The goal of this paper is to provide an overview on the Virtual Laboratory infrastructure for online economic experiments. We summarize our experience gained from performing several economic experiments on the Internet. The experiments we have run range from electronic markets to individual decision making. From there we synthesize and evaluate a set of methodological issues in performing economic experiments on the Internet. As a result for further exploration we sketch the design of an infrastructure that allows the automated execution of Internet experiments including marketing of experiments, control of application and participation, payment system integration, and evaluation of results. The infrastructure also aims at providing a generic interface for third parties to register and run experiments.

We study combinatorial prediction markets where agents bet on the sum of values at any tree node in a hierarchy of events, for example the sum of page views among all the children within a web subdomain. We propose three expressive betting languages that seem natural, and analyze the complexity of pricing using Hanson’s logarithmic market scoring rule (LMSR) market maker. Sum of arbitrary subset (SAS) allows agents to bet on the weighted sum of an arbitrary subset of values. Sum with varying weights (SVW) allows agents to set their own weights in their bets but restricts them to only bet on subsets that correspond to tree nodes in a fixed hierarchy. We show that LMSR pricing is NP-hard for both SAS and SVW. Sum with predefined weights (SPW) also restricts bets to nodes in a hierarchy, but using predefined weights. We derive a polynomial time pricing algorithm for SPW. We discuss the algorithm’s generalization to other betting contexts, including betting on maximum/minimum and betting on the product of binary values. Finally, we describe a prototype we built to predict web site page views and discuss the implementation issues that arose.

The dominantscientific methodology for short term forecasting in important elections uses trial-heat polls, in which a sample of respondents report their current electoral preferences.

We construct a novel agent-based model of prediction markets in which putative human qualities like learning, reasoning, and profit-seeking are absent. We show that the prices which emerge from a market populated by a class of distinctly inhuman agents, Zero-Intelligence agents with diffuse beliefs, replicate the findings of empirical market studies. We use this result to argue against the prevailing descriptive theories of price formation in prediction markets, which have stressed the role of expert, rational participants.

This paper provides an overview on the simulations and experiments we have done in order to better understand human-agent interaction in a market environment. We find that the introduction of software agents does not necessarily induce a more efficient market. More surprisingly, information on the existence of software agents in the market environment results in more efficient behavior of human traders.

Successful new product development is crucial for firms ’ competitive advantage. Despite various sophisticated methods and high investments, new products still face notoriously high failure rates. A very critical stage in new product development is product concept testing for the go/no-go decision in further product development. Since there is a high number of different product concepts to test, there obviously is a need for a reliable, valid and efficient method, which can benefit from the scalability and interactivity of Internet-based technologies. Internet-based information markets are a new method to support new product development, based on the market efficiency hypothesis. We empirically evaluate product concepts with information markets. Further, we compare the results of the information markets with traditional research methods.

Preference markets address the need for scalable, fast and engaging market research. For faster new product development decisions, we implement a flexible prioritization methodology for product features and concepts, one that scales up in the number of testable alternatives, limited only by the number of participants. Preferences are measured by trading stocks whose prices are based upon share of choice of new products and features. We develop a conceptual model of scalable preference markets, and test it experimentally. Our conceptual model posits that individuals: (a) develop expectations of others based on self preferences, (b) use those expectations when buying and selling stocks, (c) have their opinions differentially weighted by the market pricing mechanism, resulting in a consensus of opinions, (d) learn from trading, and further converge towards consensus. Four studies confirm (a)- (d). Beyond accuracy, advantages of the methodology include speed (less than one hour per trading experiment), scalability (question capacity grows linearly in the number of traders), flexibility

We construct a novel agent-based model of prediction markets in which putative human qualities like learning, reasoning, and profit-seeking are absent. We show that the prices which emerge from a market populated by a class of distinctly inhuman agents, Zero-Intelligence agents with diffuse beliefs, replicate the findings of empirical market studies. We use this result to argue against the prevailing descriptive theories of price formation in prediction markets, which have stressed the role of expert, rational participants.

with aggregate uncertainty

Please do not cite or circulate without permissionAre Crowds Wise 2 Although researchers have documented many instances of crowd wisdom, it is important to go beyond these demonstrations to know whether some kinds of judgments may lead the crowd astray, whether crowds ’ judgments improve with feedback over time, and whether crowds’ judgments can be improved by changing the way those decisions are elicited. We tested these hypotheses in a sports gambling context (predictions against point spreads) that features prices that are widely believed to reflect crowd wisdom. A season-long experiment in which a diverse sample of enthusiastic NFL football fans wagered more than $20,000 on NFL football games found that the crowd was systematically biased and unwise: Faulty intuitions led bettors to predict “favorites ” more than “underdogs ” against point spreads that disadvantaged favorites. This bias persisted even among bettors who were told that the point spreads were biased against favorites. Moreover, the crowd’s bias increased over time, a result that may have been caused by attributions for success and failure that caused participants to “learn ” that intuitive choices were wise. However, when the crowd was asked to predict game outcomes by estimating point differentials rather than by predicting against point spreads, its predictions were unbiased and wiser. Thus, the same crowd of bettors can emerge wise or unwise, depending on how predictions are elicited. Are Crowds Wise 3 The wisdom-of-crowds hypothesis predicts that the independent judgments of a crowd of individuals (as measured by any form of central tendency) will be relatively accurate, even when most of the individuals in the crowd are ignorant and error-prone (Surowiecki, 2004). Examples