The phenomenon of Anchoring bias refers to the influence of arbitrary numbers in decision-making under uncertainty. Humans are affected by anchors on a daily basis, especially when confronted with quantitative tasks. However, basic anchoring effects appear even when individuals are not expected to compare the value to a certain estimation task. Many researchers tried to figure out the reasons for the anchoring bias. Kahnemann et alii conducted three studies and concluded anchoring bias as the disability of adjustment processes. In an experimental setting the anchoring effects are examined by first showing the participant an arbitrary number, then comparing it to a certain targeted value and finally giving an own estimation. The anchoring bias appears in the last step as the estimation is inevitably biased toward the initial anchor. An alternative explanation is found in the studies conducted by Mussweiler et al. They suggest that the anchoring effect is rather a combination of insufficient adjustment and selective accessibility.
This paper aims to combine both explanations and test four hypotheses in an experimental setting related to financial markets. The first assumption to be tested is whether a comparative task yields to higher anchoring bias when an anchor is provided or self-generated. The second hypothesis to be proved is whether the anchoring bias can mitigate by giving explanations on the comparative task answers. Thirdly, the paper assumes a positive correlation between motivation and cognitive capacity influence the estimates significantly. Finally, this paper extents the research by asking whether risk-aversion is correlated to the anchoring bias.
Table of Contents
1 Introduction
2 Background and Related Literature
2.1 Anchoring bias in Financial Markets
3 Hypotheses Development
4 Data and Research Design
4.1 Experimental setup
4.1.1 Data analysis and theoretical results
5 General Discussion and model limitations
5.1.1 Incentivizing
5.1.2 Choice of non-professionals
5.2 Conversational inferences
5.3 Research Design and Forecast Errors
6 Conclusion and Outlook
7 Appendix
8 Publication bibliography
1 Introduction
The phenomenon of Anchoring bias refers to the influence of arbitray numbers in decision making under uncertainty. Humans are affected by anchors on a daily basis especially when confronted with quantitative tasks. However basic anchoring effects appear even when individuals are not expected to compare the value to a certain estimation task (Wilson et al. 1996). Many researchers tried to figure out the reasons for the anchoring bias. Kahnemann et al. (1974, 1995, 1996) conducted three studies and concluded anchoring bias as the disability of adjustment processes. In an experimental setting the anchoring effects are examined by first showing the participant an arbirary number, then comparing it to a certain targeted value and finally giving an own estimation. The anchoring bias appears in the last step as the estimation is inevitably biased toward the initial anchor. An alternative explanation is found in the studies conducted by Mussweiler et al. (2000, 2001). They suggest that the anchoring effect is rather a combination of insufficient adjustment and selective accessibility.
This paper aims to combine both of these explanations and test four hypotheses in an experimental setting related to financial markets. The first assumption to be tested is whether a comparative task yields to higher anchoring bias when an anchor is provided or self-generated. The second hypothesis to be proved is whether the anchoring bias can mitigate by giving explanations on the comparative task answers. Thirdly, the paper assumes a positive correlation between motivation and cognitive capacity influence the estimates significantly. Finally, this paper extents the research by asking whether risk-aversion is correlated to the anchoring bias.
This paper is organised in the following way. The paper first gives a deeper understanding of the background of anchoring effects and summerizes the related literature. Chapter 3 replicates the four research hypotheses and its underlying motivation. Section 4 begins with the research design that is suggested to answer the hypotheses and explains the data acquisition. Chapter 5 critically assesses the research design with its limitations and supplies possible extensions. Finally, chapter 6 briefly concludes and gives and outlook.
2 Background and Related Literature
In the field of Behavioral Economics, heuristics play a substantial role in daily business. Anchoring effects have first been studied by Tversky and Kahnemann (1974). The paper of Kahnemann and Tversky (1974) serves as the cornerstone for the related literature conducted so far. That is why this seminar thesis builds on it as well.
In the research paper of Tversky and Kahnemann (1974), subjects first faced a random number by spinning a wheel of chance from 0 to 100. This figure is called the anchor. Secondly, the participants had to decide whether the percentage of African countries in the United Nations was higher or lower than their assigned number (Tverksy and Kahnemann 1974). Lastly, probands had to estimate the true percentage of African countries in the UN and adjust from the individual starting point (Tversky and Kahnemann 1974). Although the anchor is not related to the question the authors state a significant bias towards the anchor. In addition, the paper proves that monetary incentives do not affect the results. Finally, the authors state that the anchoring bias occurs not only to uninformed subjects regarding the estimation task (Tversky and Kahnemann 1974). This paper first confirmed the phenomenon of anchors and explained the bias as an insufficient adjustment process.
The research followed on Tversky and Kahnemann’s (1974) paper is very widespread, and several fields of science noticed a similar phenomenon. Anchoring especially appears in negotiation processes as profit schedules (Ritov 1996) and starting points i.e. first offers (Kristensen and Gärling 1997) are of high importance. More recent studies collected evidence for the anchoring bias in pay-what-you-want settings. Subjects are significantly influenced in their payments by task-related as well as context irrelevant anchors (Jung et al. 2016). In addition, humans are affected in their payments by internal as well as external price references that serve as anchors (Roy et al. 2021). In all these studies the anchoring bias is inevitable and very robust. The hypothesis that anchoring effects are correlated to lower cognitive abilities like this is the case with conjunction fallacy and risk- or time preferences were rejected (Oechssler et al. 2009).
As mentioned in the last paragraph, numerous studies proved the existence of anchoring effects across several fields of research. One of the most convincing pieces of proof for the existence of the anchoring bias among these studies was conducted by Wilson et al. (1996) who warned the subjects of the anchoring bias before the initial estimation task. The probands had to estimate the number of physicians in the phone book (Wilson et al. 1996). Next, participants got divided into a control group (no anchor), the typical comparison group (receiving an anchor and comparing it to the targeted value), and forewarned groups (Wilson et al. 1996). A forewarned subject got e.g. the information that the anchor underestimates or overestimates the outcomes (Wilson et al. 1996). Although the subjects indicated that they were not influenced by the anchor, the adjustment process failed and resulted in similar values as the comparison group (Wilson et al. 1996).
As mentioned above Kahnemann and Tversky (1974) explained anchoring effects happening from judgments under uncertainty with the disability of sufficient adjustments. In an extended version of the first experiment, Jacowitz and Kahnemann (1995) had two groups of participants. The calibration group guessed 15 open questions (Jacowitz and Kahnemann 1995) e.g. what is the population of Chicago and provided anchors for the second group. The experimenter generated low and high anchors depending on the distribution of the calibration group estimate (15th and 85th percentile). Subjects with the anchoring condition received a low or high anchor and first answered whether the targeted value (e.g. population of Chicago) is higher or lower than the anchor (Jacowitz and Kahnemann 1995). Then the participants were asked to give their measures. This study used the Anchoring (AI) Index to quantitatively evaluate the bias. The AI is the difference of the median estimates of the high and low anchored groups, divided by the difference in the anchors. In most cases, the AI ranges from 0 (no anchoring effect) to 1, but greater values are possible (Jacowitz and Kahnemann 1995).1
The authors found that the transformed estimates were biased towards the anchor with a greater effect in the high anchored group. This is explained by the fact that the range of figures is limited to the lower boundary (zero) but not to the upper boundary (Jacowitz and Kahnemann 1995). Consistent with that fact, the mean AI was 0.51 and 0.40 for the high anchor and low anchor respectively (Jacowitz and Kahnemann 1995). The authors conclude with two additional observations: 1. the comparative task could be a non-quantitative anchor itself as subjects assume the value as a hint from the experimenter and 2. higher stated confidence of the estimate is not significantly reducing the bias (Jacowitz and Kahnemann 1995).
To better understand a possibly higher anchoring effect in the comparative task, Green, Jacowitz et al. (1998) compared yes/no questions to open-ended questions in a referendum contingent valuation framework. The authors found that subjects need more cognitive attention in open-ended questions than in yes/no questions (Green et al. 1998). When a task requires high attention subjects need more time solving it and this could explain the absolute value of the anchoring effect. Mussweiler and Strack (Mussweiler and Strack 2001) observed whether the absolute time a subject is confronted with an anchor is an exacerbating factor. Mussweiler and Strack (2001) suggest that the longer it takes to figure out whether the possibility of the anchor to be true is high or low the more the estimate will be biased towards the anchor.
Mussweiler et al. (2000) run two experiments to reduce the anchoring by giving subjects anchor-related information and letting them justify their response to the yes/no question. In this real-world setting, the experimenter asked specialists (mechanics, car salesmen) to estimate the value of a used car (Mussweiler et al. 2000). Subjects were provided with the necessary information to fulfil the task (e.g., kilometre indication and age). Between the comparative and the estimation task, half of the interviewees had to explain why the initial anchor was too high or too low (Mussweiler et al. 2000). An external expert estimated a buying (from a car dealer) and selling (to the car dealer) price for the used car that served for the anchor values (Mussweiler et al. 2000). The results exhibit that the range of final estimates is tighter in the argument group than in the control group and strengthens the researcher’s hypothesis. Table 1 contains the results made by Mussweiler et al. (2000).
[...]
1 The Appendix supplies a table of the results conducted by Jacowitz and Kahnemann (1995)