Schedule - Parallel Session 4 - Risk
WMG IDL03 (IDL Auditorium) - 11:00 - 12:30On the Accuracy of Floods Cumulative Risk Perceptions: Evidence from Judgement and Choice
Cristobal De La Maza; Ines Azevedo; Alex Davis; Cleotilde Gonzalez
Abstract
Previous studies have shown, people have difficulty understanding how risks accumulate over time. We assess how people perceive cumulative risks in the context of flooding events. For example, suppose that there is a 1% chance of flooding in a specific location each year. When asked to assess the probability of at least one flood occurring in thirty years, people use variants of two heuristics for computing the cumulative risk deviating from the normative response (26%): 1) They take the average probability (1%), or 2) They sum the probabilities for all periods (30%) (Doyle 1997, Juslin et al. 2015). Previous studies have asked participants to provide direct assessments of the cumulative risk, measuring risk from their judgments. In this work, we compare two frameworks: a judgment task, similar to what has been used in previous risk perception studies, where participants are asked to provide a direct assessment of cumulative risk, and a choice task, where we use the stated choices from alternatives of insurance coverage to understand the effect of cumulative perceived risks in subjects risky choices, allowing us to compare the perceived risk across two elicitation modes, each with their own heuristics and biases. The survey was implemented using Amazon in M-Turk. We used a randomized controlled trial where MTurk participants were assigned to either a judgment or choice condition. Also we analyzed the effect of providing additional cumulative probability information. We found that subjects judgments can be represented by a bimodal distribution, with a group that severely underestimate the risk and a group that moderately overestimate the risk. Regarding choice preferences, we observe that judgments have an influence in choices. If individuals underestimate the risk they are less inclined to pay for insurance. On the contrary, if individuals overestimate the risk they recognize the insurance coverage option as more favorable. We must highlight that the predicted behavior is not as dramatic as we predicted based on expected and non-expected utility formulation. Further, we propose a model to account for subjective judgments when modeling choices.
Are Sellers Biased? A Meta-Analysis of Buyers and Sellers' Pricing of Risky Prospects
Nathaniel Ashby
Abstract
The disparity existing between buyers and sellers valuations where buyers valuations are lower than sellers “commonly referred to as the endowment effect “has been frequently investigated. A plethora of theories have resulted from these investigations, with many positing that the effect is driven by sellers, who for whatever reason (e.g., affective attachment and loss aversion), show upward biases in their valuations. Our investigations highlight four additional asymmetries existing between the valuations provided by buyers and sellers for risky prospects (lotteries) “asymmetries which suggest that sellers rather than buyers show valuations more in line with normative bench marks (expected values). We employed a meta-analytical approach looking at 28 studies, reported in 18 papers, totaling 3,295 experimental participants “nine of which we were able to obtain raw data for. In our first analysis we found that the classic endowment effect was robust, though we also find evidence that lack of monetary incentivization likely leads to larger estimates of the difference between buyers and sellers valuations: Indicating that incentivization is important if one wishes to obtain unbiased estimates. In our second analysis we find that compared to buyers, sellers provide valuations that are closer to normative values (i.e., show less of a difference between their valuation and the expected value of the risky prospect). However, as in our analyses of the common endowment effect, we find that incentivization decreases the size of this effect, further highlighting the importance of proper incentivization. In our third analyses we find that sellers rank ordering of prices are closer to the true rank ordering of normative prices than buyers. In our final analyses we find that while the variance in sellers valuations is higher than that of buyers, the variance per unit – also known as the precision score or coefficient of variance “is lower for sellers suggesting greater precision in sellers valuations than buyers. Together, these findings challenge common explanations for buyer-seller differences suggesting that disparities between the valuations provided by buyers and sellers are driven by sellers’ showing upwards biases in their valuations. Instead, these findings indicate that sellers show greater accuracy in their valuations than do buyers, and suggest that current theories of why the endowment effect occurs likely require some revision.
From Anomalies to Forecasts
Ido Erev; Eyal Ert; Ori Plonsky; Doron Cohen; Oded Cohen
Abstract
Experimental studies of choice behavior document distinct, and sometimes contradicting, deviations from maximization in different settings and experimental paradigms. Specifically, different behavioral phenomena emerge in decisions under risk and decisions under ambiguity, in decisions from description and decisions from experience, and in choice between binary gambles and choice between multi-outcome gambles. Most previous efforts to develop descriptive models of choice behavior address the distinct results by assuming different processes and proposing different models (or parameters) capturing the different choice anomalies. Implicit in these efforts is the assumption that the task of developing descriptive models is similar to the task of solving a puzzle: It is wise to start with a focus on the easy and interesting problems (areas); and the progress (added parts) will eventually clarify the more difficult areas, and the relationship between the different areas. The current paper evaluates an alternative approach: We consider the possibility that the development of descriptive models is more similar to the game “Scratch and Guess.” The optimal strategy in this game is to distribute the data collection efforts over a wide space to facilitate evaluation of the big picture. That is, we try to develop a general model capturing the coexistence and relative importance of the contradicting tendencies shown to emerge in different settings. Three steps are taken to reduce the risk of overfitting the data. First, we replicate 14 classical anomalies in one experimental paradigm. Next, we studied 60 problems randomly selected from a space that includes all problems examined in the replication study. Finally, to reduce the danger of an arbitrary selection of feasible models, an open choice prediction competition was organized. The organizers (first three co-authors) presented their favorite model and challenged other researchers to develop better models. Models were evaluated based on their predictions of 60 new problems. The results suggest that the classical “pre-feedback” phenomena are replicable, but that feedback eliminates most and instigates choice of the prospect minimizing probability of regret. The models that best capture the results assume: (a) high sensitivity to the best estimates of the expected values, (b) the use of several feedback-dependent heuristics, and (c) reliance on small samples.
Randomized Strategies and Prospect Theory in a Dynamic Context
Vicky Henderson
Abstract
Applying prospect theory (PT) in a dynamic context brings new challenges. We study PT agents facing optimal timing decisions and consider the impact of allowing them to follow randomized strategies. In the discrete model of casino gambling of Barberis (2012) we show that allowing randomization leads to gains in PT value. In the continuous analog (Ebert and Strack (2015)) we show that allowing randomization can significantly alter the predictions of the model. Ebert and Strack show that a naive investor never stops. We show that allowing naive PT agents to use randomized strategies leads to predictions which are closer to reality and include voluntary cessation of gambling.