Bachelor Thesis
Name: Jing Wang
Student number: 10315683
Specialization: Economics
Field: Behavioral Economics
Number of credits thesis: 12
Title of thesis:
The use of Heuristics for forecasts of decisions under risk and
ambiguity
Name of supervisor: Jindi Zheng
Statement of Originality
This document is written by Student [Jing Wang] who declares
to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is
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
the supervision of completion of the work, not for the contents.
Contents
1. INTRODUCTION...4
2.THEORETICAL FRAMEWORK & LITERATURE REVIE...6
3. MODELS AND METHODS...8
4. CONCLUSION...13
5. APPENDIX...16
6. BIBLIOGRAPHY...22
Section 1: Background and introduction
Standard economics sometimes allegedly fails to predict real life behavior because of ‘cognitive issues’ such as there is no savings for retirement and people´s risk aversion behavior. By combining economics with psychology, behavioral economics provides realism to standard models and presents itself as a more practical approach. Although there are some doubts about the importance and applicability of it, the rising need of understanding deviations from standard economic models and implications of them make behavior economics an
increasingly significant and pragmatic research field. As Erik (2012) suggested: the days behavioral economics started out at the margins of the economic professions are long gone and behavior economics is also increasingly having an impact in business and government. By analyzing the violations of standard assumptions and evaluate its relevance to real-world outcomes we could gain useful insights and apply practical economics to welfare and public policy and contribute to the overall utility of the society. Governments can also use behavioral economics to encourage citizens to do the right thing.
Understanding the importance of behavioral economics entails us to think about the core and central spirit of it which is essentially people´s irrationality and there is a matching piece that makes these irrational behaviors disappear and even alter. Experimental studies of human choice behavior have documented clear violations of rational economic theory and triggered the development of behavioral
economics (Eyal et al., 2015). To address this problem Kahneman and Tversky (1979) presented a critique of expected utility theory and brought about a new approach called prospect theory to model people’s decision making behavior under risks and captured the joint effect of four of the most important deviations from maximization: the certainty effect (Allais paradox, Allais, 1953), the reflection effect, overweighting of low probability extreme events, and loss aversion (see top four rows in Table 1). To capture all of these phenomena Ido et al. (2015) ran an “estimation set” study that included randomly selected choice problems to
investigate people´s choices under risks and uncertainty, which will be discussed in the following part of the study.
Before going further to the core of this study, a concept will be introduced, that is: heuristics. As Gerd (2011) defined: a heuristic is a strategy that ignores part of the information, with the goal of making decisions more quickly, frugally, and
accurately than more complex methods. The goal of making judgments more quickly and frugally is consistent with the goal of effort reduction, where “frugal”
is often measured by the number of cues that a heuristic searches (Gerd, 2011). Kahneman & Frederick (2002) proposed that a heuristic assesses a target attribute by another property (attribute substitution) that comes more readily to mind. In real life heuristics play a significant role in problem-solving and decision-making
process. Although heuristics are helpful in many situations, they can also lead to biases and systematic mistakes. As Daniel (2003) described, prototype heuristics are associated with two major biases: (i) Violations of monotonicity: Adding elements to a set may lower the average and cause the judgment of the target variable to decrease, contrary to the logic of extensional variables. (ii) Extension neglect: Other things equal, an increase in the extension of a category will increase the value of its extensional attributes, but leave unchanged the values of its
prototype attributes. The apparent base rate neglect is an example of this ( Daniel , 2003).
Despite the fact that behavioral economists have done a great job in connecting psychology and economics, up to now, however, most have focused on cognitive illusions and anomalies in order to prove the descriptive failure of neoclassical economic models (Gerd, 2011). Some conjectured that “mental illusions should be considered the rule rather than the exception” (Thaler, 1991, p. 4). However, how are these rules determined and regulated by people and what are the real-life implications of them have remained an unsolved puzzle and required more
investigation. Naturally the use of heuristics to explain people’s decision making process and the underling cognitive biases becomes a questionable yet promising project. This study will explore the proper use of heuristics in explaining people’s rationale behind decision making and the cognitive biases anomalies and provide useful forecasts of decisions under risk and ambiguity based the experimental data from Ido et al. (2015).
This study is formed in this way. Theoretical framework and literature review are discussed in the following part. Models and explanation is
presented as the third part. Eventually, a conclusion is drawn based on the previous analysis.
Section 2.Theoretical framework & literature review
The classical view of standard economic models indicating people making choices based on the maximization of expected utility failed to explain many cognitive biases and mind pitfalls existing in our every life. A lot of scholars have studied the rationale behind people’s decision making process. One of the most influential studies: the prospect theory, proposed by Daniel Kahneman and Amos
Tversky (1992) states that people make decisions based on the potential value of losses and gains rather than the final outcome, and that people evaluate these losses and gains using certain heuristics . On the contrary of modeling optimal decisions people make, it models real-life choices. According to them, utility is defined based on a reference point which depends on people’s initial endowments. People value gains and losses differently as people are more reluctant to losses and
experience more pains when suffer from losses. Based on prospect theory
following studies continue to analyze these phenomena and the experiment results of decision making and choices making also captured several effects within the prospect theory known as certainty Effect, break-even effect or Reflection Effect, overweighting of rare events and loss aversion.
This brings us an interesting question : How are these effect defined and what do they imply? Certainty indicates that people prefer choices with certainty (100% probability in the experiment design) compared to less certain events. According to Daniel (1979), the preference between negative prospects is the mirror image of the preference between positive prospects, which means the reflection of prospects around 0 reverses the preference order. Reflection effect implies that risk aversion in the positive domain is accompanied by risk seeking in the negative domain (Daniel, 1979). Overweighting of rare events implies that a tremendous remuneration accompanied by an extremely low probability seems to be more tempting than its counterpart. Eventually loss aversion refers to people's tendency to strongly prefer avoiding losses to acquiring gains, which may also explain sunk cost effects.
Understanding the four prominent cognitive biases helps us recognize people's decision making process better and lead to the more in-depth enquiry of the
rationale of choices under risk and uncertainty. On the contrary to the sophisticated and optimal utility optimization of decisions, in our daily life judgments and
choices are normally intuitive and reasonably successful (Klein, 1998). As Daniel (2013) described, some choices are better described as expressions of an affective response than as economic preferences. Especially the prevalence of framing
effects suggests that people mostly do not think very hard and are easily influenced
by the external environment. Despite the complexity and uncertainty in people’s judgement and selection, several rules are put forward to arrive at optimal
decisions and accurate judgments. One of the complex algorithms brought about by Payne, Bettman, & Johnson, (1993) is the weighted additive rule. According to this rule, decision makers consider all of the available alternatives and cues for each alternative, and then weigh each cue based how it contributes to an
alternative’s overall value (Anuj et al. 2008).
Another significant approach is the use of heuristics, which is also the core of this paper. In judgment and decision making, heuristics are considered as an effort saving approach by reduction of one or more of the following: (a) examining fewer cues, (b) reducing the effort of retrieving cue values, (c) simplifying the weighting of cues, (d ) integrating less information, and (e) examining fewer alternatives (Shah & Oppenheimer, 2008). Heuristic strategies are used to be considered as general verbal descriptions of behaviour that indicate little about the process of information search (Gilovich et al., 2002). However, more recently Gigerenzer et al. (1999) have introduced the notion of simple heuristics that are step-by-step process models which also imply information search strategies (Gigerenzer et al., 1999). Nowadays, simple heuristics have been developed such as two-alternative choice tasks (Gigerenzer & Goldstein, 1996), estimation tasks (Hertwig, Hoffrage, & Martignon, 1999) and multi- and binary-classification tasks (Berretty, Todd, & Martignon, 1999; Dhami & Ayton, 2001).
What of great interest to this paper is Gerd’s study in 2011. He contributed to the understanding of heuristics by introducing one way to model classification: Fast-and-Frugal Trees. It has building blocks similar to “take-the-best” practice: 1. Search rule: Search through cues in a predetermined order. 2. Stopping rule: Stop search as soon as a cue leads to an exit. 3. Decision rule: Classify the object
accordingly (Martignon et al. 2003). This approach provides important implication and contributes to the construction of modeling building of this paper as the basic structure of the heuristic used here are applied in the form of trees.
The approach analyzed in this paper has gained acceptance of many scholars. According to Mandeep (2010), cognitive psychologists have widely argued that people rely on heuristic strategies when making judgments and decisions. And these psychologically plausible strategies may be well adapted to the task
environment, and so can lead to optimal or near optimal behavior (Gigerenzer et al., 1999). Relating to this issue, plenty of studies of cognitive biases have shed light on the theoretical part of the prospect theory and few provide an intuitive and
practical approach for predicting people’ choices. Moreover, many literatures have introduced the concept and underlining rationale behind the basic heuristics and few studies the application of it. This paper intends to contribute to the ongoing investigation and study of heuristics and sheds light on people’s preferences and decision making by analyzing the cognitive biases existing in our everyday life and try to predict people’s behaviors.
Section 3: Model building and explanation
For the analysis of people’s preferences under risk and uncertainty, a heuristic model is built in this paper to predict people’s behaviors facing two options with different probability and payoff in the experiment conducted by Ido et al. (2015). In order to predict people’s choices, a set of heuristics are proposed to model the behavioral phenomena regarding cognitive biases illustrated in the Table 1. The heuristics are divided into three categories: when there is no feedback and no ambiguity, when there is feedback and no ambiguity, and when there is ambiguity. Eight out of the eleven dimensions from a study conducted by Ido et al. (2015) are used as parameters of the payoff distribution to form this model.
Definitions of Terms
The eleven parameters include: LA, HA, pHA, LB, HB, pHB, ambiguity and feedback; Participates face a choice between Option A that provides HA with probability pHA or LA otherwise (with probability 1 − pHA), and Option B that provides a lottery (that has an expected value of HB) with probability pHB, and provides LB otherwise (with probability 1 − pHB).
The parameter, Ambiguity (Amb), captures the precision of the initial information the decision maker receives concerning the probabilities of the outcomes in Option B. Amb= 1 implies no initial information concerning these probabilities (they are described with undisclosed fixed parameters), and Amb= 0 implies complete information and no ambiguity (Ido et al., 2015).
Having a feedback indicates that people will be informed of the final result of the experiment. Decision makers faced each problem first without feedback, and then with full feedback (they will be told the obtained and forgone outcomes following each choice).
EVA is the expected value of option A and is defined as: HA*pHA + LA*(1-pHA).
EVB is the expected value of option B and is defined as: HB*pHB + LB*(1-pHB).
The term “significant difference” is defined as greater than or equal to a difference of 5 payoff. An “insignificant difference” is defined as less than a difference of 5 in payoff. The term “much greater” is defined as HB> 5*HA. The term “small” is
defined as a payoff ≤ 2. These definitions were based off of trends in the data from Experiments 1 and 2.
Case 1: No feedback, no ambiguity
The flow diagram is used in addition to pseudo code for this section. This diagram does not use subscripts (HA is used instead of HA) in order to make the image
clearer, however subscripts will be used in the text of this paper. Every terminal node on the flowchart has the letter “A” or “B.” A terminal node ending in letter A indicates that the predicted proportion of option B is expected to be < 50%. A terminal node ending in letter B indicates that the predicted proportion of option B is expected to be > 50%. Letters A and B are used instead of the terminology used in the pseudo code to maintain clarity in the diagram.
The pseudo code for problems with feedback and without ambiguity is as follows, going from left to right on the flow diagram where a bullet point is used for each line of pseudo code, and bullet points are indented when appropriate to make the pseudo code clearer. The term Pred1(B) is the prediction of the proportion of choosing B in trials 1-5 with no feedback.
● Calculate the expected values of option A and option B ● If EVA = EVB, determine if pHA = 1
○ If pHA = 1, determine if HA≥ 0
■ If HA ≥ 0, determine if there is a significant difference between
HA and HB
● If there is a significant difference, determine if LB< 0
○ If LB < 0, Pred1(B) < 50%
○ If LB> 0, Pred1(B) > 50%
● If there is not a significant difference, Pred1(B) < 50% ■ If HA < 0, determine if HA< HB ● If HA< HB, determine if HB = 0 ○ If HB = 0, Pred1(B) > 50% ○ If HB ≠ 0, Pred1(B) < 50% ● If HA> HB, Pred1(B) > 50% ○ If pHA≠ 1, Pred1(B) = 50%
● If EVA < EVB, determine if pHA = 1
○ If pHA = 1, determine if HA≥ 0
■ If HA ≥ 0, determine if there is a significant difference between
HA and HB
● If there is a significant difference, determine if LB≤ 0
○ If LB≤ 0, Pred1(B) < 50%
○ If LB≥ 0, Pred1(B) > 50%
● If there is not a significant difference, Pred1(B) < 50% ■ If HA< 0, Pred1(B) > 50%
○ If pHA≠ 1, determine if LB< 0
■ If LB< 0, Pred1(B) < 50%
■ If LB> 0, Pred1(B) > 50%
● If EVA > EVB, determine if: LA< 0 AND HA≥ 0 AND LB≥ 0 AND HB ≥ 0
○ If LA = 0 AND HA> 0 AND LB> 0 AND HB > 0, Pred1(B) > 50%
○ If one or more of the conditions two lines up does not hold, Pred1(B) < 50%
Case 2: Feedback, no ambiguity
The term Pred5(B) is the prediction of the proportion of choosing B in trials 21-25 with feedback. The pseudo code is organized by bullet points.
● Calculate expected values of option A and option B
● If there is a significant difference between EVA and EVB, EVA > EVB, Pred5(B) < 50%
● If there is a significant difference between EVA and EVB, EVB > EVA, Pred5(B) > 50%
● If there is an insignificant difference between EVA and EVB and/or PHB is
high and/or LA is negative AND all other payoffs (including the ones of
option B) are positive, Pred5(B) = 50%
Case 3: Ambiguity
The term Pred(B) is the prediction of the proportion of choosing B trials 1-25 regardless of feedback. The pseudo code is organized by bullet points.
● If pHA is known, calculate EVA
● If pHB is known, calculate EVB
● In the case that HB is much greater than HA (> times 5) and EVA is small (>
2) ,Pred(B) > 50%
● Otherwise, Pred(B) = 50%
Pseudo code Explanation
The pseudo code explanation is organized in the same order as above. The section regarding no feedback and no ambiguity is done by referring to the numbered final outcomes on the flow chart above.
The Certainty Effect, also known as Allais Paradox, is captured by the model in outcome [1.4 and 2.4]. By specifying pHA=1, a certain payoff is included in the decision problems. Furthermore, according to the Experiments 1 and 2 data, we could come to the conclusion that HA>0 and no significance difference between HA
and HB must hold in order for the certainty effect to be present. If HA< 0, the
Break-even effect or Reflection Effect come into play (although this model does not take the Reflection Effect into account as there is no evidence for it in the data of Experiments 1 and 2). If there is a significance difference between HA and HB,
people may be willing to play the lottery. Overweighting of rare events may lead to this outcome if the probability of the higher outcome in B is extremely small. If LB=0, still under the condition of HB being significantly larger than HA, people
choose B due to the Get-Something Effect [1.5].
If HA<0 (people choose B if EVA<EVB [2.3]) and HA<HB a further phenomenon
may arise based on one more specification. Namely, if HB=0, people choose B
more, which can be attributed to the Break-Even Effect [1.3]. As mentioned before, Reflection Effect is not included in the model because people still choose option A although the payoffs are negative when analyzing the data of the experiments. Consequently, people are not risk-loving in the negative domain, as Kahneman and Tversky claim.
Relaxing the condition of a certain payoff, we assume Loss Aversion will influence people to choose A when LB<0, even though HB is significantly larger than HA [1.6
and 2.2] (in the case where EVA>EVB, we observe loss aversion when LA is the
only negative payoff and people tend to choose B [3.2] ). Note, if LB is not strongly
negative, meaning that it has a low magnitude, Loss Aversion is credited to shifting people’s decisions towards option B, according to Low Magnitude Eliminates Loss Aversion Effect (Kahneman and Tversky, 1979) . Yet, more than 50% of the people choose A and are not prone to this effect in the data of experiments 1 and 2 and thus, it is not included in the model [1.6].
Based on the experiments, people choose A and B in equal proportion when there is no certain payoff and the expected values are equal [1.7]. It can be argued that the relative stakes and probabilities could have an impact on people’s decisions.
Yet, the model is kept as simple as possible so that the heuristics are more flexible and able to apply to more situations.
When feedback is present, it can be concluded that people tend to choose the option with the higher expected payoff. However, there are some exceptions. The most important exception is the case where the difference between the two expected payoffs is very small or barely noticed. In that case, people’s decisions cannot be modeled and there is a 50% chance of choosing either option. Finally, if the probability of the higher outcome of option B (pHB) is perceived as high and if
the lower outcome of option A is negative, whereas all other payoffs, including the ones of option B, are positive, people will choose B more often. This conclusion is based on trends in the Experiments 1 and 2 data.
In most cases (except for the 1 case outlined in the pseudo code above), ambiguity cannot be modeled because there is insufficient reasoning. When p = unknown comes into play, heuristics cannot be made due to the unpredictable nature and lack of identifiable patterns of ambiguous questions. Looking at the ambiguous questions in Experiments 1 and 2 reveals that there is nearly no trend in behavior towards ambiguity. Therefore, to model ambiguity, the best model is a uniform distribution where Option A will be selected 50% of the time and Option B will be selected 50% of the time.
Section 4. Conclusion and summary
From the section above, one could conclude that the heuristics proposed here worked properly in predicting people’ choices under risk and uncertainty.
However, the limitations of the experiment data applied here may jeopardize the validity of the conclusion. Thus, the potential problems and shortcoming of the experiment design and this study will be discussed in the following part.
In order to reveal people’s underlying rationale, sample selection plays a pivotal role in the experiment design. In the experiment conducted by Ido et al., data is collected from students. Consequentially, the data set is not diversified, and a sampling bias could be present. In order to draw a more accurate conclusion, a
larger and more diversified sample selection should be provided. The 4-5 second time constraint for making decisions should also be noted because that is possibly too short for people to really understand the question and make rational choices. Also, incentives present a significant component of the experiment design. In their experiment, real payments are used as financial incentives. Financial incentives move the data closer to the game/decision-theoretic prediction, but also reduce the variability of the data (Ortmann, 2009). People may not take losses seriously when the losses are not substantial. In their experiment, high stake questions should also be asked in order to better model real behavior. Whether this reward system works for this forecasting is still questionable and it makes the experiment vulnerable to further questions.
There is ample room to make improvement of this model and draw a more valid conclusion by increasing the significance of the result. Additionally, a more advanced dynamic analysis could relax assumptions made in this study and therefore increase the validity of the results. Last but not least, a few more
parameters could be introduced to make the model here more explanatory. And the application of more valid and significant data could improve the confidence of the conclusion as well.
To conclude, the heuristic proposed based on the experiment conducted by Ido et al. (2015) is disputable and imperfect in its nature. As a consequence, the conclusion drawn here remains it validity in some way but it still requires further investigation and examination.
Decision making in the real world typically involves heuristics because the conditions for rational models rarely hold in an uncertain world (Gerd, 2011).
Heuristics are a good starting point but they could lead to faulty conclusions in some cases. As Gerd (2011) argued, a heuristic is not good or bad, rational or irrational; its accuracy depends on the structure of the environment, which is covered as the ecological rationality. Despite the various problems presented in this study, it cannot be denied that behavioral economics is important and the application of it to make predictions remains a useful method. In the real world people don’t follow classical economic models. Although coming up with a comprehensive model is very challenging, the use of heuristics to make predictions
of people’s behaviors incorporating cognitive biases and mind pitfalls can be considered as a valuable approach.
Section 5.Appendix
Table 1 (resourse from: Ido Erev, Eyal Ert, Ori Plonsky, 2015)
Table 2 (resourse from: Ido Erev, Eyal Ert, Ori Plonsky, 2015)
Bibliography
Andreas Ortmann (2009). The Way in which an Experiment is Conducted is Unbelievably Important.Experimentation Practices of Economists and
Psychologists.CESifo Working Paper No. 2887, pp.28-55.
Daniel Kahneman, Amos Tversky (1979). Prospect Theory: An Analysis of Decision under Risk.Econometrica, 47(2), pp. 263-291.
Daniel Kahneman (2013). Maps of Bounded Rationality.Psychology for
Behavioral Economics. The American Economic Review.
Erik Angner (2012). Course in Behavioral Economics. Palgrave Macmillan, New York.
Gary Klein (1998). Sources of power: How people make decisions. Cambridge, MA: MIT Press.
Gerd Gigerenzer, Wolfgang Gaissmaier (2011). Heuristic Decision Making.
Annual Review of Psychology, 62, 107-39.
Gigerenzer, G., Todd, P. M., & The ABC Research Group. (1999). Simple heuristics that make us smart. Oxford: Oxford University Press.
Gilovich, T., Griffin, D., & Kahneman, D. (2002). Heuristics and biases: The psychology of intuitive Judgment. Cambridge, UK: Cambridge University Press.
Ido Erev, Eyal Ert, Ori Plonsky (2015). From Anomalies to Forecasts:
A Choice Prediction Competition for Decisions under Risk and Ambiguity. Kahneman D, Frederick S. (2002). Representativeness revisited: attribute
substitution in intuitive judgment in Heuristics and Biases: The Psychology of
Intuitive Judgment, New York: Cambridge Univ. Press, pp. 49–81.
Martignon L, Vitouch O, Takezawa M, Forster M. (2003). Naive and yet enlightened: from natural frequencies to fast and frugal trees. In Thinking: Psychological Perspectives on Reasoning, Judgment, and Decision Making, ed. D Hardman, L Macchi, pp. 189–211. Chichester, UK: Wiley.
SefaHayibor, David M. Wasieleski (2008). Effects of the Use of the Availability Heuristic on Ethical Decision-Making in Organizations.Journal of Business
Ethics, 84, pp.151–165.DOI 10.1007/s10551-008-9690-7.
Shah AK, Oppenheimer DM. (2008). Heuristics made easy: an effort-reduction framework. Psychol. Bull.137:207–22.
Thaler, R.H. (1991). Quasi Rational Economics.Sage, New York. Payne, J. W., Bettman, J. R., Johnson, E. J. (1993). The adaptive decision maker. New York: Cambridge University Press.
