Market Design for Gaming Platforms: Using Stopping Behavior in Matching Algorithms (job market paper) with Ala Avoyan and Giorgi Mekerishvili
We investigate market design for online gaming platforms. Since a large part of such platforms’ income is generated by advertisement, it is essential to know what influences how long users stay on the platform. We provide evidence of history-dependent stopping behavior in non-monetary environments. We find that there are two types of people: those who are more likely to stop playing after a loss; and others, who are more likely to keep playing until a win. We find that an individual’s type is time-invariant over the years. We propose a behavioral dynamic choice model where utility from playing another game is directly affected by the outcome of the previous game. We structurally estimate this time non-separable preference model and conduct counterfactual analyses to evaluate alternative market designs. We find that in the context of online chess games a matching algorithm that incorporates stopping behavior can increase the length of play by 5.44%.
Work in Progress
Collaboration Cycles of R&D Teams: Theory and Experiment (draft available soon)
I investigate the possibility of collaboration and potential dynamics among two R&D teams, which are called research joint venture (RJV). I show that with observable types, collaboration and first-best outcomes are sustainable for some parameter values. Next, I consider two cases with asymmetric information, in which one player’s ability is observable (player A), while another player’s ability is private information (player B). First, I show that if the ability of player A is moderately low, RJV and first-best outcomes are sustainable for a limited period of time, but there is a break up with positive probability in every equilibrium. Break up happens because of accumulating a high reputation. Second, when the ability of player A is very high, the RJV is sustainable for some time, but the low ability player B is shirking on the equilibrium path, so first-best outcomes are not attainable. In this case, no matter what player B’s actual type is, there will be a break up and re-establishment of the RJV with positive probability. The last result of this paper explores the case of multiple projects. If player B has two projects to complete — one with a high type player A and another with a low type player A — order which player B chooses depends on his actual type. High type player B starts a collaboration with a low type player A and moves with an established good reputation with high type player A. Contrary, low type player B starts from a high type player A and moves to a new project (collaborated with low type player A) with a depleted reputation. I also designed an experiment to test theoretical predictions in the laboratory. In line with the theory prediction, I found that reputation building might lead to the break up of the RJV.
Belief Updating Under Uncertainty in Financially Relevant Situations
I study how people respond to the arrival of new information after they have already made an investment decision and are considering whether to pull out their investment or not. The genesis of this project came from my analysis of proprietary data from an online bookmaker. While suffering from selection issues, these data showed evidence that subjects updated their beliefs in the face of new information in a way that was both heterogeneous and systematic. To better understand how new information affects investors decision I created a new game, the “cash-out” game, and run it in the laboratory. The patterns I observe in the lab data are similar to the patterns we see in the field data. In particular, we can divide the population into three main groups. The first group is more likely to withdraws an investment if they get better news – under-interpret good news. The second group is more likely to withdraw investment after getting worth news – over-interpret bad news. In the last group, investors decide to “cash-out” with a higher chance if the news is less informative – over-interpret less informative good and bad news. This information can be used by bookmakers to better design their platforms and increase profits.