Sanmay Das

A learning market-maker in the Glosten-Milgrom model

This paper develops a model of a learning market-maker by extending the Glosten–Milgrom model of dealer markets. The market-maker tracks the changing true value of a stock in settings with informed traders (with noisy signals) and liquidity traders, and sets bid and ask prices based on its estimate of the true value. We empirically evaluate the performance of the market-maker in markets with different parameter values to demonstrate the effectiveness of the algorithm, and then use the algorithm to derive properties of price processes in simulated markets. When the true value is governed by a jump process, there is a two regime behaviour marked by significant heterogeneity of information and large spreads immediately following a price jump, which is quickly resolved by the market-maker, leading to a rapid return to homogeneity of information and small spreads. We also discuss the similarities and differences between our model and real stock market data in terms of distributional and time series properties of returns.This paper develops a model of a learning market-maker by extending the Glosten-Milgrom model of dealer markets. The market-maker tracks the changing true value of a stock in settings with informed traders (with noisy signals) and liquidity traders, and sets bid and ask prices based on its estimate of the true value. We empirically evaluate the performance of the market-maker in markets with different parameter values to demonstrate the effectiveness of the algorithm, and then use the algorithm to derive properties of price processes in simulated markets. When the true value is governed by a jump process, there is a two regime behaviour marked by significant heterogeneity of information and large spreads immediately following a price jump, which is quickly resolved by the market-maker, leading to a rapid return to homogeneity of information and small spreads. We also discuss the similarities and differences between our model and real stock market data in terms of distributional and time series properties of returns.

Anarchy, stability, and utopia: creating better matchings

Historically, the analysis of matching has centered on designing algorithms to produce stable matchings as well as on analyzing the incentive compatibility of matching mechanisms. Less attention has been paid to questions related to the social welfare of stable matchings in cardinal utility models. We examine the loss in social welfare that arises from requiring matchings to be stable, the natural equilibrium concept under individual rationality. While this loss can be arbitrarily bad under general preferences, when there is some structure to the underlying graph corresponding to natural conditions on preferences, we prove worst case bounds on the price of anarchy. Surprisingly, under simple distributions of utilities, the average case loss turns out to be significantly smaller than the worst-case analysis would suggest. Furthermore, we derive conditions for the existence of approximately stable matchings that are also close to socially optimal, demonstrating that adding small switching costs can make socially (near-)optimal matchings stable. Our analysis leads to several concomitant results of interest on the convergence of decentralized partner-switching algorithms, and on the impact of heterogeneity of tastes on social welfare.

A bayesian market maker

Ensuring sufficient liquidity is one of the key challenges for designers of prediction markets. Variants of the logarithmic market scoring rule (LMSR) have emerged as the standard. LMSR market makers are loss-making in general and need to be subsidized. Proposed variants, including liquidity sensitive market makers, suffer from an inability to react rapidly to jumps in population beliefs. In this paper we propose a Bayesian Market Maker for binary outcome (or continuous 0-1) markets that learns from the informational content of trades. By sacrificing the guarantee of bounded loss, the Bayesian Market Maker can simultaneously offer: (1) significantly lower expected loss at the same level of liquidity, and, (2) rapid convergence when there is a jump in the underlying true value of the security. We present extensive evaluations of the algorithm in experiments with intelligent trading agents and in human subject experiments. Our investigation also elucidates some general properties of market makers in prediction markets. In particular, there is an inherent tradeoff between adaptability to market shocks and convergence during market equilibrium.

Anarchy, Stability, and Utopia: Creating Better Matchings

We consider the loss in social welfare caused by individual rationality in matching scenarios. We give both theoretical and experimental results comparing stable matchings with socially optimal ones, as well as studying the convergence of various natural algorithms to stable matchings. Our main goal is to design mechanisms that incentivize agents to participate in matchings that are socially desirable. We show that theoretically, the loss in social welfare caused by strategic behavior can be substantial. However, under some natural distributions of utilities, we show experimentally that stable matchings attain close to the optimal social welfare. Furthermore, for certain graph structures, simple greedy algorithms for partner-switching (some without convergence guarantees) converge to stability remarkably quickly in expectation. Even when stable matchings are significantly socially suboptimal, slight changes in incentives can provide good solutions. We derive conditions for the existence of approximately stable matchings that are also close to socially optimal, which demonstrates that adding small switching costs can make socially optimal matchings stable.

Identifying Relevant Data for a Biological Database: Handcrafted Rules versus Machine Learning

With well over 1,000 specialized biological databases in use today, the task of automatically identifying novel, relevant data for such databases is increasingly important. In this paper, we describe practical machine learning approaches for identifying MEDLINE documents and Swiss-Prot/TrEMBL protein records, for incorporation into a specialized biological database of transport proteins named TCDB. We show that both learning approaches outperform rules created by hand by a human expert. As one of the first case studies involving two different approaches to updating a deployed database, both the methods compared and the results will be of interest to curators of many specialized databases.

Expert-mediated sequential search

This paper studies markets, such as Internet marketplaces for used cars or mortgages, in which consumers engage in sequential search. In particular, we consider the impact of information-brokers (experts) who can, for a fee, provide better information on true values of opportunities. We characterize the optimal search strategy given a price and the terms of service set by the expert, and show how to use this characterization to solve the monopolist expert’s service pricing problem. Our analysis enables the investigation of three common pricing schemes (pay-per-use, unlimited subscription, and package pricing) that can be used by the expert. We demonstrate that in settings characteristic of electronic marketplaces, namely those with lower search costs for consumers and lower costs of production of expert services, unlimited subscription schemes are favored. Finally, we show that the platform that connects consumer and experts can improve social welfare by subsidizing the purchase of expert services. The optimal level of subsidy forces the buyer to exactly fully internalize the marginal cost of provision of expert services. In electronic markets, this cost is minimal, so it may be worthwhile for the platform to make the expert freely available to consumers.

Instructor rating markets

We describe the design of Instructor Rating Markets (IRMs) where human participants interact through intelligent automated market-makers in order to provide dynamic collective feedback to instructors on the progress of their classes. The markets are among the first to enable the empirical study of prediction markets where traders can affect the very outcomes they are trading on. More than 200 students across the Rensselaer campus participated in markets for ten classes in the Fall 2010 semester. In this paper, we describe how we designed these markets in order to elicit useful information, and analyze data from the deployment. We show that market prices convey useful information on future instructor ratings and contain significantly more information than do past ratings. The bulk of useful information contained in the price of a particular class is provided by students who are in that class, showing that the markets are serving to disseminate insider information. At the same time, we find little evidence of attempted manipulation by raters. The markets are also a laboratory for comparing different market designs and the resulting price dynamics, and we show how they can be used to compare market making algorithms.

Matching, cardinal utility, and social welfare

Matching markets have historically been an important topic in economics research. On the positive (descriptive) side, researchers have modeled everything ranging from marriage markets to labor markets using the framework of matching. Matching was also one of the first areas in which market design made a name for itself, perhaps most famously in the redesign of the market that matches graduating M.D.s to their first residency programs in the United States. The arrival of computer scientists to the field of market design in general can be traced to many of the reasons suggested recently by Conitzer [2010] (in a broader context than just market design) in an article in Communications of the ACM, including the effects of new markets that have been made possible by advances in networking and Internet technology, a more computational mindset in general, and also the ability to view problems from a different perspective. In the case of matching markets in particular, in addition to the (often) constructive nature of computational approaches, there is also the historical fact that computer scientists have studied matching from many different perspectives, perhaps because matching markets have a very natural representation in the language of graphs.

Finding Transport Proteins in a General Protein Database

The number of specialized databases in molecular biology is growing fast, as is the availability of molecular data. These trends necessitate the development of automatic methods for finding relevant information to include in specialized databases. We show how to use a comprehensive database (SwissProt) as a source of new entries for a specialized database (TCDB, the Transport Classification Database). Even carefully constructed keyword-based queries perform poorly in determining which SwissProt records are relevant to TCDB; we show that a machine learning approach performs well. We describe a maximum-entropy classifier, trained on SwissProt records, that achieves high precision and recall in cross-validation experiments. This classifier has been deployed as part of a pipeline for updating TCDB that allows a human expert to examine only about 2% of SwissProt records for potential inclusion in TCDB. The methods we describe are flexible and general, so they can be applied easily to other specialized databases.

Team Formation in Social Networks

It is now recognized that the performance of an individual in a group depends not only on her own skills but also on her relationship with other members of the group. It may be possible to exploit such synergies by explicitly taking into account social network topology. We analyze team-formation in the context of a large organization that wants to form multiple teams composed of its members. Such organizations could range from intelligence services with many analysts to consulting companies with many consultants, all having different expertise. The organization must divide its members into teams, with each team having a specified list of interrelated tasks to complete, each of which is associated with a different reward. We characterize the skill level of a member for a particular task type by her probability of successfully completing that task. Members who are connected to each other in the social network provide a positive externality: they can help each other out on related tasks, boosting success probabilities. We propose a greedy approximation for the problem of allocating interrelated tasks to teams of members while taking social network structure into account. We demonstrate that the approximation is close to optimal on problems where the optimal allocation can be explicitly computed, and that it provides significant benefits over the optimal allocation that does not take the network structure into account in large networks. We also discuss the types of networks for which the social structure provides the greatest boost to overall performance.