How do risk preferences influence the speed and character of social learning? I consider this question with theoretical modeling and a laboratory experiment. In the theoretical model, I generalize the information cascade settings (Banerjee, 1992; Bikhchandani et al., 1992), assuming that agents differ in risk preferences and choose between asymmetrically risky options. In the laboratory experiment, I compare the herding decisions of agents in settings with symmetrically and asymmetrically risky options. Preliminary results of the model and the pilot experiment show that heterogeneity of agents’ risk preferences may slow down convergence speed and change the probability of convergence to different options.
Experimental design
In this study I investigate how risk aversion influences the sequential choice process in the generalized version of the information cascades model. Specifically, I modify the classical models of information cascades (Banerjee, 1992; Bikhchandani et al., 1992), and test the implications of the generalized model in a laboratory experiment. In contrast to the classical information cascades models (Banerjee, 1992; Bikhchandani et al., 1992), where all the agents are risk-neutral and choose between some symmetrically risky options, I assume that the options differ in the risk values, and the agents differ in their risk preferences. In the laboratory experiment, I compare the cascading and herding decisions of participants in settings where they have to choose between two symmetrically risky options or between two asymmetrically risky options.
Design details
Randomization method
Randomization is done in the office as a public lottery -- the seating in the laboratory is randomized (participants draw their computer number from a bag with all computer numbers randomly before entering the room). Each computer number corresponds to a particular role in the experiment, so each participant chooses their role in the experiment themselves randomly, drawing a number of the computer they will be using during the experiment.
Randomization unit
Individual
Clusters
270 participants
Observations
Minimum 1890 observations (minimum 7 observations per participant, maximum 22 observations per participant)
Arms
90 participants per treatment arm
Primary outcomes
1. Percentage of actual cascades happening in all potential information cascade situations for players 3, 4 and 5;
2. Percentage of incorrect cascades happening in all actual information cascade situations for players 3, 4 and 5;
3. Percentage of actual cascades happening in all potential information cascade situations for players 4 and 5;
4. Percentage of incorrect cascades happening in all actual information cascades situations for players 4 and 5;
5. Percentage of actual herding happening in all potential herding situations for players 3, 4 and 5;
6. Percentage of incorrect herding happening in all actual herding situations for players 3, 4 and 5;
7. Percentage of actual herding happening in all potential herding situations for players 4 and 5;
8. Percentage of incorrect herding happening in all actual herding situations for players 4 and 5.
In all these outcomes, potential herding means that players 1 and 2 chose the same option, and player 3, 4 or 5 may choose to follow their choices or not. Actual herding means that potential herding happened for player 3 (or 4,5), and player 3 (or 4,5) followed the choices of the first two players. Incorrect herding means that actual herding happened and the option which the players chose brought them the lower outcomes, given the state of nature (realization of the coin flip). The same definitions hold for information cascades, but in this case the signal of the player 3 (or 4,5) is different from the option that players 1 and 2 chose.
Cited by
No citation data available for this registration yet.