, as we had to manually add literature to the bibliometric analysis based on excerpts. In the next two sections, we review the basic theories on ambiguity aversion as well as their theoretical applications (Sect. 3) and present experimental and empirical evidence based on these models (Sect. 4).

In Ellsberg‘s (1961) seminal paper, he questiones the Sure Thing Principle of Savage‘s (1951) Subjective Expected Utility (SEU) theory with a thought experiment. SEU assumes individuals to assess subjective probabilities to alternatives of decision problems under uncertainty if these cannot be objectively quantified (see the Expected Utility theory of von Neumann and Morgenstern 1944). Ellsberg (1961) presents two experimental designs. The first design comprises two urns with 100 balls. Urn A contains 50 white and 50 black balls, urn B has an unknown distribution of white and black balls. An ambiguity-averse decision-maker prefers the known probability in urn A. The preference for urn A, no matter if the decision-maker is paid when (1) a black or (2) a white ball is drawn, leads to a contradiction. The first decision indicates the belief that urn B contains fewer black balls than urn A, the second decision implies the belief that urn B contains more black balls than urn A.

The second experimental design is based on an urn with 90 balls – 30 red balls and 60 blue or yellow balls. In the first decision, the decision-maker receives a payoff if the color is drawn on which he or she betted (either red or blue), in the second decision the decision-maker additionally receives a payoff if a yellow ball is drawn. Table 4 provides an example.

An ambiguity-averse decision-maker prefers action \(f\) over \(g\) (thus he or she would win $100 if a red ball is drawn) and action \(g^{\prime}\) over \(f^{\prime}\) (thus winning $100 if either a blue or a yellow ball is drawn). Again, this leads to the contradiction that the decision-maker seems to believe that the urn contains less than 30 blue balls when deciding for \(f\) but more than 30 blue balls when deciding for \(g^{\prime}\):

f ≿ g implies:

f' ≿ g' implies:

After Ellsberg (1961) introduced the experiment, many applications and extensions of the design were published (Di Mauro and Maffioletti 2004; Du and Budescu 2005; Halevy 2007; Moore and Eckel 2006; Yates and Zukowski 1976, see also Sect. 4).

The Choquet Expected Utility (CEU) model developed by Schmeidler (1989) is based on capacity \(v\), which reflects the degree of uncertainty experienced by the decision-maker and his or her attitude towards uncertainty (Agliardi et al. 2016). The capacity represents a set of non-additive probability distributions, and the utility is calculated with the Choquet integral (Etner et al. 2012). Schmeidler (1989) describes the decision criterion as follows:

If \(S = \left\{ {s_{1} ,s_{2} , \ldots ,s_{n} } \right\}\) and \(f\left( {s_{i} } \right) = x_{i} ,i = 1, \ldots ,n\) with \(x_{i} \le x_{i + 1}\),

then

\(\begin{aligned} \mathop \int \nolimits_{Ch}^{{}} u\left( f \right)dv = & \, u\left( {x_{1} } \right) + \left( {u\left( {x_{2} } \right) - u\left( {x_{1} } \right)} \right)v\left( {\left\{ {s_{2} ,s_{3} , \ldots ,s_{n} } \right\}} \right) + \cdots \\ & \quad + \left( {u\left( {x_{i} + 1} \right) - u\left( {x_{i} } \right)} \right)v\left( {\left\{ {s_{i + 1} , \ldots ,s_{n} } \right\}} \right) + \cdots + \left( {u\left( {x_{n} } \right) - u\left( {x_{n - 1} } \right)} \right)v\left( {\left\{ {s_{n} } \right\}} \right). \\ \end{aligned}\)

Thus, a decision-maker first considers the worst outcome of an alternative and continues with better ones until the whole outcome space is evaluated. In contrast to Savage's SEU, different utility functions can be applied for different actions (Agliardi et al. 2016). Furthermore, the non-additivity of CEU can be used to model ambiguity aversion. An individual is ambiguity-averse if his or her capacity \(v\) is convex and the utility function concave or linear (Schmeidler 1989).

While Savage's SEU assumes that an individual prefers a single alternative, the Maxmin Expected Utility (MEU) model assumes that several alternatives can be preferred simultaneously. In the model, the decision-maker chooses the option that promises the maximum of the minimal expected utilities (Gilboa and Schmeidler 1989). Formally, an action \(f\) is preferred to an action \(g\) if and only if (Etner et al. 2012).

Hence, MEU can model ambiguity aversion. An action \(f\) in a state \(s\) leads to a prize \(x\) with a probability distribution of \(f\left( s \right)\). Imagine an experiment in which the decision-maker is allowed to choose between the draw from urn A or B (containing black and white balls each), urn B representing ambiguity. Formally, the process can be described by \(\alpha f + \left( {1 - \alpha } \right)g\) with \(\alpha \in \left[ {0,1} \right]\). If state \(s\) occurs, then the prize the decision-maker wins (outcome) depends on the lottery \(\alpha f\left( s \right) + \left( {1 - \alpha } \right)g\left( s \right)\) (Segal 1990). \(\alpha\) represents the probability of choosing action \(f\), \(1 - \alpha\) the probability of choosing action \(g\), and \(f\left( s \right)\) the probability distribution of the lottery. According to MEU, \(\alpha f\left( s \right) + \left( {1 - \alpha } \right)g\left( s \right){ \succsim }f\) applies (Gilboa and Schmeidler 1989). As an example,\(f\) can denote the bet on a white ball in urn B, while \(0.5f\left( s \right) + 0.5g\left( s \right)\) represents the bet on white in urn A (\(g\) is the draw of a black ball). Under ambiguity, the decision-maker tries to find the option that maximizes his or her utility of the worst outcome (Gilboa and Schmeidler 1989), which means minimizing the possibility of drawing a black ball from one of the urns in our example. This leads to an aversion to non-quantifiable probabilities and a preference for urn A.

Several studies refer to MEU (Kochov 2015; Maccheroni et al. 2006; Gilboa and Marinacci 2016), apply it (Bidder and Dew-Becker 2016; Zheng et al. 2015; Trojani and Vanini 2004), or extend it (Ghirardato et al. 2004; Li et al. 2016; Hansen and Sargent 2001; Chen and Epstein 2002; Miao and Wang 2011).

The two models described so far are based on a two-stage decision-making process. Before betting on an action, the decision-maker mentally estimates the probabilities of the actions. The Smooth Ambiguity Model (SAM) does not need the mental reduction of two-stage lotteries (Lang 2017).

Segal (1987, 1990) proposes two axioms, the Reduction of Compound Lotteries Axiom and the Compound Independence Axiom, which provide a new explanation for the Ellsberg paradox. Klibanoff et al. (2005) develop these further and designed the SAM. In this model, the decision criterion is based on a real-value function on a state space \(S\). The decision-maker’s utility function \(u\) follows von Neumann and Morgenstern (1944) and includes the risk attitude. The decision-maker’s attitude towards ambiguity is captured by \(\phi\). \(\mu\) is the subjective prior over \(\Delta\), the set of possible probabilities over the state space, and \(\pi\) a probability measure on the state space. The criterion results in

A concave function of \(\phi\) defines ambiguity aversion, while a convex function represents ambiguity seeking. When \(\phi\) is linear, the decision-maker is ambiguity neutral and acts as an SEU maximizer. Assuming ambiguity aversion, Klibanoff et al. (2005) give the following example of the decision maker’s thinking:”My best guess of the chance that the return distribution [of an investment decision] is ‘\(\pi\)’ is 20%. However, this is based on ‘softer’ information than knowing that the chance of a particular outcome in an objective lottery is 20%. Hence, I would like to behave with more caution with respect to the former risk.”

After the introduction of SAM, many other papers followed the approach (Battigalli et al. 2015; Battigalli et al. 2016; Ju and Miao 2012; Maccheroni et al. 2013; Strzalecki 2013; Thimme and Völkert 2015; van de Kuilen and Wakker 2011; Wong 2015).

Models on ambiguity aversion have been used in various theoretical applications. Gao and Driouchi (2018), for instance, apply CEU to model the influence of ambiguity aversion on outsourcing decisions. Some papers study the role of ambiguity aversion in the design of auctions. Salo and Weber’s (1995) decision-makers, e.g., are characterized by CEU, Lo (1998) and Turocy (2008) use SEU with multiple priors as a framework, and the recent work by Koçyiğit et al. (2020) is based on MEU.

Analyzing investment decisions, Dow and Werlang (1992) use the CEU model for portfolio choices. Likewise, Berger et al. (2013) model the propensity of investments and portfolio compositions considering ambiguity aversion and learning. Escobar et al. (2015) investigate portfolio management for financial derivatives modeling ambiguity-averse investors. Using a similar approach, Bergen et al. (2018) study portfolios composed of derivatives and equities. Furthermore, Vardas and Xepapadeas (2015) analyze ambiguity aversion in portfolio selection using the Robust Portfolio Choice Theory.

Chateauneuf et al. (2000) apply the CEU model to the analysis of market equilibria and identify the decision-maker's tendency to diversify risk. Rigotti and Shannon (2005) use CEU to study factors influencing ambiguity aversion in the market with a model developed by Bewley (2002). Mukerji and Tallon (2001) attribute the imperfection of financial markets to ambiguity aversion utilizing the CEU model (see Rinaldi 2009 for a similar analysis with SAM). Chau and Vayanos (2008) analyze a market model in which some individuals have insider knowledge. Vitale (2018) extend this approach and apply it specifically to the insurance market. Also, Epstein and Wang (1994), Chen and Epstein (2002), and Liu (2011) study the impact of ambiguity aversion on market equilibria.

Epstein and Schneider (2008) model the valuation of assets for which decision-makers have access to information of different quality – assuming that low-quality information is associated with a high degree of uncertainty. Leippold et al. (2008) follow the same approach and extend the assumptions of Epstein and Schneider (2008) by learning of the decision-maker. The theoretical analysis of Dicks and Fulghieri (2019) implies that ambiguity aversion accelerated the financial crises. However, Condie (2008) argues that ambiguity aversion has very limited explanatory power for the long-term price development of assets.

Halevy (2007) analyzes how far his experimental data can be explained by SEU, MEU, Recursive Non-Expected Utility, or Recursive Expected Utility. None of these models universally represents all the students' preferences. Similarly, Eichberger et al. (2012) do not find evidence for ambiguity-aversion when not only probabilities but also the stake sizes are ambiguous. However, Ahn et al. (2014) reveal that most of their subjects express SEU preferences and ambiguity aversion in line with MEU and CEU. Hey et al. (2010) are better able to explain behavior in their experiment with relatively simple models like the MEU compared to more sophisticated models like the CEU.

Stahl (2014) examines the preferences of decision-makers in a two-urn design. The subjects can decide to bet on a risky urn with a 50% probability to win $10 or on an ambiguous urn leading to a win of $10, $12, or $15. Most of the subjects prefer the risky over the $ 10 ambiguous urn. A large number of subjects are indifferent between the risky and the $12 ambiguous urn, and the majority prefer the $15 ambiguous over the risky urn. Stahl (2014) categorizes the subject pool into three groups: (1) subjects who behave according to SEU (12%), (2) subjects who behave according to MEU (26%), and (3) subjects whose decisions cannot be assigned to a theoretical model (60%). In line with his findings, most of the subjects in Charness et al. (2013) cannot be classified as ambiguity-averse.

Curley and Yates (1989) let subjects choose between a risky bet (with a 25% probability of winning) and a draw from an urn representing ambiguity with 5 winning, 55 losing, and 40 unknown balls. In contrast to the Ellsberg paradox, most of the subjects prefer the ambiguous urn, either because of the rather low winning probability in the risky bet or because of optimism (see also Dimmock et al. 2013, 2016b; Kahn and Sarin 1988).

Cohen et al. (1987) compare decisions on potential gains with those on potential losses. Similar to Kahnemann and Tversky (1979), the authors observe that the majority of decision-makers are ambiguity-averse in the domain of gains but indifferent in the domain of losses. Likewise, Friedl et al. (2014) assess the willingness to pay (WTP) for insurance policies and observe no differences in the preferences for risky or ambiguous options in the domain of losses. For comparable results with low winning probabilities (in the range from 0.1% to 30%) of the risky alternative in the domain of losses see Curley and Yates (1985), Einhorn and Hogarth (1986), Di Mauro and Maffioletti (1996), Lara Resende and Wu (2010) and Tymula et al. (2012). With moderate (30% to 60%) winning probabilities, subjects tend to decide in favor of the ambiguous option (Baillon et al. 2018b, 2018a; Chakravarty and Roy 2009; Ho et al. 2002; Liu and Onculer 2017). In line with the Ellsberg paradox, higher winning probabilities lead to ambiguity aversion (Abdellaoui et al. 2011).

The ambiguity premium is comparable to the risk premium and reflects the difference between the WTPs of the risky and ambiguous option. Grou and Tabak (2008) observe that business and economics students in Brazil (in contrast to students from the University of Chicago) do not want to pay a premium to reduce ambiguity. Trautmann and van de Kuilen (2015) review several studies on ambiguity premiums and suppose that the magnitude of the premiums depends on the valuation method, the stake size, and the incentive method. They leave a systematic analysis of the heterogeneity of ambiguity premiums to future research.

Analyzing the Standard & Poor’s (S&P) 500 index, Brenner and Izhakian (2018) find evidence for an ambiguity premium in the stock market. They assume that the equity premium in asset pricing theory contains both a risk and an ambiguity premium. Jeong et al. (2015) develop a method for separately measuring premiums for risks and ambiguity. They apply this method to the S&P 500 index and confirm that investors pay an ambiguity premium. Although there have been publications on ambiguity premiums, research on their causes and the factors that may increase or decrease these premiums is still pending.

Sarin and Weber (1993) compare the behavior of students with bank executives' choices in sealed-bid auctions and double oral auctions. They find evidence for ambiguity aversion in both samples. Koudstaal et al. (2016) observe that entrepreneurs express a similar level of ambiguity aversion compared to employees and managers. Furthermore, Sutter et al. (2013) show that even children between 10 and 18 years are ambiguity-averse.

In a lab in the field experiment in Ethiopia, Akay et al. (2012) find that farmers do not differ in their level of ambiguity aversion from Dutch students in a classroom experiment. Likewise, Engle-Warnick et al. (2007) investigate the decision-making behavior of Peruvian farmers about the use of new cultivation technologies. Combining survey data with a lab in the field experiment, they observe a positive correlation between ambiguity aversion and conservative choices concerning new technologies. Similarly, Ross et al. (2012) find that ambiguity-averse farmers from Laos have a lower propensity to use new rice varieties.

Dimmock et al. (2016a, 2016b) confirm ambiguity aversion for US and Dutch households using online experiments implemented in representative surveys. Dimmock et al. (2016b) conduct five experiments with subjects of the “RAND American Life Panel” on household portfolio choice puzzles. They show a negative association of ambiguity aversion with stock market participation, the fraction of financial assets, and foreign stock ownership. Wakker et al. (2007) experimentally analyze the WTP for insurances with a representative sample from the general Dutch public. Using in-depth individual interviews to gain more information about the decision-maker's choices, they find evidence for ambiguity seeking rather than aversion.

Muthukrishnan et al. (2009) analyze the tendency of customers to prefer well-known over unknown brands. They observe that ambiguity-averse subjects prefer established brands, even if the product specifications are inferior compared to the unknown brands. Liu and Colman (2009) examine the long-term change in preferences for marketing strategies with a repeated experiment. They find that the intensity of ambiguity aversion decreases with an increasing number of repetitions. The authors assess decisions on marketing strategies as a control for classical Ellsberg urn decisions.

Moreover, Berger et al. (2013) investigate patients' decisions regarding vaccination against swine flu. They find that medical advice intended to support patients' decisions does not consider their ambiguity aversion. Courbage and Peter (2021) extend this study by examining the influence of ambiguity on personal vaccination decisions. Hoy et al. (2014) show that patients are reluctant to use free genetic tests due to ambiguity aversion. Segal and Stein (2006) find evidence that ambiguity-averse defendants in court tend to prefer bench trials over jury trials whenever their acquittal chances are substantial. Otherwise, they prefer jury trials. Furthermore, Ryall and Sampson (2017) see that contract partners tend to assure themselves against ambiguous actions in joint contracts.

Analyzing data of mutual fund investors, Li et al. (2017) argue that investors seem to place greater weight on the worst signal when confronted with information of ambiguous quality. Breuer et al. (2016) analyze the level of capital reserves held by companies. They find that managers reduce capital reserves when their investors are more ambiguity-averse. Similarly, Antoniou et al. (2015) observe that an increase of ambiguity in the stock market yields outflows from equity funds. Finally, Anderson (2019) examines the behavior of investors during the financial crisis in 2008. She finds that market outcomes can change more abruptly under ambiguity than under risk.