Robizon Khubulashvili

In this paper, we investigate market design for online gaming platforms. A significant fraction of such platforms' revenue is generated by advertisements, in-app purchases, and subscriptions. Thus, it is necessary to understand which factors influence how much time users spend on the platform. We focus on one such factor - the outcome of the previous game. Using data from an online chess platform, we find strong evidence of history-dependent stopping behavior. We identify two primary types of players: those who are more likely to stop playing after a loss and those who are more likely to stop playing after a win. We propose a behavioral dynamic choice model in which the utility from playing another game is directly affected by the previous game's outcome. We structurally estimate this time non-separable preference model and then conduct counterfactual analyses to evaluate alternative market designs. In the context of online chess games, a matching algorithm that incorporates stopping behavior can substantially alter the length of play. 

An important function of a crowdfunding platform is to mitigate information asymmetry between entrepreneurs and investors by transmitting private information from the former to the latter. But can the platform be trusted? Using data from Kickstarter, we estimate a dynamic model of cheap talk, develop a statistical test confirming that the platform's incentives undermine the credibility of its signals, propose regulations that would curb those incentives, and quantify their welfare consequences. These regulations enable the platform to commit to an information disclosure rule and lead to Pareto improvements. We show that the platform's long-run reputation concerns could substitute for commitment.

I explore the dynamics of collaboration under incomplete information, focusing on the tension between the benefits and costs of collaboration. Specifically, I examine the case of two agents: one is an incumbent with well-known ability, and the other is an entrant with unobservable ability. If the incumbent's incentive to free ride depends on the entrant's possible types and learns the collaborator’s type based on history, then accumulating the reputation of being a high-ability type will lead to a breakup of the partnership. The breakup occurs because the incumbent's incentive to free ride increases if the entrant accumulates a high enough reputation. I design an experiment to study the incumbent's incentives to free ride after observing different paths of the entrant's reputation building. As predicted by theory, I find that reputation-building might hinder collaboration. 

We study a trade-off between innovation and investor protection on reward-based crowdfunding platforms. Informing investors about the risks of an investment opportunity protects them from failure, but comes at the cost of dissuading innovation. We show that a regulator, who values investor protection, may find it optimal to choose disclosure requirements that are not fully informative about projects. Partial disclosure enables investors to commit to sometimes funding bad projects, encouraging further innovation. We provide conditions under which a profit-motivated platform sets regulator-optimal disclosure requirements and study substitutability between regulation of disclosure and reputation systems. 

We investigate how international conflicts impact the behavior of hostile nationals in online games. Utilizing data from the largest online chess platform, where players can see their opponents' country flags, we observed behavioral responses based on the opponents' nationality. Specifically, there is a notable decrease in the share of games played against hostile nationals, indicating a reluctance to engage. Additionally, players show different strategic adjustments: they opt for safer opening moves and exhibit higher persistence in games, evidenced by longer game durations and fewer resignations. This study provides unique insights into the impact of geopolitical conflicts on strategic interactions in an online setting, offering contributions to further understanding human behavior during international conflicts.

Matching mechanisms that elicit strength-of-preference can exhibit efficiency gains over those that do not. To quantify these gains, we propose a measure of approximate ex-ante Pareto efficiency. We use this notion to quantify the efficiency improvement of the raffles mechanism (which we define) over the deferred acceptance mechanism (DA). We complement our theoretical analyses with experimental results. Using human subjects, we find that the raffles mechanism yields higher average payoffs as predicted by the theory, despite the fact that subjects play only approximately-optimal strategies.

In many situations both luck and effort are needed to secure a good outcome. Behavioral patterns such as outcome bias - a tendency to discount luck and judge individuals on (ex-post) outcomes rather than the effort they put in - could discourage individuals from investing effort. This is especially true for real-effort settings, where random outcomes (eg a correct medical diagnosis) are often separated from the qualitatively rich process of ex-ante effort investment (e.g. studying conscientiously) which is inadequately captured by numbers (e.g four years of medical school). In this novel online study, agents use mouse clicks to complete a real-effort task that increases principals' chance for a lottery win. In our Numerical treatment, we provide principals with statistics related to agents' mouse clicks. In our Visual treatment, we additionally provide a 30-second video of the effort investment process (replaying agents' mouse clicks).  We observe outcome bias in the Numerical treatment: principals reward agents 18 cents for each 1 std deviation increase in effort when it results in a win but only by 8 cents when it results in a loss. This bias disappears in the Visual treatment (1 std dev in effort = 12 cents in both wins and losses). This paper provides suggestive evidence that outcome bias in real effort tasks can be reduced by providing principals a window into the process of agents' effort production.

With the help of more than 700 reviewers, we assess the reproducibility of nearly 500 articles published in the journal Management Science before and after the introduction of a new Data and Code Disclosure policy in 2019. When considering only articles for which data accessibility and hardware and software requirements were not an obstacle for reviewers, the results of more than 95% of articles under the new disclosure policy could be fully or largely computationally reproduced. However, for 29% of articles, at least part of the data set was not accessible to the reviewer. Considering all articles in our sample reduces the share of reproduced articles to 68%. These figures represent a significant increase compared with the period before the introduction of the disclosure policy, where only 12% of articles voluntarily provided replication materials, of which 55% could be (largely) reproduced. Substantial heterogeneity in reproducibility rates across different fields is mainly driven by differences in data set accessibility. Other reasons for unsuccessful reproduction attempts include missing code, unresolvable code errors, weak or missing documentation, and software and hardware requirements and code complexity. Our findings highlight the importance of journal code and data disclosure policies and suggest potential avenues for enhancing their effectiveness.

In this paper, we describe the implementation of an information-sharing platform, got-toilet-paper.com. We create this web page in response to the COVID-19 pandemic to help the Pittsburgh, PA community share information about congestion and product shortages in supermarkets. We show that the public good problem of the platform makes it difficult for the platform to operate. In particular, there is a sizable demand for the information, but supply satisfies only a small fraction of demand. We provide a theoretical model and show that the first best outcomes cannot be obtained in a free market and the best symmetric equilibrium outcome decreases as the number of participants increases. Also, the best symmetric equilibrium has two problems, cost inefficiency and positive probability of termination. We discuss two potential solutions. The first is a uniform random sharing mechanism, which implies randomly selecting one person every period who will be responsible for information sharing. It is ex-post individually rational but hard to implement. The second solution is the one that we began implementing. It implies selecting a person at the beginning and make her responsible to share information every period while reimbursing her cost. We discuss the reasons for high demand and low supply both qualitatively and quantitatively.

''An advanced city is not a place where the poor move about in cars, rather it's where even the rich use public transportation''. This is what Enrique Peñalosa, the celebrated ex-mayor of Bogota once said. However, in order to achieve this objective, one of the crucial properties that the public transportation systems need to satisfy is reliability. While reliability is often referenced with respect to on-schedule arrivals and departures, in this study we are interested in the ability of the system to satisfy the total passenger demand. This is crucial, since if the capacity of the system is not enough to satisfy all the passengers, then ridership will inevitably drop. However, quantifying this excess demand is not straightforward since public transit data, and in particular data from bus systems that we focus on in this study, only include information for people that got on the bus and not those that were left behind at a stop due to a full bus. In this work, we design a framework for estimating this excess demand. Our framework includes a Poisson regression model for the demand for a given bus route and stop and mechanism for identifying instances of potential excess demand. These instances are filtered out from the training phase of the Poisson regression. We show through simulated data that this filtering is able to remove the bias introduced by the censored data observed/logged by the system that results in underestimation of the excess demand. We then apply our approach to real data collected from the Pittsburgh Port Authority and estimate the excess demand over a one-year period.