CS286r Questions/Comments 10/6/2008

Xi Alice Gao

This paper uses Google's internal prediction markets to analyse how an organization like Google process information. Two findings are: (1) Google's markets are reasonably efficient, but are still affected by bias. (2) Opinions on topics are correlated among employees who are proximate in some sense, including physical proximity, and association in a social network. The accuracy of prediction markets is not the main concern of this paper. Rather, prediction markets are used in this context to analyze information flow without an organization. This could potentially lead to applications where corporations attempt to utilize prediction markets or other discovered communication median to better convey important information through the organization. The four biases can also help us to understand what factors drive employees in an organization to have different opinions on certain topics.

One thing interesting about Google's prediction markets is that the participants belong to a rather restricted group of people and they are thus not representative of the organization. This brings up the question of whether a group of more diverse people would exhibit different behaviour when participating in these prediction markets. When the authors are trying to derive the conclusions related to opinion correlation, some significant assumptions are made such that they can essentially rule out unwanted correlations possibly hidden in the data. In fact, in real world, it is quite possible that the effects of several factors could mix up together in one set of data. In this paper, however, the frequent moves of employees did help the authors to get a cleaner result.

I thought this paper relates to a talk I attended last week. The talk was primarily about investigating the interactions of social influence and similar interests in a social network. I think these two works are similar because they are both in some sense trying to understand the interactions in a social network. The difference is that this paper uses corporate internal prediction markets as the tool whereas the other work uses data from Wikipedia.

Xi Alice Gao

This paper considers share markets in Iowa Political Markets and attempts to evaluate the absolute efficiency of these markets as well as the efficiency of these markets relative to polls. The authors argue that predictions given by markets are more accurate and stable than those given by polls. Iowa Electronic Markets are interesting for researchers because it attempts to fill the gap between traditional experimental markets and the real world markets.

It seems to me that this paper can be viewed as a high level survey of related previous results. The paper is quite limited in the sense that it presents a list of claims without giving any supporting evidence. Little insight was given in how the results were obtained. This is especially apparent when I noticed that the paragraph discussing the reasons for market efficiency does not mention the data at all, and the paragraph discussing the data does not give much account for the trends in the data. Also, when discussing the absolute market accuracy, the paper states three factors explaining the variance in accuracy without explaining the reason for the occurrence of each factor.

The main thing unclear to me was the reasons why poll results are less accurate than market results. Also, the article uses the market price as of midnight on election eve as a measure for the market prediction, and this seems to me a rather arbitrary choice. An idea for extending the result could be to explore other measures for the market prediction and compare their relative accuracies. The paper also mentions many topics for future research projects that could use data obtained from Iowa Electronic Markets.

Using Prediction Markets to Track Information Flows: Evidence from Google
By Cowgill, Wolfers, and Zitzewitz

This paper presents a study of an electronic market simulation conducted within Google and among the Google employees. Such an experiment provides a unique opportunity to study the traits and habits of traders in a closed, controlled environment. In particular, the authors studies the affects of the traders' geographic proximity (e.g. same office, same floor) on the correlation between traders' behaviors with respect to the market. In addition, the authors search for common biases in trader behavior.

The paper's main results were twofold: 1) The paper finds strong correlations in trading between employees that are in very close proximity (i.e. within a few feet) of each other in the office, but not necessarily between employees working on different floors of the same office, nor between employees located in different offices. 2) The traders' exhibit an optimistism bias, especially when the Google stock appreciates and especially among new hires. The authors document several other biases as well, including overpricing of favorites, short aversion, and underpricing of extreme outcomes.

It is posited that such findings shed additional light on the behavior of entrepreneurial firms, with regards to the optimism bias. I am personally unclear on much extra insight such empirical results give, since it is already true and moreover obvious that entrepreneurs are generally optimistic, risk-averse people. Also, I think optimism biases are likely to occur in markets that do not involve real money, including these particular Google prediction markets. This is probable since traders may not take much risk assessment into account when acting, generally under-estimating the likelihood of negative events (and over-estimating the likelihood of positive events).

In addition, the results regarding the affect on geographic proximity on trader correlation are said to have applications to information flows between firms. However, it is noted in the paper that geographic distances as close as a single floor presented enough separation to prevent any such correlation. Therefore, it is difficult for me to believe that these results can be extended to the economic geography literature, which usually show that firms in the same counties, and even larger demographic boundaries, exhibit information flows and information spillovers.

I would like to see the same empirical work performed on a prediction market where real money is involved. For example, if the same amount of user data can be collected through the Iowa Electronic Markets (IEM) or intrade.com, the resulting study could be more plausible.

Results from a Dozen Years of Election Futures Markets Research
By Berg, Forsythe, Nelson, and Rietz

This paper studies the predictive power of the elections markets in the Iowa Electronic Markets (IEM). The results conclude that these markets have considerable predictive power, with midnight-before-election prices outperforming polls in most cases as forecasting devices.

In general, this paper (along with others that study electronic prediction markets) show great promise for using markets as forecast devices. It is particularly interesting how such markets' participants come from a very narrow demographic, consisting mostly of well-educated, high-income, young-to-middle-aged males. I wonder if a market could be designed to encourage other demographic groups to trade in the prediction markets, providing a more diverse and well-balanced sample of participants. I suspect that the accuracy of the prediction market as a forecast device would increase, since the average person (e.g. the average voter in America in the case of the election market) would be more well represented by the aggregate of the traders.

I would also like to see the results of this paper extended in the following way: instead of just comparing the bid and ask prices of the candidates on midnight before election day to the most recent poll results, I would like to see a time series analysis and comparison between the stock prices and the polls in the months leading up to the election. This may elucidate the connection, if there is any, between the prediction market prices and the poll results. The paper does not have be convinced that the prices are in some way a result of traders' beliefs as a result of receiving information from the polls. Future areas of research could also include studying the accuracy of other prediction markets besides election markets.

Results from a Dozen Years of Election Futures Markets (Berg, Forsythe, Nelson, Rietz)

Comments by Brian Young

The paper deals with the results of the Iowa Political Markets from 1988 to 2000, arguing that these prediction markets serve a similar purpose to pre-election polls, with a comparable or even superior degree of accuracy.

My biggest concern with the results from this paper are that, although they do cover a large number of countries, the time period addressed is by nature limited, since the markets have only been in operation for so long. However, that means the study only covers four US presidential election cycles (although the foreign elections do provide corroborative detail). Despite the strong correlation between the predicted and actual outcomes, the limited scope of the study as pertains to American presidential elections, which to my myopic view seem rather more widely covered than the foreign and regional elections, makes me wonder whether we may actually generalize these results.

Unlike polls, markets are opt-in, meaning that theoretically, they are manipulable, although I confess I'm not entirely sure what end that would serve. Still, if we begin to emphasize the markets in news reporting to the same degree that polls are emphasized now, it seems that the possibility of tampering is one we must recognize -- recall the (similarly manipulable) Internet polls early in the 2008 cycle that showed consistently high ratings for Republican candidate Rep. Ron Paul, whose primary showings did not bear out the poll results.

Although I'm willing to concede the authors' belief that "the differences between election markets and polls give the markets an edge in prediction" (6), I believe that poll results cannot (yet) be replaced by prediction markets about elections (a claim the authors, to be fair, do not make), if only because investors' beliefs are so dependent on poll results. Also, polls can give information that is hard to capture in a prediction market, such as potential voters' specific beliefs about candidates' personalities and character traits, as well as opinions about their policies.

Using Prediction Markets to Track Information Flows: Evidence from Google (Cowgill, Wolfers, Zitzewitz)

Comments by Brian Young

The paper deals with a Google internal prediction market, in which Google employees traded securities representing various outcomes, on subjects both Google-specific and otherwise.

The authors placed a great deal of emphasis on the interesting optimistic bias they discovered. This seems to be, among other things, a promising sign for Google, if its employees feel favorably about its future. Could other companies employ such a scheme? In describing the demographics of most investors, the authors suggest that most of the high-volume traders fit a certain profile, one that seems pretty common among Google employees but might be less prevalent in other industries.

I am curious about the reasons for the mellowing of this optimistic bias in more experienced traders. Is this caused by (as seems probable) experience leading to more astute predictions, or is there some kind of jading going on? It would be interesting to survey traders about their confidence in the company and compare this to their trading behavior.

The authors also note that their study covered a time period during which Google was largely successful. I wonder how the results would have been different over a less optimistic period of time, whether a down period for the company or for the economy as a whole. Trading volume might have been depressed as people faced more important concerns than the trading game. Without short selling, too, people might have been more reluctant to invest in complete sets of securities and more inclined to keep their Goobles in their pockets (as it were).

l This paper states market mechanism in an election, the reliability of polls and furthermore, the psychological biases in trading activities.

l I think this paper is important. It shows that the market is efficient by solid empirical results. Using future market to test the efficiency of the market and probe the mechanism in an election is quite a clever technique.

l A company can use the same technique on evaluating their products and strategies.

l I think the future market in this paper is similar to the extensive form game with perfect information we have learned from this class. With good communication of information (perfect information), the game (future market) has an equilibrium, where each player (trader) chose his own best response.

--
Hao-Yuh Su
Master student in Engineering Science
School of Engineering and Applied Sciences
Harvard University
Tel: 857-869-0104
E-mail: hysu@fas.harvard.edu

l Using IEM system, this paper analyzes the factors of psychological biases and information transmission between traders. It provided several different groups of data and used statistical measure to derive the vigorous conclusion.

l When measuring the transmission of communication, the author made an assumption that like-minded people do not proximate. I think it is overly-simplified, though I cannot deny the explanation the author offered.

l In this paper, one point I cannot fully understand is about the Figure 2 and its corresponding conclusion. Why can Figure 2 confirm the favorite bias? Why does it use 2-outcome market and 5-outcome market?

l I think the psychological biases can be viewed as a belief in the player's strategy, and it was part of the topics in the previous lectures.

--
Hao-Yuh Su
Master student in Engineering Science
School of Engineering and Applied Sciences
Harvard University
Tel: 857-869-0104
E-mail: hysu@fas.harvard.edu

CS286: Article Response
Peter Blair

Results from a Dozen Years of Election Futures Markets Research

In this article, the writers explain the structure of the Iowa Electronic Markets (IEM). Political securities are traded on the IEM. These securities are priced in a way that reflects the market's expectation concerning the election's outcome e.g. based on which candidate/party win or alternatively based on vote or seat share.

The article is clear and well-written. It explains the different types of contracts that exists on the market (unit portfolios and contracts based on vote/seat share) and then gives examples of the markets predictions compare to actual outcomes of elections both in the USA and other countries. The idea of an election market is a clever one -- by coupling a financial pay-off to the client's prediction, the market incentivizes the client giving a true opinion. This feature of the market stands in contrast to the partisan who may be polled by a political operative.

The one questions that I have about this article is: 1). What does the actions of "typical traders" tell us about arbitrage opportunities in a market where irrational considerations affect a client's participation in the market. An interesting follow-up research question would be to find ways to identify dominant biases of potential traders, quantify those effects and then come up with a set of bid/ask strategies that would exploit these arbitrage opportunities. One shining example of such an arbitrage opportunity would be the effect of race in this year's presidential elections -- some investors may process the impact of race in this election in a manner that identifies them as a "typical voter," which in turn creates an arbitrage opportunity for the savvy rational advisers.

In terms of future research direction,it is evident that prediction markets of this type can be used to track expected stock prices on the Dow and other international exchanges. Imagine clients not buying actual stock in a company, which can be very costly (e.g. few hundred dollar for Google's stock), but rather purchasing contracts that predict the probability that a given stock, or even the Dow Jones Industrial Average rises/falls. Such a market would be a great asset in times like these were there is great uncertainty about how the bailout will affect the long term stability of the global economy. Besides, an actual investor in the NYSE can use information from this predictive market as a reference point for considering how to participate in the market (NYSE).

In conclusion this article presents a very cogent argument for the utility of predictive electoral markets such as IEM.

--
Peter Blair
Physics PhD Candidate
Jefferson Laboratory, Harvard University

CS286r Article Review

Using Prediction Markets to Track Information Flows: Evidence from Google
by: Peter Blair

In "Using Prediction Markets to Track Information Flows: Evidence from Google", Cowgil Wolfers and Zitzewitz study correlations among sercuirties purchasing decisions of Google employees participating in the company's internal exchange market. Google's internal exchange market mimics the setup of a traditional financial market in which there are buyers and sellers interacting via a common currency "Goobles". The authors report on the effects of spacial separation of employees (mostly "quants, so hardly a random sample), membership to certain groups ("demographic similarity" e.g. professional associations and email lists) and other social ties, on correlations with their choice of market portfolios. Two central results are that neighboring employees have highly correlated purchases and secondly that employees are very optimistic about the performance of securities linked to Google's business operations, e.g. how many new users will Gmail attract this month (particularly on days when Google's real life stock was up). The authors claim that their findings "contribute to three quite different literatures: on the role of optimisms in entrepreneurial firms, on employee communication in organization, and on social networks and information flows among investors."

My main criticism of this paper is that there seem to be two papers in the one paper. The first paper would address the important problem of biases in the data arising from "over pricing of favorites, short aversion, optimism and underpricing of extreme outcomes" (pg 5). The second potential would be one discussion the correlations in the portfolio choices of employees based on demographic similarity and proximity. A separation of this single paper into these two papers mentioned above would be suitable in that the first paper would contribute directly to the literature on the role of optimism in entrepreneurial firms, while the second paper would contribute to the literature on employee communication in organization, and on social networks and information flows among investors.

My second criticism of this paper is that the correlations arising from spatial closeness is loosely discussed. Given the author's access to specific longitudinal and latitudinal coordinates of the employees in the study, they could potential do a plot of correlated security purchase as a function of distance. As a reader, I would be interested to see if such a correlation would be isotropic, for example, or if there are preferred directions of communication, perhaps selected based on some natural/physical structure e.g. a scenic window view, proximity to the vending machines or bathrooms or employee lounge etc. Notwithstanding this gentle criticism, the importance of spatial closeness is a surprising result, particularly given the minimal cost of communicating via alternative means e.g. IM, facebook, cell phones, e-mail -- a point that the authors make very successfully in the article. Further on the point of spatial separation, the reader is interested to see how this model of information flow would be mapped onto a larger open system of investors -- think a group of home business owners who are also investors. Otherwise put, is the spatial correlation in employees choices only significant because it exist within the context of an umbrella organization that unifies the employees, or does it apply to random groups of unrelated people working on similar questions? This is an important research question that the results of this paper motivate.

My third and final criticism relates to the homogeneity of the sample group. On the one hand, it is interesting that correlations owing to demographic similarities are relatively weak given that we have an ostensibly homogeneous demographic of "quants" and computer scientists/engineers -- it is certainly reasonable to expect that this might not be the case. On the other hand, it could be that this homogeneity washes out many correlations that would be based on demographic similarity (think race, gender, age, marital status) since these share characteristics would simply provide a background (as in constant baseline) for the signal of correlations based on other factors that are not controlled for by sampling from a homogeneous demographic.

This article is strong in its novel attempt to study a closed market system and from its study propose stylized facts about social interactions, information flow and investor decisions. The broad scope of issues and effects suggested, realized and explained provides the reader with lots of follow-up questions that are useful for contributing the the three litteratures mentioned in the article. This great strength of the article -- it's breadth, is also one of its weak points: because it packs in so many contributions to the respective litteratures, it comes across as unfocused on a first read, when in fact its filled with many gems that sow seeds for future research projects on information flow in investing markets.

--
Peter Blair
Physics PhD Candidate
Jefferson Laboratory, Harvard University

This paper provides empirical data testing the accuracy of prediction markets in predicting political events. The types of contracts discussed are such that a player is rewarded by playing according to the true nature of his belief. Prediction market data compares favorably to polling data both in terms of accuracy on the actual day of the vote and stability over time (i.e. less volatility compared to polling data).

One interesting thing to note about the nature of prediction markets is that market makers (marginal traders) drive prices and therefore predictions. This seems to run somewhat counter to the notion of prediction markets as aggregators of information from many sources as traders who play most frequently will have their information "count" more. I'd be interested in knowing how the number of traders and frequency at which they trade impacts the accuracy of prediction markets.

In comparing prediction markets to polls, the authors don't seem to go into detail about how polls are fundamentally different than prediction markets. Polls ask an individual "Who would you vote for?" not "Who do you think is most likely to win, and with what probability?" Thus, while prediction markets seem to beat traditional polls in terms of getting more accurate predictions, what about modified polls? Do prediction markets outperform those?

Another general question about prediction markets: How do play money markets perform vs. real money markets?

The focus on this pape= r was to describe not only how well internal prediction markets work in predicting future events, but also on how they can be used as a tool to better underst= and information flow within a firm. The paper found that traders in the same location (i.e. sharing a cubicle or office) tend to have strong positive correlation in prediction market positions. This was found to be more than = just a case of like-minded people being located in proximate positions since Goo= gle has people rotate desks often. Sitting on the same floor has little effect = of trading behavior, nor does being friends with an individual. The importance= of proximity is somewhat surprising to me despite the fact that chat clients, email, and telephone can effectively diminish the necessity of being next t= o someone to share information. There is one important caveat to the assumpti= on that everyone needs to be sitting on top of each other for optimal informat= ion flow, however. The questions posed in prediction market contracts were not necessarily related to employees' main jobs. For instance, if I am a coder working on optimizing an algorithm =96 how much more likely am I to ask the= coder sitting next to me for help than sending my code over to someone else to ta= ke a look? Intuitively, it seems that the first is more likely =96 but the paper= does not answer this question. If the cost of setting up a meeting, emailing, or chatting is in fact much more difficult and close proximity is the key to information flow =96 why is it that professors at Harvard work in offices r= ather than in cubicles?

Another interesting fi= nding of the paper was that there is a significant optimistic bias among Google employee= s. This raises some interesting social questions. Does optimism about a compan= y's prospects play a role in increasing worker productivity? Rephrased, how doe= s a declining stock price impact employee moral and worker productivity?

This paper's main objective is to show how the Iowa Electronic Markets (IEM) can be used to predict election results. The IEM is a real-money futures market conducted by University of Iowa College of Business and is touted to predict election results in many different markets in various countries. This concept and its implementation are important and interesting because, if precise, they can accurately predict a variety of election result data types, such as the margin of victory and anticipated election effects based on news.

The market itself doesn't model the actual election since the participants are well educated, male, and high-income. The market mechanism is very similar to financial markets in the sense that traders can see the current market price, set an asking bid, and a selling bid. From what we have seen in class about rationality and irrationality, the paper states that typical traders, or those that have hopeful beliefs for their preferred candidates, tell make irrational mistakes. This is because they decrease their payoff values regardless of the outcomes. Marginal traders who regularly trade drive the predictions.

The future work that can be drawn from this study involves applying this concept not only to elections but also to all zero sum events. Examples include sports and various tournaments. One project is to apply prediction markets to a specific sport such as basketball and predict the winner of the NBA championship over the course of the season. As the paper states, the results of the prediction markets are very accurate when taken over the whole season rather than over the few weeks towards the end.

-Subhash Arja

The main purpose of this paper is to show how the concept of prediction markets can be applied to studying how a large company, Google, processing information. Also, it shows how events that have no direct effect on the company's or employee's performance are modeled in the prediction markets. This study is important because it takes the concept which was explained in the Iowa Electronic Markets and applies them in a corporate setting rather than predicting elections. Such application, if extended further, has the potential to help the company find and make improvements to help the employees work more efficiently. This will, as a result, lead to higher productivity for the company and more profits.

As was stated in the previous IEM paper, the main insight is to treat an event like a company event or movie premiere as a market. In this case, the market asks a question that has 2-5 possible mutually exclusive outcomes. The study makes two assumptions. First, it is assumed that two like-minded people will not be physically close to each other. Second, observed measures of proximity are not correlated with unobserved proximity. In my opinion, both of these assumptions are valid because they do not constrict the possibilities but lay a foundation on which to base the study.

This paper directly relates to the Iowa Election Market paper and even references the work done there. One project, which is an extension of this paper, is to apply a prediction market between two competing companies. This could potentially give insight into whether the employees' mindset is different and what factors lead to one company eventually beating out the other.

-Subhash Arja

Xiaolu Yu

This paper analyzes the largest experiment with prediction markets.

Main conclusions and contributions:

1. Corporate prediction markets reveal some biases, but may perform better as collective experience increases.

a. The finding of a favorite bias in Google's markets (overpricing of favorites and underpricing of longshots) contradicts with Ali and Manski's results. The authors suggest a possible reason as the belief bias of the median trader.

b. Short selling aversion creates arbitrage opportunity.

c. The market traders have extreme aversion.

d. Optimistic bias exists in two categories of contracts where outcomes are most directly under the control of Google employees: company news and performance. The optimistic is larger in two outcome markets, early in sample period, and earlier in each quarter.

e. Number of shares transacted and sample period matters to the magnitude of biases.

f. Stock price variation induced mainly by optimistic bias, extreme aversion and favorite bias actually reflect ex post surprise. (Some thoughts: the economy meltdown lately in the U.S. causes similar reactions of the market: more disappointing, more unconfident, more extreme, and harder to predict than rational traders anticipated.)

g. Individual behaviors in driving prediction market biases: newly hired employees are more likely to take advantage of the reverse favorite-longshot and short aversion bias in prices, though trade optimistically; more experienced traders profit from optimism, favorite, and short aversion biases, but contributes to extreme aversion.

2. Proximity importantly matters in terms of the correlation between opinions on specific topics among employees. Physical proximity is the most important, whereas demographic does not affect information sharing significantly.

Applications:

1. Entrepreneurs are optimistically biased.

2. Considerable thought is put into optimizing physical locations in order to encourage communication between managers and workers, and among peers, which in turn facilitate information flow.

3. Physical proximity and social network are important in information sharing among investors.

How valid to 'using prediction market to track information flow'?

Assumptions:

1. Like-minded people are not proximate (proved by examing sharp changes in proximity). Information sharing forming more rapidly when employees share and office than when they are not, and when they switch offices frequently than when they are not.

2. Observed measures of proximity are not correlated with unobserved proximity. (Limitations: each employee name ten colleagues that they turned to for ideas and advice. However, the ten may not be those from whom he got the most information from, but those who were consulted by him most recently and hence he remember most clearly. Also, two employees have to name each other simultaneously in order to build the professional relationship. This process is sort of random.)

Caveat:

The prediction market subjects of information flow are ancillary to employees' main job. This caveat may explains the significance of physical proximity, but still assumably meaningful in better understanding product innovation and other creative processes.

Xiaolu Yu

One of the best known markets of IEM, the Iowa Political Markets shows considerable accuracy in the market forecasts, especially for large, U.S. election markets. The variance in accuracy may be explained by three factors. 1. Presidential election markets perform better than congressional, state and local election markets (larger, more active markets with fewer contracts are more accurate). 2. Markets with more volume near the election perform better than those with less (the marginal traders are much less prone to bias because of information update along the progression of the game.) 3. Markets with fewer contracts predict better than those with more. Comparing to polls, over the majority of the time the market ran, the predictions were dramatically more accurate and stable, and hence more valuable as long run forecasting devices.

In order for the market to work, there must be enough traders so that the aggregate of their knowledge can correctly forecast the outcome of the election. Meanwhile, the market mechanism must facilitate aggregation of their disparate information so that the prevailing market price becomes a sufficient statistic for their collective information. Core group traders/"marginal traders"/"market makers" that tend to set market prices appears less biased and error prone than typical traders, and make mistakes much less often.

IEM have great research potential. For example, Google's prediction markets are patterned on the Iowa Electronic Markets.

The main contribution of the paper is presenting the results of prediction markets of the past few years in an aggregated format. This review is important because it gathers together the findings of other papers and tries to justify the value of prediction markets as an accurate method of predicting event outcomes.

The limitation of this paper is that it talks about prediction market results and trades/market makers that make it accurate, however these conclusions are drawn using various results from different papers. As a result, there does not appear to be a single point that the author is trying to make, the topics range from the accuracy of the prediction markets, to the types of investors that make it accurate. Perhaps more detailed examination of each claim is required. However reading the cited papers should clear up these questions.

The main insight of the paper is that prediction markets can offer accurate long term results. This is however subject to the quality of the investors. Investors who are "market makers" and less likely affecteb by their personal choice in an outcome will make accurate predictions. The paper also talks about the accuracy of the prediction markets against traditional polls, and discusses how it out performs the polls.

After reading this paper, one can assume that using prediction market results as an indicator of how well candidates are doing in a election, might offer better insights into the outcome than polls. Furthermore, the current predictions of the market are based on the price the evening before the election and the average price a week ahead of the election. Using these sets of data, the prediction markets have been deemed accurate. A possible project would be to analyze prices further back in time, and to determine a starting point at which the markets will present accurate predictions.

Malvika Rao	This paper analyzes prediction markets internal to Google. Google employees participate voluntarily in prediction markets that raise questions ranging from “fun” topics such as the quality of the latest Star Wars movie to more serious topics such as the pricing of securities. The paper reveals that while Google internal prediction market results did approximate the event probabilities, there were some biases. First of all, it was discovered that there is an optimism bias - especially among newer employees, and more pronounced regarding projects under Google’s control. Also, a correlation in information among employees that sit close to each other was established. While the idea of tracking information flow within an organization is interesting, I feel that this paper does not provide any great insights. Obviously employees that sit close to each other would share information and might even play with these prediction markets together. The discussion section of the paper elaborates on why this may be so (low-cost communication). The optimism bias is also explained and indeed it is easy to see why there would be an optimism bias. I am uncertain if the range of subjects used in these prediction markets is large enough. Employees of a company may be optimistic about certain things but pessimistic about other issues. For example, these workers have chosen industry work over staying in academia. What would be their optimism level regarding university endowment issues, for example? Providing a very specific information medium such as the predictions market and measuring information correlation amongst people, levels and groups might not be the best way to track information flow in a company. It is more interesting to try to characterize what are the different types of information that flow across levels, groups, and people. What if the information exchanged had real consequences – would this still show a correlation between physically proximate people or might it show greater correlation between friends and trusted professional contacts? However it is tricky to track information in general – as it raises privacy questions. The problem with such experiments (D. Kahneman) is that the very act of conducting an experiment/survey biases the types of results we would get. Somewhat like the Heisenberg uncertainty principle.
Malvika Rao	This paper examines the Iowa Electronic Markets and more specifically the Iowa Political Markets. These markets are designed so that the contracts’ prices predict election outcomes. The markets appear to be quite accurate compared to the polls and in some cases even outperformed the polls. This might be because traders are paid for correct guesses about eventual election outcomes. The paper mainly describes the design of the market and outlines the market forecasts resulting from trading. I found this paper to be interesting in the way the market is designed to elicit behaviour that ultimately leads to accurate predictions. It might be worthwhile to explore what would be the efficiency of such a system if the auction mechanism were altered. Also, the aggregation of information might be looked into - could we improve it and how. There appear to be lots of further research questions here tied into the topic of market design.
Alice Gao	------=_Part_32837_31496944.1223263505196 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline Xi Alice Gao This paper uses Google's internal prediction markets to analyse how an organization like Google process information. Two findings are: (1) Google's markets are reasonably efficient, but are still affected by bias. (2) Opinions on topics are correlated among employees who are proximate in some sense, including physical proximity, and association in a social network. The accuracy of prediction markets is not the main concern of this paper. Rather, prediction markets are used in this context to analyze information flow without an organization. This could potentially lead to applications where corporations attempt to utilize prediction markets or other discovered communication median to better convey important information through the organization. The four biases can also help us to understand what factors drive employees in an organization to have different opinions on certain topics. One thing interesting about Google's prediction markets is that the participants belong to a rather restricted group of people and they are thus not representative of the organization. This brings up the question of whether a group of more diverse people would exhibit different behaviour when participating in these prediction markets. When the authors are trying to derive the conclusions related to opinion correlation, some significant assumptions are made such that they can essentially rule out unwanted correlations possibly hidden in the data. In fact, in real world, it is quite possible that the effects of several factors could mix up together in one set of data. In this paper, however, the frequent moves of employees did help the authors to get a cleaner result. I thought this paper relates to a talk I attended last week. The talk was primarily about investigating the interactions of social influence and similar interests in a social network. I think these two works are similar because they are both in some sense trying to understand the interactions in a social network. The difference is that this paper uses corporate internal prediction markets as the tool whereas the other work uses data from Wikipedia. ------=_Part_32837_31496944.1223263505196 Content-Type: text/html; charset=ISO-2022-JP Content-Transfer-Encoding: 7bit Content-Disposition: inline Xi Alice Gao This paper uses Google's internal prediction markets to analyse how an organization like Google process information. Two findings are: (1) Google's markets are reasonably efficient, but are still affected by bias. (2) Opinions on topics are correlated among employees who are proximate in some sense, including physical proximity, and association in a social network. The accuracy of prediction markets is not the main concern of this paper. Rather, prediction markets are used in this context to analyze information flow without an organization. This could potentially lead to applications where corporations attempt to utilize prediction markets or other discovered communication median to better convey important information through the organization. The four biases can also help us to understand what factors drive employees in an organization to have different opinions on certain topics. One thing interesting about Google's prediction markets is that the participants belong to a rather restricted group of people and they are thus not representative of the organization. This brings up the question of whether a group of more diverse people would exhibit different behaviour when participating in these prediction markets. When the authors are trying to derive the conclusions related to opinion correlation, some significant assumptions are made such that they can essentially rule out unwanted correlations possibly hidden in the data. In fact, in real world, it is quite possible that the effects of several factors could mix up together in one set of data. In this paper, however, the frequent moves of employees did help the authors to get a cleaner result. I thought this paper relates to a talk I attended last week. The talk was primarily about investigating the interactions of social influence and similar interests in a social network. I think these two works are similar because they are both in some sense trying to understand the interactions in a social network. The difference is that this paper uses corporate internal prediction markets as the tool whereas the other work uses data from Wikipedia. ------=_Part_32837_31496944.1223263505196--
Alice Gao	------=_Part_32855_16988235.1223263563366 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline Xi Alice Gao This paper considers share markets in Iowa Political Markets and attempts to evaluate the absolute efficiency of these markets as well as the efficiency of these markets relative to polls. The authors argue that predictions given by markets are more accurate and stable than those given by polls. Iowa Electronic Markets are interesting for researchers because it attempts to fill the gap between traditional experimental markets and the real world markets. It seems to me that this paper can be viewed as a high level survey of related previous results. The paper is quite limited in the sense that it presents a list of claims without giving any supporting evidence. Little insight was given in how the results were obtained. This is especially apparent when I noticed that the paragraph discussing the reasons for market efficiency does not mention the data at all, and the paragraph discussing the data does not give much account for the trends in the data. Also, when discussing the absolute market accuracy, the paper states three factors explaining the variance in accuracy without explaining the reason for the occurrence of each factor. The main thing unclear to me was the reasons why poll results are less accurate than market results. Also, the article uses the market price as of midnight on election eve as a measure for the market prediction, and this seems to me a rather arbitrary choice. An idea for extending the result could be to explore other measures for the market prediction and compare their relative accuracies. The paper also mentions many topics for future research projects that could use data obtained from Iowa Electronic Markets. ------=_Part_32855_16988235.1223263563366 Content-Type: text/html; charset=ISO-2022-JP Content-Transfer-Encoding: 7bit Content-Disposition: inline Xi Alice Gao This paper considers share markets in Iowa Political Markets and attempts to evaluate the absolute efficiency of these markets as well as the efficiency of these markets relative to polls. The authors argue that predictions given by markets are more accurate and stable than those given by polls. Iowa Electronic Markets are interesting for researchers because it attempts to fill the gap between traditional experimental markets and the real world markets. It seems to me that this paper can be viewed as a high level survey of related previous results. The paper is quite limited in the sense that it presents a list of claims without giving any supporting evidence. Little insight was given in how the results were obtained. This is especially apparent when I noticed that the paragraph discussing the reasons for market efficiency does not mention the data at all, and the paragraph discussing the data does not give much account for the trends in the data. Also, when discussing the absolute market accuracy, the paper states three factors explaining the variance in accuracy without explaining the reason for the occurrence of each factor. The main thing unclear to me was the reasons why poll results are less accurate than market results. Also, the article uses the market price as of midnight on election eve as a measure for the market prediction, and this seems to me a rather arbitrary choice. An idea for extending the result could be to explore other measures for the market prediction and compare their relative accuracies. The paper also mentions many topics for future research projects that could use data obtained from Iowa Electronic Markets. ------=_Part_32855_16988235.1223263563366--
Angela Ying	I thought this paper was extremely interesting and addressed a lot of different aspects of the problem. The main purpose of the paper was to study how the outcomes of the Google prediction market were affected by factors such as proximity of coworkers, Google stock, and the experience of the employee in trading. It found correlations between trading due to a variety of factors, but the most important was the proximity to coworkers. In addition, it characterized the typical trader as someone who comes from a more quantitative background, most likely an engineer / programmer, since they are more inclined to trade stocks. The study also found that traders were generally averse to trading long-shot categories and tended to be overly optimistic in the more popular trades. Overall, although the paper addressed many aspects of the problem, I wish that it had gone into more detail and numbers about the effectiveness of the market itself (like what margin of error did these markets have?) It would be interesting to look at the characteristics of the employees that Google hires. It seems that the engineers hired by Google would share common backgrounds and probably common ways of thinking about trading. Furthermore, Google presumably sends some kind of pro-Google propaganda to its employees, which may explain the overall optimism that we see, especially in new employees who simply may be grateful for getting hired and thus have a more positive outlook on the company. Finally, it would be interesting to conduct a study where employees are allowed to short sell, since the paper mentioned that this was forbidden.
Angela Ying	This paper is a survey on the overall success of the Iowa Electronic Markets, where traders can enter into a market to predict the outcome of a future event, such as a presidential election. The system is unique in that it forces traders to think about the general population's preferences, while typical political surveys only ask about the subject's own preferences. The latter type of prediction can be statistically biased depending on which subjects are chosen to participate in the survey, while the former has the advantage of not requiring a diverse group of people to participate. These results are extremely important because they often predict the outcome of the election better than political polls, which have a larger margin of error. If this is proven for a longer period of time, we could potentially see results from such markets on CNN.com and other political news site that keep track of how the election is going. It would be an interesting extension to this paper to isolate the effects of trader characteristics with how the market performs. For example, would a group of Republicans who are going to vote for John McCain going to bet on him in a predictions market? To do this, the markets would have to have certain requirements for a trader to enter the market, but this would reduce liquidity and may not produce good results. I am also wondering what other electronic prediction markets were available over this same time period. Today, there is InTrade.com and presumably other electronic markets - how well do these perform against IEM? Overall, I think that predictions markets are very interesting and extremely effective.
Avner May	This is a multi-part message in MIME format. --------------040809070404030307060405 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit I think that the main contribution of this paper is its study of information flow within an organization. Secondly, I think its discussion of biases, most notably an optimism bias, within an entrepreneurial setting is very important. I thought it was incredibly insightful that the authors looked into how correlated the trading patterns of people were based on different notions of proximity. This is something which has value in a context completely separate from prediction markets. It gives insight into the way people share information, and influence each others beliefs and actions; its answers questions like, who do people share the most information with? How do ideas spread over time within an organization? I think that using the setting of prediction markets to study these questions is extremely innovative and insightful. Additionally, I thought it was very interesting how employees within an organization tended to be optimistic with regard to the performance of that organization; as discussed in the paper, this has implications with regard to motivation in the workplace. It also presents the question of whether this optimism actually leads to higher performance than the organization would have reached otherwise. The work in this paper has a lot of implications with regard to how to set up work environments in order for people to communicate most effectively. I think that the same approach taken in this paper can be taken to studying information flow in different, maybe larger, settings, and would provide a lot of insight into information flow in these areas. --------------040809070404030307060405 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit I think that the main contribution of this paper is its study of information flow within an organization. Secondly, I think its discussion of biases, most notably an optimism bias, within an entrepreneurial setting is very important. I thought it was incredibly insightful that the authors looked into how correlated the trading patterns of people were based on different notions of proximity. This is something which has value in a context completely separate from prediction markets. It gives insight into the way people share information, and influence each others beliefs and actions; its answers questions like, who do people share the most information with? How do ideas spread over time within an organization? I think that using the setting of prediction markets to study these questions is extremely innovative and insightful. Additionally, I thought it was very interesting how employees within an organization tended to be optimistic with regard to the performance of that organization; as discussed in the paper, this has implications with regard to motivation in the workplace. It also presents the question of whether this optimism actually leads to higher performance than the organization would have reached otherwise. The work in this paper has a lot of implications with regard to how to set up work environments in order for people to communicate most effectively. I think that the same approach taken in this paper can be taken to studying information flow in different, maybe larger, settings, and would provide a lot of insight into information flow in these areas. --------------040809070404030307060405--
Avner May	This is a multi-part message in MIME format. --------------090101080001080809090103 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit I think the main contribution of this paper is the data it is presenting. It shows that these prediction markets are very accurate (often more accurate than large polls) in predicting the actual results of elections. I find the difference between these markets and polls to be very interesting; as stated in the article, polls ask people who they think they will vote for, whereas these markets financially reward people for correcting answering who /everyone/ will vote for on election day. I think that one of the reasons that the markets often outperform the polls is due to the fact that the people trading in the markets have access to all the poll data, as well as other information. Thus, more information is being processed by the markets than by the polls. Furthermore, giving people real /incentive/ to act truthfully based on their beliefs bolsters the success of this type of market. The relates closely to the system we discussed in class wherein a whether reporter has incentive to report truthfully The results in this paper seems like they would be of high interest to political analysts. In terms of the technical elements of the paper, I found it to not be incredibly interesting. I did not think there was much original thought, or any ground-breaking techniques, presented in this paper. I think the main limitation of the paper is that it does not discuss much of the theory behind why the results presented looked the way they did. The main thing the paper does is simply present the basic results to the readers. --------------090101080001080809090103 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit I think the main contribution of this paper is the data it is presenting. It shows that these prediction markets are very accurate (often more accurate than large polls) in predicting the actual results of elections. I find the difference between these markets and polls to be very interesting; as stated in the article, polls ask people who they think they will vote for, whereas these markets financially reward people for correcting answering who everyone will vote for on election day. I think that one of the reasons that the markets often outperform the polls is due to the fact that the people trading in the markets have access to all the poll data, as well as other information. Thus, more information is being processed by the markets than by the polls. Furthermore, giving people real incentive to act truthfully based on their beliefs bolsters the success of this type of market. The relates closely to the system we discussed in class wherein a whether reporter has incentive to report truthfully The results in this paper seems like they would be of high interest to political analysts. In terms of the technical elements of the paper, I found it to not be incredibly interesting. I did not think there was much original thought, or any ground-breaking techniques, presented in this paper. I think the main limitation of the paper is that it does not discuss much of the theory behind why the results presented looked the way they did. The main thing the paper does is simply present the basic results to the readers. --------------090101080001080809090103--
Brett Harrison	------=_Part_39263_31677795.1223220955688 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline Using Prediction Markets to Track Information Flows: Evidence from Google By Cowgill, Wolfers, and Zitzewitz This paper presents a study of an electronic market simulation conducted within Google and among the Google employees. Such an experiment provides a unique opportunity to study the traits and habits of traders in a closed, controlled environment. In particular, the authors studies the affects of the traders' geographic proximity (e.g. same office, same floor) on the correlation between traders' behaviors with respect to the market. In addition, the authors search for common biases in trader behavior. The paper's main results were twofold: 1) The paper finds strong correlations in trading between employees that are in very close proximity (i.e. within a few feet) of each other in the office, but not necessarily between employees working on different floors of the same office, nor between employees located in different offices. 2) The traders' exhibit an optimistism bias, especially when the Google stock appreciates and especially among new hires. The authors document several other biases as well, including overpricing of favorites, short aversion, and underpricing of extreme outcomes. It is posited that such findings shed additional light on the behavior of entrepreneurial firms, with regards to the optimism bias. I am personally unclear on much extra insight such empirical results give, since it is already true and moreover obvious that entrepreneurs are generally optimistic, risk-averse people. Also, I think optimism biases are likely to occur in markets that do not involve real money, including these particular Google prediction markets. This is probable since traders may not take much risk assessment into account when acting, generally under-estimating the likelihood of negative events (and over-estimating the likelihood of positive events). In addition, the results regarding the affect on geographic proximity on trader correlation are said to have applications to information flows between firms. However, it is noted in the paper that geographic distances as close as a single floor presented enough separation to prevent any such correlation. Therefore, it is difficult for me to believe that these results can be extended to the economic geography literature, which usually show that firms in the same counties, and even larger demographic boundaries, exhibit information flows and information spillovers. I would like to see the same empirical work performed on a prediction market where real money is involved. For example, if the same amount of user data can be collected through the Iowa Electronic Markets (IEM) or intrade.com, the resulting study could be more plausible. ------=_Part_39263_31677795.1223220955688 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline Using Prediction Markets to Track Information Flows: Evidence from Google By Cowgill, Wolfers, and Zitzewitz This paper presents a study of an electronic market simulation conducted within Google and among the Google employees. Such an experiment provides a unique opportunity to study the traits and habits of traders in a closed, controlled environment. In particular, the authors studies the affects of the traders' geographic proximity (e.g. same office, same floor) on the correlation between traders' behaviors with respect to the market. In addition, the authors search for common biases in trader behavior. The paper's main results were twofold: 1) The paper finds strong correlations in trading between employees that are in very close proximity (i.e. within a few feet) of each other in the office, but not necessarily between employees working on different floors of the same office, nor between employees located in different offices. 2) The traders' exhibit an optimistism bias, especially when the Google stock appreciates and especially among new hires. The authors document several other biases as well, including overpricing of favorites, short aversion, and underpricing of extreme outcomes. It is posited that such findings shed additional light on the behavior of entrepreneurial firms, with regards to the optimism bias. I am personally unclear on much extra insight such empirical results give, since it is already true and moreover obvious that entrepreneurs are generally optimistic, risk-averse people. Also, I think optimism biases are likely to occur in markets that do not involve real money, including these particular Google prediction markets. This is probable since traders may not take much risk assessment into account when acting, generally under-estimating the likelihood of negative events (and over-estimating the likelihood of positive events). In addition, the results regarding the affect on geographic proximity on trader correlation are said to have applications to information flows between firms. However, it is noted in the paper that geographic distances as close as a single floor presented enough separation to prevent any such correlation. Therefore, it is difficult for me to believe that these results can be extended to the economic geography literature, which usually show that firms in the same counties, and even larger demographic boundaries, exhibit information flows and information spillovers. I would like to see the same empirical work performed on a prediction market where real money is involved. For example, if the same amount of user data can be collected through the Iowa Electronic Markets (IEM) or intrade.com, the resulting study could be more plausible. ------=_Part_39263_31677795.1223220955688--
Brett Harrison	------=_Part_39912_29271812.1223224107846 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline Results from a Dozen Years of Election Futures Markets Research By Berg, Forsythe, Nelson, and Rietz This paper studies the predictive power of the elections markets in the Iowa Electronic Markets (IEM). The results conclude that these markets have considerable predictive power, with midnight-before-election prices outperforming polls in most cases as forecasting devices. In general, this paper (along with others that study electronic prediction markets) show great promise for using markets as forecast devices. It is particularly interesting how such markets' participants come from a very narrow demographic, consisting mostly of well-educated, high-income, young-to-middle-aged males. I wonder if a market could be designed to encourage other demographic groups to trade in the prediction markets, providing a more diverse and well-balanced sample of participants. I suspect that the accuracy of the prediction market as a forecast device would increase, since the average person (e.g. the average voter in America in the case of the election market) would be more well represented by the aggregate of the traders. I would also like to see the results of this paper extended in the following way: instead of just comparing the bid and ask prices of the candidates on midnight before election day to the most recent poll results, I would like to see a time series analysis and comparison between the stock prices and the polls in the months leading up to the election. This may elucidate the connection, if there is any, between the prediction market prices and the poll results. The paper does not have be convinced that the prices are in some way a result of traders' beliefs as a result of receiving information from the polls. Future areas of research could also include studying the accuracy of other prediction markets besides election markets. ------=_Part_39912_29271812.1223224107846 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline Results from a Dozen Years of Election Futures Markets Research By Berg, Forsythe, Nelson, and Rietz This paper studies the predictive power of the elections markets in the Iowa Electronic Markets (IEM). The results conclude that these markets have considerable predictive power, with midnight-before-election prices outperforming polls in most cases as forecasting devices. In general, this paper (along with others that study electronic prediction markets) show great promise for using markets as forecast devices. It is particularly interesting how such markets' participants come from a very narrow demographic, consisting mostly of well-educated, high-income, young-to-middle-aged males. I wonder if a market could be designed to encourage other demographic groups to trade in the prediction markets, providing a more diverse and well-balanced sample of participants. I suspect that the accuracy of the prediction market as a forecast device would increase, since the average person (e.g. the average voter in America in the case of the election market) would be more well represented by the aggregate of the traders. I would also like to see the results of this paper extended in the following way: instead of just comparing the bid and ask prices of the candidates on midnight before election day to the most recent poll results, I would like to see a time series analysis and comparison between the stock prices and the polls in the months leading up to the election. This may elucidate the connection, if there is any, between the prediction market prices and the poll results. The paper does not have be convinced that the prices are in some way a result of traders' beliefs as a result of receiving information from the polls. Future areas of research could also include studying the accuracy of other prediction markets besides election markets. ------=_Part_39912_29271812.1223224107846--
Brian Young	------=_Part_47308_4051053.1223266169892 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline Results from a Dozen Years of Election Futures Markets (Berg, Forsythe, Nelson, Rietz) Comments by Brian Young The paper deals with the results of the Iowa Political Markets from 1988 to 2000, arguing that these prediction markets serve a similar purpose to pre-election polls, with a comparable or even superior degree of accuracy. My biggest concern with the results from this paper are that, although they do cover a large number of countries, the time period addressed is by nature limited, since the markets have only been in operation for so long. However, that means the study only covers four US presidential election cycles (although the foreign elections do provide corroborative detail). Despite the strong correlation between the predicted and actual outcomes, the limited scope of the study as pertains to American presidential elections, which to my myopic view seem rather more widely covered than the foreign and regional elections, makes me wonder whether we may actually generalize these results. Unlike polls, markets are opt-in, meaning that theoretically, they are manipulable, although I confess I'm not entirely sure what end that would serve. Still, if we begin to emphasize the markets in news reporting to the same degree that polls are emphasized now, it seems that the possibility of tampering is one we must recognize -- recall the (similarly manipulable) Internet polls early in the 2008 cycle that showed consistently high ratings for Republican candidate Rep. Ron Paul, whose primary showings did not bear out the poll results. Although I'm willing to concede the authors' belief that "the differences between election markets and polls give the markets an edge in prediction" (6), I believe that poll results cannot (yet) be replaced by prediction markets about elections (a claim the authors, to be fair, do not make), if only because investors' beliefs are so dependent on poll results. Also, polls can give information that is hard to capture in a prediction market, such as potential voters' specific beliefs about candidates' personalities and character traits, as well as opinions about their policies. ------=_Part_47308_4051053.1223266169892 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline Results from a Dozen Years of Election Futures Markets (Berg, Forsythe, Nelson, Rietz) Comments by Brian Young The paper deals with the results of the Iowa Political Markets from 1988 to 2000, arguing that these prediction markets serve a similar purpose to pre-election polls, with a comparable or even superior degree of accuracy. My biggest concern with the results from this paper are that, although they do cover a large number of countries, the time period addressed is by nature limited, since the markets have only been in operation for so long. However, that means the study only covers four US presidential election cycles (although the foreign elections do provide corroborative detail). Despite the strong correlation between the predicted and actual outcomes, the limited scope of the study as pertains to American presidential elections, which to my myopic view seem rather more widely covered than the foreign and regional elections, makes me wonder whether we may actually generalize these results. Unlike polls, markets are opt-in, meaning that theoretically, they are manipulable, although I confess I'm not entirely sure what end that would serve. Still, if we begin to emphasize the markets in news reporting to the same degree that polls are emphasized now, it seems that the possibility of tampering is one we must recognize -- recall the (similarly manipulable) Internet polls early in the 2008 cycle that showed consistently high ratings for Republican candidate Rep. Ron Paul, whose primary showings did not bear out the poll results. Although I'm willing to concede the authors' belief that "the differences between election markets and polls give the markets an edge in prediction" (6), I believe that poll results cannot (yet) be replaced by prediction markets about elections (a claim the authors, to be fair, do not make), if only because investors' beliefs are so dependent on poll results. Also, polls can give information that is hard to capture in a prediction market, such as potential voters' specific beliefs about candidates' personalities and character traits, as well as opinions about their policies. ------=_Part_47308_4051053.1223266169892--
Brian Young	------=_Part_47311_17129592.1223266193675 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline Using Prediction Markets to Track Information Flows: Evidence from Google (Cowgill, Wolfers, Zitzewitz) Comments by Brian Young The paper deals with a Google internal prediction market, in which Google employees traded securities representing various outcomes, on subjects both Google-specific and otherwise. The authors placed a great deal of emphasis on the interesting optimistic bias they discovered. This seems to be, among other things, a promising sign for Google, if its employees feel favorably about its future. Could other companies employ such a scheme? In describing the demographics of most investors, the authors suggest that most of the high-volume traders fit a certain profile, one that seems pretty common among Google employees but might be less prevalent in other industries. I am curious about the reasons for the mellowing of this optimistic bias in more experienced traders. Is this caused by (as seems probable) experience leading to more astute predictions, or is there some kind of jading going on? It would be interesting to survey traders about their confidence in the company and compare this to their trading behavior. The authors also note that their study covered a time period during which Google was largely successful. I wonder how the results would have been different over a less optimistic period of time, whether a down period for the company or for the economy as a whole. Trading volume might have been depressed as people faced more important concerns than the trading game. Without short selling, too, people might have been more reluctant to invest in complete sets of securities and more inclined to keep their Goobles in their pockets (as it were). ------=_Part_47311_17129592.1223266193675 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline Using Prediction Markets to Track Information Flows: Evidence from Google (Cowgill, Wolfers, Zitzewitz) Comments by Brian Young The paper deals with a Google internal prediction market, in which Google employees traded securities representing various outcomes, on subjects both Google-specific and otherwise. The authors placed a great deal of emphasis on the interesting optimistic bias they discovered. This seems to be, among other things, a promising sign for Google, if its employees feel favorably about its future. Could other companies employ such a scheme? In describing the demographics of most investors, the authors suggest that most of the high-volume traders fit a certain profile, one that seems pretty common among Google employees but might be less prevalent in other industries. I am curious about the reasons for the mellowing of this optimistic bias in more experienced traders. Is this caused by (as seems probable) experience leading to more astute predictions, or is there some kind of jading going on? It would be interesting to survey traders about their confidence in the company and compare this to their trading behavior. The authors also note that their study covered a time period during which Google was largely successful. I wonder how the results would have been different over a less optimistic period of time, whether a down period for the company or for the economy as a whole. Trading volume might have been depressed as people faced more important concerns than the trading game. Without short selling, too, people might have been more reluctant to invest in complete sets of securities and more inclined to keep their Goobles in their pockets (as it were). ------=_Part_47311_17129592.1223266193675--
Nick Wells	CS 286r Nick Wells Placing a bet in a prediction market gives a person incentive to be honest. In the last few years, a number of corporations have set up prediction markets among their employees, notably Google. Researchers drew two main conclusions from their study of Google 1.) Markets were accurate but overly optimistic regarding favorable outcomes; and 2.) opinions were correlated where employees were proximate with one another. The general optimism of employees may be tied to the notion of the “entrepreneur’s curse” where the entrepreneur is the most optimistic person about a project. In this case the employees maintain the curse and researchers also note a marked correlation with stock performance. Proximity of employees mattered, whether physical, social or other. Workers that are connected to each other hold higher correlation in their bets.
Nick Wells	CS 286r Nick Wells Futures markets are places where people are allowed to bet on the outcome of future events. The example given is that of a presidential election. Bets are placed on the outcome of the election, giving a monetary incentive for people to give an accurate prediction. To evaluate the market’s accuracy, researchers compare the market prediction at midnight before the election and the outcome. They found that the markets were fairly accurate predictors. Markets with high a higher volume of trades typically were much more accurate. Higher-profile events such as presidential elections were more accurate than lower profile ones. The authors also described areas of possible further research, such as how prediction markets are affected by events that should have an impact on the campaign, for example “surprise” events vs. events in general.
Hao-Yuh Su	------=_Part_31484_2095691.1223268188246 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline l This paper states market mechanism in an election, the reliability of polls and furthermore, the psychological biases in trading activities. l I think this paper is important. It shows that the market is efficient by solid empirical results. Using future market to test the efficiency of the market and probe the mechanism in an election is quite a clever technique. l A company can use the same technique on evaluating their products and strategies. l I think the future market in this paper is similar to the extensive form game with perfect information we have learned from this class. With good communication of information (perfect information), the game (future market) has an equilibrium, where each player (trader) chose his own best response. -- Hao-Yuh Su Master student in Engineering Science School of Engineering and Applied Sciences Harvard University Tel: 857-869-0104 E-mail: hysu@fas.harvard.edu ------=_Part_31484_2095691.1223268188246 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline l This paper states market mechanism in an election, the reliability of polls and furthermore, the psychological biases in trading activities. l I think this paper is important. It shows that the market is efficient by solid empirical results. Using future market to test the efficiency of the market and probe the mechanism in an election is quite a clever technique. l A company can use the same technique on evaluating their products and strategies. l I think the future market in this paper is similar to the extensive form game with perfect information we have learned from this class. With good communication of information (perfect information), the game (future market) has an equilibrium, where each player (trader) chose his own best response. -- Hao-Yuh Su Master student in Engineering Science School of Engineering and Applied Sciences Harvard University Tel: 857-869-0104 E-mail: hysu@fas.harvard.edu ------=_Part_31484_2095691.1223268188246--
Hao-Yuh Su	------=_Part_31675_29553807.1223271755911 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline l Using IEM system, this paper analyzes the factors of psychological biases and information transmission between traders. It provided several different groups of data and used statistical measure to derive the vigorous conclusion. l When measuring the transmission of communication, the author made an assumption that like-minded people do not proximate. I think it is overly-simplified, though I cannot deny the explanation the author offered. l In this paper, one point I cannot fully understand is about the Figure 2 and its corresponding conclusion. Why can Figure 2 confirm the favorite bias? Why does it use 2-outcome market and 5-outcome market? l I think the psychological biases can be viewed as a belief in the player's strategy, and it was part of the topics in the previous lectures. -- Hao-Yuh Su Master student in Engineering Science School of Engineering and Applied Sciences Harvard University Tel: 857-869-0104 E-mail: hysu@fas.harvard.edu ------=_Part_31675_29553807.1223271755911 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline l Using IEM system, this paper analyzes the factors of psychological biases and information transmission between traders. It provided several different groups of data and used statistical measure to derive the vigorous conclusion. l When measuring the transmission of communication, the author made an assumption that like-minded people do not proximate. I think it is overly-simplified, though I cannot deny the explanation the author offered. l In this paper, one point I cannot fully understand is about the Figure 2 and its corresponding conclusion. Why can Figure 2 confirm the favorite bias? Why does it use 2-outcome market and 5-outcome market? l I think the psychological biases can be viewed as a belief in the player's strategy, and it was part of the topics in the previous lectures. -- Hao-Yuh Su Master student in Engineering Science School of Engineering and Applied Sciences Harvard University Tel: 857-869-0104 E-mail: hysu@fas.harvard.edu ------=_Part_31675_29553807.1223271755911--
Haoqi Zhang	The main contribution of the paper is in the empirical study of the largest corporate prediction market and the analysis of geographic proximity and length of hire to the participants' behavior in prediction markets. The results are significant because it shows that prediction results and biases can be used to study the employees of the firm and the culture of the firm itself. For example, the relations between optimism and new hires and between optimism and a firm's success is an interesting one that may lead to the design of more effective incentive schemes and happier employers (also e.g., how to raise the optimism of company veterans). The discovery that employees sitting close together are likely to have correlated trading behavior suggests that physical proximity plays a major role in shaping people's views and behaviors (albeit, perhaps about lower-priority subjects). One possible extension along this line of thought is to see what physical proximity brings in terms of creativity, and what aspects of operations particularly depend on physical setup and low cost communication (for example, debugging may not be considered one of them, judging from the success of open source projects). While this wasn't extensively discussed in the paper, I wonder if Google's office moving strategy is to counteract the drying up of ideas that occur once nearby people converge on their beliefs and in a sense becomes less creative. In terms of predicting outcome, I wonder if prediction markets can benefit from having restrictions on market participation, e.g., new hires shouldn't be allowed to participate. I wonder if such restrictions increase the prediction of the market or actually harm it (e.g., less irrational people to profit off).
Haoqi Zhang	The main contribution of this paper is in empirically showing the effectiveness of prediction markets as an aggregator of information to predict outcomes to future events. The results are significant because the margin of error is lower than polls conducted before the week of the election in more than half of the cases and that the prediction market does not require a randomly distributed sample of users (in fact, they don't even have to be voters and are mainly well-educated young males with high income). As a platform, the IEM contains more information than typical financial markets, and can served as a research tool. For the market to operate, they needed a large number of traders and the market price to serve as a sufficient statistic. An interesting observation in the paper is the distinction between the marginal trader and the average trader, where the marginal traders are the experts who are making money off average traders while providing the best predictions. But knowing this, wouldn't the average trader learn over time from their losses and become less biased? It was unclear to me in the paper why there was a large influx of new traders late in the 1996 campaign, and what the implications of that are on such a market. Also, how do prediction markets affect people's actual behavior? As early release of exit polls have had the effect of driving some people to vote or not vote (election called too early), can prediction markets play similar roles?
Peter Blair	------=_Part_49734_28373425.1223260548810 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline CS286: Article Response Peter Blair* Results from a Dozen Years of Election Futures Markets Research* In this article, the writers explain the structure of the Iowa Electronic Markets (IEM). Political securities are traded on the IEM. These securities are priced in a way that reflects the market's expectation concerning the election's outcome e.g. based on which candidate/party win or alternatively based on vote or seat share. The article is clear and well-written. It explains the different types of contracts that exists on the market (unit portfolios and contracts based on vote/seat share) and then gives examples of the markets predictions compare to actual outcomes of elections both in the USA and other countries. The idea of an election market is a clever one -- by coupling a financial pay-off to the client's prediction, the market incentivizes the client giving a true opinion. This feature of the market stands in contrast to the partisan who may be polled by a political operative. The one questions that I have about this article is: 1). What does the actions of "typical traders" tell us about arbitrage opportunities in a market where irrational considerations affect a client's participation in the market. An interesting follow-up research question would be to find ways to identify dominant biases of potential traders, quantify those effects and then come up with a set of bid/ask strategies that would exploit these arbitrage opportunities. One shining example of such an arbitrage opportunity would be the effect of race in this year's presidential elections -- some investors may process the impact of race in this election in a manner that identifies them as a "typical voter," which in turn creates an arbitrage opportunity for the savvy rational advisers. In terms of future research direction,it is evident that prediction markets of this type can be used to track expected stock prices on the Dow and other international exchanges. Imagine clients not buying actual stock in a company, which can be very costly (e.g. few hundred dollar for Google's stock), but rather purchasing contracts that predict the probability that a given stock, or even the Dow Jones Industrial Average rises/falls. Such a market would be a great asset in times like these were there is great uncertainty about how the bailout will affect the long term stability of the global economy. Besides, an actual investor in the NYSE can use information from this predictive market as a reference point for considering how to participate in the market (NYSE). In conclusion this article presents a very cogent argument for the utility of predictive electoral markets such as IEM. -- Peter Blair Physics PhD Candidate Jefferson Laboratory, Harvard University ------=_Part_49734_28373425.1223260548810 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline CS286: Article Response Peter Blair Results from a Dozen Years of Election Futures Markets Research In this article, the writers explain the structure of the Iowa Electronic Markets (IEM). Political securities are traded on the IEM. These securities are priced in a way that reflects the market's expectation concerning the election's outcome e.g. based on which candidate/party win or alternatively based on vote or seat share. The article is clear and well-written. It explains the different types of contracts that exists on the market (unit portfolios and contracts based on vote/seat share) and then gives examples of the markets predictions compare to actual outcomes of elections both in the USA and other countries. The idea of an election market is a clever one -- by coupling a financial pay-off to the client's prediction, the market incentivizes the client giving a true opinion. This feature of the market stands in contrast to the partisan who may be polled by a political operative. The one questions that I have about this article is: 1). What does the actions of "typical traders" tell us about arbitrage opportunities in a market where irrational considerations affect a client's participation in the market. An interesting follow-up research question would be to find ways to identify dominant biases of potential traders, quantify those effects and then come up with a set of bid/ask strategies that would exploit these arbitrage opportunities. One shining example of such an arbitrage opportunity would be the effect of race in this year's presidential elections -- some investors may process the impact of race in this election in a manner that identifies them as a "typical voter," which in turn creates an arbitrage opportunity for the savvy rational advisers. In terms of future research direction,it is evident that prediction markets of this type can be used to track expected stock prices on the Dow and other international exchanges. Imagine clients not buying actual stock in a company, which can be very costly (e.g. few hundred dollar for Google's stock), but rather purchasing contracts that predict the probability that a given stock, or even the Dow Jones Industrial Average rises/falls. Such a market would be a great asset in times like these were there is great uncertainty about how the bailout will affect the long term stability of the global economy. Besides, an actual investor in the NYSE can use information from this predictive market as a reference point for considering how to participate in the market (NYSE). In conclusion this article presents a very cogent argument for the utility of predictive electoral markets such as IEM. -- Peter Blair Physics PhD Candidate Jefferson Laboratory, Harvard University ------=_Part_49734_28373425.1223260548810--
Peter Blair	------=_Part_57028_9190208.1223249017712 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline CS286r Article Review Using Prediction Markets to Track Information Flows: Evidence from Google by: Peter Blair In "Using Prediction Markets to Track Information Flows: Evidence from Google", Cowgil Wolfers and Zitzewitz study correlations among sercuirties purchasing decisions of Google employees participating in the company's internal exchange market. Google's internal exchange market mimics the setup of a traditional financial market in which there are buyers and sellers interacting via a common currency "Goobles". The authors report on the effects of spacial separation of employees (mostly "quants, so hardly a random sample), membership to certain groups ("demographic similarity" e.g. professional associations and email lists) and other social ties, on correlations with their choice of market portfolios. Two central results are that neighboring employees have highly correlated purchases and secondly that employees are very optimistic about the performance of securities linked to Google's business operations, e.g. how many new users will Gmail attract this month (particularly on days when Google's real life stock was up). The authors claim that their findings "contribute to three quite different literatures: on the role of optimisms in entrepreneurial firms, on employee communication in organization, and on social networks and information flows among investors." My main criticism of this paper is that there seem to be two papers in the one paper. The first paper would address the important problem of biases in the data arising from "over pricing of favorites, short aversion, optimism and underpricing of extreme outcomes" (pg 5). The second potential would be one discussion the correlations in the portfolio choices of employees based on demographic similarity and proximity. A separation of this single paper into these two papers mentioned above would be suitable in that the first paper would contribute directly to the literature on the role of optimism in entrepreneurial firms, while the second paper would contribute to the literature on employee communication in organization, and on social networks and information flows among investors. My second criticism of this paper is that the correlations arising from spatial closeness is loosely discussed. Given the author's access to specific longitudinal and latitudinal coordinates of the employees in the study, they could potential do a plot of correlated security purchase as a function of distance. As a reader, I would be interested to see if such a correlation would be isotropic, for example, or if there are preferred directions of communication, perhaps selected based on some natural/physical structure e.g. a scenic window view, proximity to the vending machines or bathrooms or employee lounge etc. Notwithstanding this gentle criticism, the importance of spatial closeness is a surprising result, particularly given the minimal cost of communicating via alternative means e.g. IM, facebook, cell phones, e-mail -- a point that the authors make very successfully in the article. Further on the point of spatial separation, the reader is interested to see how this model of information flow would be mapped onto a larger open system of investors -- think a group of home business owners who are also investors. Otherwise put, is the spatial correlation in employees choices only significant because it exist within the context of an umbrella organization that unifies the employees, or does it apply to random groups of unrelated people working on similar questions? This is an important research question that the results of this paper motivate. My third and final criticism relates to the homogeneity of the sample group. On the one hand, it is interesting that correlations owing to demographic similarities are relatively weak given that we have an ostensibly homogeneous demographic of "quants" and computer scientists/engineers -- it is certainly reasonable to expect that this might not be the case. On the other hand, it could be that this homogeneity washes out many correlations that would be based on demographic similarity (think race, gender, age, marital status) since these share characteristics would simply provide a background (as in constant baseline) for the signal of correlations based on other factors that are not controlled for by sampling from a homogeneous demographic. This article is strong in its novel attempt to study a closed market system and from its study propose stylized facts about social interactions, information flow and investor decisions. The broad scope of issues and effects suggested, realized and explained provides the reader with lots of follow-up questions that are useful for contributing the the three litteratures mentioned in the article. This great strength of the article -- it's breadth, is also one of its weak points: because it packs in so many contributions to the respective litteratures, it comes across as unfocused on a first read, when in fact its filled with many gems that sow seeds for future research projects on information flow in investing markets. -- Peter Blair Physics PhD Candidate Jefferson Laboratory, Harvard University ------=_Part_57028_9190208.1223249017712 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline CS286r Article Review Using Prediction Markets to Track Information Flows: Evidence from Google by: Peter Blair In "Using Prediction Markets to Track Information Flows: Evidence from Google", Cowgil Wolfers and Zitzewitz study correlations among sercuirties purchasing decisions of Google employees participating in the company's internal exchange market. Google's internal exchange market mimics the setup of a traditional financial market in which there are buyers and sellers interacting via a common currency "Goobles". The authors report on the effects of spacial separation of employees (mostly "quants, so hardly a random sample), membership to certain groups ("demographic similarity" e.g. professional associations and email lists) and other social ties, on correlations with their choice of market portfolios. Two central results are that neighboring employees have highly correlated purchases and secondly that employees are very optimistic about the performance of securities linked to Google's business operations, e.g. how many new users will Gmail attract this month (particularly on days when Google's real life stock was up). The authors claim that their findings "contribute to three quite different literatures: on the role of optimisms in entrepreneurial firms, on employee communication in organization, and on social networks and information flows among investors." My main criticism of this paper is that there seem to be two papers in the one paper. The first paper would address the important problem of biases in the data arising from "over pricing of favorites, short aversion, optimism and underpricing of extreme outcomes" (pg 5). The second potential would be one discussion the correlations in the portfolio choices of employees based on demographic similarity and proximity. A separation of this single paper into these two papers mentioned above would be suitable in that the first paper would contribute directly to the literature on the role of optimism in entrepreneurial firms, while the second paper would contribute to the literature on employee communication in organization, and on social networks and information flows among investors. My second criticism of this paper is that the correlations arising from spatial closeness is loosely discussed. Given the author's access to specific longitudinal and latitudinal coordinates of the employees in the study, they could potential do a plot of correlated security purchase as a function of distance. As a reader, I would be interested to see if such a correlation would be isotropic, for example, or if there are preferred directions of communication, perhaps selected based on some natural/physical structure e.g. a scenic window view, proximity to the vending machines or bathrooms or employee lounge etc. Notwithstanding this gentle criticism, the importance of spatial closeness is a surprising result, particularly given the minimal cost of communicating via alternative means e.g. IM, facebook, cell phones, e-mail -- a point that the authors make very successfully in the article. Further on the point of spatial separation, the reader is interested to see how this model of information flow would be mapped onto a larger open system of investors -- think a group of home business owners who are also investors. Otherwise put, is the spatial correlation in employees choices only significant because it exist within the context of an umbrella organization that unifies the employees, or does it apply to random groups of unrelated people working on similar questions? This is an important research question that the results of this paper motivate. My third and final criticism relates to the homogeneity of the sample group. On the one hand, it is interesting that correlations owing to demographic similarities are relatively weak given that we have an ostensibly homogeneous demographic of "quants" and computer scientists/engineers -- it is certainly reasonable to expect that this might not be the case. On the other hand, it could be that this homogeneity washes out many correlations that would be based on demographic similarity (think race, gender, age, marital status) since these share characteristics would simply provide a background (as in constant baseline) for the signal of correlations based on other factors that are not controlled for by sampling from a homogeneous demographic. This article is strong in its novel attempt to study a closed market system and from its study propose stylized facts about social interactions, information flow and investor decisions. The broad scope of issues and effects suggested, realized and explained provides the reader with lots of follow-up questions that are useful for contributing the the three litteratures mentioned in the article. This great strength of the article -- it's breadth, is also one of its weak points: because it packs in so many contributions to the respective litteratures, it comes across as unfocused on a first read, when in fact its filled with many gems that sow seeds for future research projects on information flow in investing markets. -- Peter Blair Physics PhD Candidate Jefferson Laboratory, Harvard University ------=_Part_57028_9190208.1223249017712--
Rory Kulz	I found this paper to be a fascinating read. It discusses Google's experimentation with internal prediction markets on a variety of Google-related and non-Google-related topics. While they discuss a so-called "optimistic bias" quite a bit (and hint towards its importance in both assimilation into a social structure and the functioning of an entrepreneurial endeavor), the focus really appears to be on two other items. Primarily, the researchers seem interested in using information about how the market functions to model information flows within the organizational structure of Google itself. This yields a number of unexpected insights. For one, that the influence of the topology of the social network is minimal and that of demographics even less so. For another, that physical proximity alone, divorced from things like friendship or work groups that might result in physical proximity, may cause like-mindedness on predictions. The reason the authors give for this latter insight, in the conclusion, is particularly neat: "...if prediction market topics are lower-priority subjects on which to exchange information, then information exchange may require the opportunities for low-opportunity-cost communication created by physical proximity." Secondarily, the researchers investigate a number of properties of the market itself. While they don't conduct a direct comparison with the Iowa Electronic Markets, having read the IEM paper, I can see a number of analogous ideas. For example, the Google researchers talk about how new employees tend to suffer from certain biases -- optimism, short aversion, overpricing of favorites -- whereas long-term employees do not (although the experienced traders do tend to underprice extreme alternatives in the 5-security markets). I found this to be very similar to the IEM paper's discussion of marginal traders versus typical traders. Specifically, the people who make the market are more experienced and so (a.) know what sort of moves to make and (b.) are not as influenced by their own desired outcome. I was a little intrigued by the paragraph which runs on pages 6--7, where they state that theoretical analysis and experimental evidence suggest that Google should not be witnessing the direction of favorite bias observed. They offer that one "possibility is that the favorite bias in prices reflects a larger favorite bias in the beliefs of the median trader." That seems a little obscure or opaque an explanation. I wondered if this is more simply explainable by the fact that the traders on Google's prediction markets form a well-defined social network (with, e.g., a "small-world" property), whereas this is not the case for, say, Intrade or IEM. In particular, everyone at Google has some incentive towards not only Google doing well but also seeing their colleagues do well (i.e., a "magnanimity" factor).
Rory Kulz	This paper is a report on the Iowa Electronic Markets (IEM) which gives both a description of how they function -- explaining share and winner-take-all contracts, describing the typical investor and available data, etc. -- and a survey of the research that has been conducted using IEM data. This survey also presents some rudimentary statistical results that try to convince the reader that the IEM are an excellent testbed for research. There are three main components to this process of convincing. First, the authors compare the IEM to the alternatives. They argue that the IEM are larger than the typical experimental markets found in "laboratories," and hence they allow the study of real problems rather than toy problems. Furthermore, they argue that the IEM can provide more specific and interesting data than real-world markets typically allow, particularly demographic profiles and portfolio positions of the individual investors. Second, they examine how IEM predictions have performed relative to polls. In some cases, like the 1988 U.S. presidential election, the difference is dramatic. Overall, though, even if the market doesn't serve as a much better predictor in terms of absolute error, such as occured in the '96 U.S. election, it certainly seems no worse, and better still, reasonable evidence has been found that the market-based prediction is more robust and less volatile over time compared to polls (this is illustrated in their Figure 4). Finally, at the end, they offer up many avenues of investigation that have been explored and could be explored, demonstrating the potential for IEM research. I for one was convinced that the IEM, besides being interesting in their own right, could be very useful for some specific sorts of questions, especially related to behavioral finance and investigating market psychology, although I am not fully convinced that such results would transfer wholesale over to more liquid markets such as, say, NASDAQ or currency.
Sagar Mehta	------=_Part_32764_21384114.1223248328360 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline This paper provides empirical data testing the accuracy of prediction markets in predicting political events. The types of contracts discussed are such that a player is rewarded by playing according to the true nature of his belief. Prediction market data compares favorably to polling data both in terms of accuracy on the actual day of the vote and stability over time (i.e. less volatility compared to polling data). One interesting thing to note about the nature of prediction markets is that market makers (marginal traders) drive prices and therefore predictions. This seems to run somewhat counter to the notion of prediction markets as aggregators of information from many sources as traders who play most frequently will have their information "count" more. I'd be interested in knowing how the number of traders and frequency at which they trade impacts the accuracy of prediction markets. In comparing prediction markets to polls, the authors don't seem to go into detail about how polls are fundamentally different than prediction markets. Polls ask an individual "Who would you vote for?" not "Who do you think is most likely to win, and with what probability?" Thus, while prediction markets seem to beat traditional polls in terms of getting more accurate predictions, what about modified polls? Do prediction markets outperform those? Another general question about prediction markets: How do play money markets perform vs. real money markets? ------=_Part_32764_21384114.1223248328360 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline This paper provides empirical data testing the accuracy of prediction markets in predicting political events. The types of contracts discussed are such that a player is rewarded by playing according to the true nature of his belief. Prediction market data compares favorably to polling data both in terms of accuracy on the actual day of the vote and stability over time (i.e. less volatility compared to polling data). One interesting thing to note about the nature of prediction markets is that market makers (marginal traders) drive prices and therefore predictions. This seems to run somewhat counter to the notion of prediction markets as aggregators of information from many sources as traders who play most frequently will have their information "count" more. I'd be interested in knowing how the number of traders and frequency at which they trade impacts the accuracy of prediction markets. In comparing prediction markets to polls, the authors don't seem to go into detail about how polls are fundamentally different than prediction markets. Polls ask an individual "Who would you vote for?" not "Who do you think is most likely to win, and with what probability?" Thus, while prediction markets seem to beat traditional polls in terms of getting more accurate predictions, what about modified polls? Do prediction markets outperform those? Another general question about prediction markets: How do play money markets perform vs. real money markets? ------=_Part_32764_21384114.1223248328360--
Sagar Mehta	------=_Part_32766_9710727.1223248358697 Content-Type: text/plain; charset=WINDOWS-1252 Content-Transfer-Encoding: quoted-printable Content-Disposition: inline The focus on this paper was to describe not only how well internal prediction markets work in predicting future events, but also on how they can be used as a tool to better understand information flow within a firm. The paper found that traders in the same location (i.e. sharing a cubicle o= r office) tend to have strong positive correlation in prediction market positions. This was found to be more than just a case of like-minded people being located in proximate positions since Google has people rotate desks often. Sitting on the same floor has little effect of trading behavior, nor does being friends with an individual. The importance of proximity is somewhat surprising to me despite the fact that chat clients, email, and telephone can effectively diminish the necessity of being next to someone t= o share information. There is one important caveat to the assumption that everyone needs to be sitting on top of each other for optimal information flow, however. The questions posed in prediction market contracts were not necessarily related to employees' main jobs. For instance, if I am a coder working on optimizing an algorithm =96 how much more likely am I to ask the coder sitting next to me for help than sending my code over to someone else to take a look? Intuitively, it seems that the first is more likely =96 but the paper does not answer this question. If the cost of setting up a meeting, emailing, or chatting is in fact much more difficult and close proximity is the key to information flow =96 why is it that professors at Harvard work in offices rather than in cubicles? Another interesting finding of the paper was that there is a significant optimistic bias among Google employees. This raises some interesting social questions. Does optimism about a company's prospects play a role in increasing worker productivity? Rephrased, how does a declining stock price impact employee moral and worker productivity? ------=_Part_32766_9710727.1223248358697 Content-Type: text/html; charset=WINDOWS-1252 Content-Transfer-Encoding: quoted-printable Content-Disposition: inline The focus on this pape= r was to describe not only how well internal prediction markets work in predicting future events, but also on how they can be used as a tool to better underst= and information flow within a firm. The paper found that traders in the same location (i.e. sharing a cubicle or office) tend to have strong positive correlation in prediction market positions. This was found to be more than = just a case of like-minded people being located in proximate positions since Goo= gle has people rotate desks often. Sitting on the same floor has little effect = of trading behavior, nor does being friends with an individual. The importance= of proximity is somewhat surprising to me despite the fact that chat clients, email, and telephone can effectively diminish the necessity of being next t= o someone to share information. There is one important caveat to the assumpti= on that everyone needs to be sitting on top of each other for optimal informat= ion flow, however. The questions posed in prediction market contracts were not necessarily related to employees' main jobs. For instance, if I am a coder working on optimizing an algorithm =96 how much more likely am I to ask the= coder sitting next to me for help than sending my code over to someone else to ta= ke a look? Intuitively, it seems that the first is more likely =96 but the paper= does not answer this question. If the cost of setting up a meeting, emailing, or chatting is in fact much more difficult and close proximity is the key to information flow =96 why is it that professors at Harvard work in offices r= ather than in cubicles? Another interesting fi= nding of the paper was that there is a significant optimistic bias among Google employee= s. This raises some interesting social questions. Does optimism about a compan= y's prospects play a role in increasing worker productivity? Rephrased, how doe= s a declining stock price impact employee moral and worker productivity? ------=_Part_32766_9710727.1223248358697--
Subhash Arja	------=_Part_22494_11813128.1223156488073 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline This paper's main objective is to show how the Iowa Electronic Markets (IEM) can be used to predict election results. The IEM is a real-money futures market conducted by University of Iowa College of Business and is touted to predict election results in many different markets in various countries. This concept and its implementation are important and interesting because, if precise, they can accurately predict a variety of election result data types, such as the margin of victory and anticipated election effects based on news. The market itself doesn't model the actual election since the participants are well educated, male, and high-income. The market mechanism is very similar to financial markets in the sense that traders can see the current market price, set an asking bid, and a selling bid. From what we have seen in class about rationality and irrationality, the paper states that typical traders, or those that have hopeful beliefs for their preferred candidates, tell make irrational mistakes. This is because they decrease their payoff values regardless of the outcomes. Marginal traders who regularly trade drive the predictions. The future work that can be drawn from this study involves applying this concept not only to elections but also to all zero sum events. Examples include sports and various tournaments. One project is to apply prediction markets to a specific sport such as basketball and predict the winner of the NBA championship over the course of the season. As the paper states, the results of the prediction markets are very accurate when taken over the whole season rather than over the few weeks towards the end. -Subhash Arja ------=_Part_22494_11813128.1223156488073 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline This paper's main objective is to show how the Iowa Electronic Markets (IEM) can be used to predict election results. The IEM is a real-money futures market conducted by University of Iowa College of Business and is touted to predict election results in many different markets in various countries. This concept and its implementation are important and interesting because, if precise, they can accurately predict a variety of election result data types, such as the margin of victory and anticipated election effects based on news. The market itself doesn't model the actual election since the participants are well educated, male, and high-income. The market mechanism is very similar to financial markets in the sense that traders can see the current market price, set an asking bid, and a selling bid. From what we have seen in class about rationality and irrationality, the paper states that typical traders, or those that have hopeful beliefs for their preferred candidates, tell make irrational mistakes. This is because they decrease their payoff values regardless of the outcomes. Marginal traders who regularly trade drive the predictions. The future work that can be drawn from this study involves applying this concept not only to elections but also to all zero sum events. Examples include sports and various tournaments. One project is to apply prediction markets to a specific sport such as basketball and predict the winner of the NBA championship over the course of the season. As the paper states, the results of the prediction markets are very accurate when taken over the whole season rather than over the few weeks towards the end. -Subhash Arja ------=_Part_22494_11813128.1223156488073--
Subhash Arja	------=_Part_28782_21458473.1223255987530 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline The main purpose of this paper is to show how the concept of prediction markets can be applied to studying how a large company, Google, processing information. Also, it shows how events that have no direct effect on the company's or employee's performance are modeled in the prediction markets. This study is important because it takes the concept which was explained in the Iowa Electronic Markets and applies them in a corporate setting rather than predicting elections. Such application, if extended further, has the potential to help the company find and make improvements to help the employees work more efficiently. This will, as a result, lead to higher productivity for the company and more profits. As was stated in the previous IEM paper, the main insight is to treat an event like a company event or movie premiere as a market. In this case, the market asks a question that has 2-5 possible mutually exclusive outcomes. The study makes two assumptions. First, it is assumed that two like-minded people will not be physically close to each other. Second, observed measures of proximity are not correlated with unobserved proximity. In my opinion, both of these assumptions are valid because they do not constrict the possibilities but lay a foundation on which to base the study. This paper directly relates to the Iowa Election Market paper and even references the work done there. One project, which is an extension of this paper, is to apply a prediction market between two competing companies. This could potentially give insight into whether the employees' mindset is different and what factors lead to one company eventually beating out the other. -Subhash Arja ------=_Part_28782_21458473.1223255987530 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline The main purpose of this paper is to show how the concept of prediction markets can be applied to studying how a large company, Google, processing information. Also, it shows how events that have no direct effect on the company's or employee's performance are modeled in the prediction markets. This study is important because it takes the concept which was explained in the Iowa Electronic Markets and applies them in a corporate setting rather than predicting elections. Such application, if extended further, has the potential to help the company find and make improvements to help the employees work more efficiently. This will, as a result, lead to higher productivity for the company and more profits. As was stated in the previous IEM paper, the main insight is to treat an event like a company event or movie premiere as a market. In this case, the market asks a question that has 2-5 possible mutually exclusive outcomes. The study makes two assumptions. First, it is assumed that two like-minded people will not be physically close to each other. Second, observed measures of proximity are not correlated with unobserved proximity. In my opinion, both of these assumptions are valid because they do not constrict the possibilities but lay a foundation on which to base the study. This paper directly relates to the Iowa Election Market paper and even references the work done there. One project, which is an extension of this paper, is to apply a prediction market between two competing companies. This could potentially give insight into whether the employees' mindset is different and what factors lead to one company eventually beating out the other. -Subhash Arja ------=_Part_28782_21458473.1223255987530--
Victor Chan	Victor Chan (I accidentally sent an extra copy from my gmail) Paper Coments:Results from a Dozen Years of Election Futures Markets The main contribution of the paper is presenting the results of prediction markets of the past few years in an aggregated format. This review is important because it gathers together the findings of other papers and tries to justify the value of prediction markets as an accurate method of predicting event outcomes. The limitation of this paper is that it talks about prediction market results and trades/market makers that make it accurate, however these conclusions are drawn using various results from different papers. As a result, there does not appear to be a single point that the author is trying to make, the topics range from the accuracy of the prediction markets, to the types of investors that make it accurate. Perhaps more detailed examination of each claim is required. However reading the cited papers should clear up these questions. The main insight of the paper is that prediction markets can offer accurate long term results. This is however subject to the quality of the investors. Investors who are "market makers" and less likely affecteb by their personal choice in an outcome will make accurate predictions. The paper also talks about the accuracy of the prediction markets against traditional polls, and discusses how it out performs the polls. After reading this paper, one can assume that using prediction market results as an indicator of how well candidates are doing in a election, might offer better insights into the outcome than polls. Furthermore, the current predictions of the market are based on the price the evening before the election and the average price a week ahead of the election. Using these sets of data, the prediction markets have been deemed accurate. A possible project would be to analyze prices further back in time, and to determine a starting point at which the markets will present accurate predictions.
Victor Chan	Victor Chan Paper Review: Using Prediction Markets to Track Information Flows: Evidence from Google. The main contribution of the paper by Cowgill et al, is that it analyzes one of the largest coporate run prediction markets. The main findings show that forecasting demand and internal performance type markets tend to have a bias towards the optimistic outcome. This is explained in the paper as a result of the an enterprenuerial enviornment, and newly hired employees giving way to overpricing the optimist outcome. Furthermore the paper also discusses the correlation of the traders behaviour. First, when they are close to each other in the office, they are more likely to trade similarily. This is explained as a result of lower cost information flow. Next, being within in the same team, or working on the same project also tended to cause people to trade in a similar fashion. The influence of others, is also seen in that traders tend to behave more like their "professional" contacts, rather than their friends. Out of the three, it is determined that physical proximity demonstrates the highest correlation. The limitation of the paper is that Google's prediction markets were mostly participated by engineers and not a diverse group of traders. This could affect the accuracy of the results, since other types of employees might not have a high correlation in trading even if they are sitting close to each other. Another limitation is that the prediction markets are a past-time or secondary to the engineer's primary role at the company, therefore some might not have taken trading as seriously. (ie, when their own money is being invested). The main insight from the results is that although, there was bias in company related markets, the predictions were fairly accurate for events that were not related to Google. This reaffirms that prediction markets can be used to determine event outcomes fairly accurately. Another interesting point is suggested in the article about the value of proximity if direct talking was not the lowest cost communication method. The results could change, if emailing was a cheaper method of information flow. To extend the results, the authors can look at the correlation between people who the have lunch with eachother and their trading habits. In a company like Google, most socializing would be done during lunch, and it is at these hours that employees will most likely discuss prediction market related topics. It is possible that the results will show that having lunch together will be even more compelling than being within a close proximity. If this were to be true, then it is the social interaction that plays the biggest role in correlation bewteen traders. The results of the study could be applied to prediction markets and looking at the geography of the trades being made. This would be especially interesting if applied to the US elections, where one could determine a buy/sell correlation with a win/lose outome for a candidate in a specific geographic region.
Xiaolu Yu	------=_Part_59695_14561128.1223255526003 Content-Type: multipart/alternative; boundary="----=_Part_59696_5966689.1223255526003" ------=_Part_59696_5966689.1223255526003 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline Xiaolu Yu This paper analyzes the largest experiment with prediction markets. Main conclusions and contributions: 1. Corporate prediction markets reveal some biases, but may perform better as collective experience increases. a. The finding of a favorite bias in Google's markets (overpricing of favorites and underpricing of longshots) contradicts with Ali and Manski's results. The authors suggest a possible reason as the belief bias of the median trader. b. Short selling aversion creates arbitrage opportunity. c. The market traders have extreme aversion. d. Optimistic bias exists in two categories of contracts where outcomes are most directly under the control of Google employees: company news and performance. The optimistic is larger in two outcome markets, early in sample period, and earlier in each quarter. e. Number of shares transacted and sample period matters to the magnitude of biases. f. Stock price variation induced mainly by optimistic bias, extreme aversion and favorite bias actually reflect ex post surprise. (Some thoughts: the economy meltdown lately in the U.S. causes similar reactions of the market: more disappointing, more unconfident, more extreme, and harder to predict than rational traders anticipated.) g. Individual behaviors in driving prediction market biases: newly hired employees are more likely to take advantage of the reverse favorite-longshot and short aversion bias in prices, though trade optimistically; more experienced traders profit from optimism, favorite, and short aversion biases, but contributes to extreme aversion. 2. Proximity importantly matters in terms of the correlation between opinions on specific topics among employees. Physical proximity is the most important, whereas demographic does not affect information sharing significantly. Applications: 1. Entrepreneurs are optimistically biased. 2. Considerable thought is put into optimizing physical locations in order to encourage communication between managers and workers, and among peers, which in turn facilitate information flow. 3. Physical proximity and social network are important in information sharing among investors. How valid to 'using prediction market to track information flow'? Assumptions: 1. Like-minded people are not proximate (proved by examing sharp changes in proximity). Information sharing forming more rapidly when employees share and office than when they are not, and when they switch offices frequently than when they are not. 2. Observed measures of proximity are not correlated with unobserved proximity. (Limitations: each employee name ten colleagues that they turned to for ideas and advice. However, the ten may not be those from whom he got the most information from, but those who were consulted by him most recently and hence he remember most clearly. Also, two employees have to name each other simultaneously in order to build the professional relationship. This process is sort of random.) Caveat: The prediction market subjects of information flow are ancillary to employees' main job. This caveat may explains the significance of physical proximity, but still assumably meaningful in better understanding product innovation and other creative processes. ------=_Part_59696_5966689.1223255526003 Content-Type: text/html; charset=ISO-2022-JP Content-Transfer-Encoding: 7bit Content-Disposition: inline Xiaolu Yu This paper analyzes the largest experiment with prediction markets. Main conclusions and contributions: 1. Corporate prediction markets reveal some biases, but may perform better as collective experience increases. a. The finding of a favorite bias in Google's markets (overpricing of favorites and underpricing of longshots) contradicts with Ali and Manski's results. The authors suggest a possible reason as the belief bias of the median trader. b. Short selling aversion creates arbitrage opportunity. c. The market traders have extreme aversion. d. Optimistic bias exists in two categories of contracts where outcomes are most directly under the control of Google employees: company news and performance. The optimistic is larger in two outcome markets, early in sample period, and earlier in each quarter. e. Number of shares transacted and sample period matters to the magnitude of biases. f. Stock price variation induced mainly by optimistic bias, extreme aversion and favorite bias actually reflect ex post surprise. (Some thoughts: the economy meltdown lately in the U.S. causes similar reactions of the market: more disappointing, more unconfident, more extreme, and harder to predict than rational traders anticipated.) g. Individual behaviors in driving prediction market biases: newly hired employees are more likely to take advantage of the reverse favorite-longshot and short aversion bias in prices, though trade optimistically; more experienced traders profit from optimism, favorite, and short aversion biases, but contributes to extreme aversion. 2. Proximity importantly matters in terms of the correlation between opinions on specific topics among employees. Physical proximity is the most important, whereas demographic does not affect information sharing significantly. Applications: 1. Entrepreneurs are optimistically biased. 2. Considerable thought is put into optimizing physical locations in order to encourage communication between managers and workers, and among peers, which in turn facilitate information flow. 3. Physical proximity and social network are important in information sharing among investors. How valid to 'using prediction market to track information flow'? Assumptions: 1. Like-minded people are not proximate (proved by examing sharp changes in proximity). Information sharing forming more rapidly when employees share and office than when they are not, and when they switch offices frequently than when they are not. 2. Observed measures of proximity are not correlated with unobserved proximity. (Limitations: each employee name ten colleagues that they turned to for ideas and advice. However, the ten may not be those from whom he got the most information from, but those who were consulted by him most recently and hence he remember most clearly. Also, two employees have to name each other simultaneously in order to build the professional relationship. This process is sort of random.) Caveat: The prediction market subjects of information flow are ancillary to employees' main job. This caveat may explains the significance of physical proximity, but still assumably meaningful in better understanding product innovation and other creative processes.
Xiaolu Yu	------=_Part_59819_11172435.1223257192139 Content-Type: multipart/alternative; boundary="----=_Part_59820_6059759.1223257192139" ------=_Part_59820_6059759.1223257192139 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline Xiaolu Yu One of the best known markets of IEM, the Iowa Political Markets shows considerable accuracy in the market forecasts, especially for large, U.S. election markets. The variance in accuracy may be explained by three factors. 1. Presidential election markets perform better than congressional, state and local election markets (larger, more active markets with fewer contracts are more accurate). 2. Markets with more volume near the election perform better than those with less (the marginal traders are much less prone to bias because of information update along the progression of the game.) 3. Markets with fewer contracts predict better than those with more. Comparing to polls, over the majority of the time the market ran, the predictions were dramatically more accurate and stable, and hence more valuable as long run forecasting devices. In order for the market to work, there must be enough traders so that the aggregate of their knowledge can correctly forecast the outcome of the election. Meanwhile, the market mechanism must facilitate aggregation of their disparate information so that the prevailing market price becomes a sufficient statistic for their collective information. Core group traders/"marginal traders"/"market makers" that tend to set market prices appears less biased and error prone than typical traders, and make mistakes much less often. IEM have great research potential. For example, Google's prediction markets are patterned on the Iowa Electronic Markets. ------=_Part_59820_6059759.1223257192139 Content-Type: text/html; charset=ISO-2022-JP Content-Transfer-Encoding: 7bit Content-Disposition: inline Xiaolu Yu One of the best known markets of IEM, the Iowa Political Markets shows considerable accuracy in the market forecasts, especially for large, U.S. election markets. The variance in accuracy may be explained by three factors. 1. Presidential election markets perform better than congressional, state and local election markets (larger, more active markets with fewer contracts are more accurate). 2. Markets with more volume near the election perform better than those with less (the marginal traders are much less prone to bias because of information update along the progression of the game.) 3. Markets with fewer contracts predict better than those with more. Comparing to polls, over the majority of the time the market ran, the predictions were dramatically more accurate and stable, and hence more valuable as long run forecasting devices. In order for the market to work, there must be enough traders so that the aggregate of their knowledge can correctly forecast the outcome of the election. Meanwhile, the market mechanism must facilitate aggregation of their disparate information so that the prevailing market price becomes a sufficient statistic for their collective information. Core group traders/"marginal traders"/"market makers" that tend to set market prices appears less biased and error prone than typical traders, and make mistakes much less often. IEM have great research potential. For example, Google's prediction markets are patterned on the Iowa Electronic Markets.
Ziyad Aljarboua	This paper discuses a different kind of prediction market: The Iowa Political Markets where prices reflect election outcomes. This kind of market capitalize on the aggregate knowledge of traders to correctly forecasts the outcome of an election. This way can minimize error and bias of individual traders by averaging out the prediction over all traders. Unlike financial markets, Iowa market operates all day and investments are limited to $500. This paper examines accuracy of prices in these markets and compare results of national elections to polls. Unlike polls, traders in the Iowa markets tie their answer and prediction to a financial reward or loss. In this case, unlike polls, they have a incentive to know how other voters think and predict the winers and invest in the market accordingly. Whereas in the polls, voters should only consult their preference. In the Iowa markets, traders purchase and sell contracts from the system for a fee based on current best bid and ask prices. The main purpose of this paper is evaluating efficiency of this kind of market. the efficiency is determined by comparing market predictions at midnight the evening before the election to the actual election outcome. The analysis of the market data shows no obvious biases. It shows that the market on average was accurate for large markets with many traders. Data showed that high profile elections such as the presidential election markets performed better that lower profile ones. This can be attributed to the fact that high profile markets have larger volume. Also, it was observed that the number of contracts (candidates in election) affects the accuracy too. Markets with fewer contracts perform better. When comparing market data to polls, it was found out that market prediction outperformed polls in the some of the elections and had a smaller absolute error. But on average, polls yield same accuracy as prediction markets. Some limitation of this market is the fact that the majority of traders are male, educated, with high income and of young age. This is not representative of all voters and thus is not likely to predict true winner of election. Also, accuracy of this market is limited by number of traders and size of market. It was shown that large markets such as the US had a higher accuracy. The example of the 1996 US presidential election shows a serious shortcoming of the political prediction market. In this example, the accuracy of the market was greatly affected by the influx of cash towards the end of the election period. This shows that outside factors can affect the effectiveness and accuracy of such prediction markets unlike polls. Finally, accuracy of this market is completely dependent on number of traders. So, as can be seen, such prediction market can be easily influenced and diverted. --Ziyad Aljarboua
Ziyad Aljarboua	This paper analysis Google's prediction markets which is modeled after Iowa Electronic Markets. In Google' prediction market, qualified google's affiliates receive endowment (goobles) that they can use to buy securities. Trading in the market is simply a double action in the securities. Google asks questions through the market ranging from forecasting demand to internal performance and fun questions that have no relation to google's strategic business. Participents can invest in these securities. at the end of each quarter, Goobles are converted into raffle tickets. basic analysis of this experiment concluded that participant, compared to non participants, are more likely to be: -programmers. -based in googl's mountain view and new york campuses. -long time employed. -more deeply embedded in the organization. -more senior. in this paper, authors argue that internal prediction can provide an insight into how organization process information. In Google's experiment, while the author concluded that the market is efficient, they noticed some bias in the internal markets. They noticed four kinds of biases: optimism, extreme aversion, favorite bias. One factor in this bias was attributed to the trading of newly hired employees who tend to overprice favorites and underprice securities with optimistic outcomes. Also, participants often fail to predict extreme outcomes. Generally, "performance" and "company News" categories are usually optimistically biased on average. Next, testing correlation between trading habits and physical proximity of employees. A strong correlation between location of employee's offices and prediction market positions was observed. This conclusion was inforced by the results they obtained after the studied subjects have moved their offices and correlation declined. I think this conclusion is very obvious as close approximate in location implies exchange of ideas which leads to similar thinking and similar descion. The fact that many innovative firms are willing to pay higher costs to locate their office in areas like Silicon Valley and New York city indicates that importance of physical proximity and its effects on information sharing regardless of communication costs. The study also found correlation to affiliation of social networks. A very predictable factor was the fact that there is a correlation between trading habits an common non-English native languages. This is an easy one to predict since non English speakers tend to converse in their own language limiting anybody else from sharing their information. This leads to similar trading habits. The take away conclusion from this paper is that Google markets exhibited pricing predictabilites which arise from short aversion, optimistic, extremeness aversion and favorite biases. This paper has many limitation. It is based on a specific market that has relatively small number of active traders (less that 2000). The fact that the market participants are few makes it easy to predict trader's behaviors. This paper, however, gives a good insight into google's prediction market and nicely summarizes findings based on data collected. I think that results can be extended from google's experiment to oater prediction markets in other companies. Some issues need to be considered such as the effect of having "fun questions" that some other companies try to avoid to legitimize their prediction market and make it look more serious. --Ziyad Aljarboua
aubourg	The main topic is how markets can be used to study how an organization processes information. Prediction markets can capture changes in opinion at a much higher frequency than surveys, allowing to track information moves within the company. That's why it is interesting. Some fact are confirmation of the common believe that people talk together and do influence themselves : Prediction market positions were most correlated among employees sharing an office than correlation declined with distance for employees. But it is surprising to see that the correlation between two different floors is as low as two offices located in two different cities. Belonging to the same mailing list is also a factor which increases the correlation between two person's results. We can generalize this statement : -each noticeable correlation between persons should results from a complicity or proximity on a certain level. It can be by social classes, buy geographic location, on line groups, mailing lists... I fin the interest the have in entrepreneurial firms relevant because for the leader, the entrepreneur, it is crucial to know if his employees who work for him trust him, and then can do their best, or not. Google "copied" the IEM, which is a proof of good reputation and quality. Like this one, the short selling is forbidden. So people cannot borrow shares or other securities. The Gooble is equivalent to the unit portfolio from the IEM. Michael Aubourg
aubourg	The main topic is how markets can be used to study how an organization processes information. Prediction markets can capture changes in opinion at a much higher frequency than surveys, allowing to track information moves within the company. That's why it is interesting. Some fact are confirmation of the common believe that people talk together and do influence themselves : Prediction market positions were most correlated among employees sharing an office than correlation declined with distance for employees. But it is surprising to see that the correlation between two different floors is as low as two offices located in two different cities. Belonging to the same mailing list is also a factor which increases the correlation between two person's results. We can generalize this statement : -each noticeable correlation between persons should results from a complicity or proximity on a certain level. It can be by social classes, buy geographic location, on line groups, mailing lists... I fin the interest the have in entrepreneurial firms relevant because for the leader, the entrepreneur, it is crucial to know if his employees who work for him trust him, and then can do their best, or not. Google "copied" the IEM, which is a proof of good reputation and quality. Like this one, the short selling is forbidden. So people cannot borrow shares or other securities. The Gooble is equivalent to the unit portfolio from the IEM. Michael Aubourg
luxiudi	------=_Part_32771_11382610.1223264279246 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline The main contribution of the paper is presenting the results of prediction markets of the past few years in an aggregated format. This review is important because it gathers together the findings of other papers and tries to justify the value of prediction markets as an accurate method of predicting event outcomes. The limitation of this paper is that it talks about prediction market results and trades/market makers that make it accurate, however these conclusions are drawn using various results from different papers. As a result, there does not appear to be a single point that the author is trying to make, the topics range from the accuracy of the prediction markets, to the types of investors that make it accurate. Perhaps more detailed examination of each claim is required. However reading the cited papers should clear up these questions. The main insight of the paper is that prediction markets can offer accurate long term results. This is however subject to the quality of the investors. Investors who are "market makers" and less likely affecteb by their personal choice in an outcome will make accurate predictions. The paper also talks about the accuracy of the prediction markets against traditional polls, and discusses how it out performs the polls. After reading this paper, one can assume that using prediction market results as an indicator of how well candidates are doing in a election, might offer better insights into the outcome than polls. Furthermore, the current predictions of the market are based on the price the evening before the election and the average price a week ahead of the election. Using these sets of data, the prediction markets have been deemed accurate. A possible project would be to analyze prices further back in time, and to determine a starting point at which the markets will present accurate predictions. ------=_Part_32771_11382610.1223264279246 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline The main contribution of the paper is presenting the results of prediction markets of the past few years in an aggregated format. This review is important because it gathers together the findings of other papers and tries to justify the value of prediction markets as an accurate method of predicting event outcomes. The limitation of this paper is that it talks about prediction market results and trades/market makers that make it accurate, however these conclusions are drawn using various results from different papers. As a result, there does not appear to be a single point that the author is trying to make, the topics range from the accuracy of the prediction markets, to the types of investors that make it accurate. Perhaps more detailed examination of each claim is required. However reading the cited papers should clear up these questions. The main insight of the paper is that prediction markets can offer accurate long term results. This is however subject to the quality of the investors. Investors who are "market makers" and less likely affecteb by their personal choice in an outcome will make accurate predictions. The paper also talks about the accuracy of the prediction markets against traditional polls, and discusses how it out performs the polls. After reading this paper, one can assume that using prediction market results as an indicator of how well candidates are doing in a election, might offer better insights into the outcome than polls. Furthermore, the current predictions of the market are based on the price the evening before the election and the average price a week ahead of the election. Using these sets of data, the prediction markets have been deemed accurate. A possible project would be to analyze prices further back in time, and to determine a starting point at which the markets will present accurate predictions. ------=_Part_32771_11382610.1223264279246--
michael.aubourg	They step back from their results by comparing them to a natural benchmark like polls, as often as possible. That is a good point to have critic spirit for its own research. the point which is very interesting in the Iowa Markets is that it rely on very different mechanisms for collecting the data. Indeed. a few people speak in the name of other people. That is a first difference. The second one is that their is a reward at the end, so this compells traders to think about the answer more carrefully, and to do some research. Since this category of people are highly educated, the trust we can give to these results is different. Maybe higher. The orders the traders can place are comparable to the ones in the real financial market. There is also requests for immediate execution of a trade, or limit order. However, the point which interested me is the fact that they introduced unit portfolios in order to make sure the market operates as a zero-sum game. Normal markets, are not zero-sum game. Indeed everybody can win (in average) when the economics is good, and everybody can lose (in average) during a crisis like the current Mortgage subprime crisis. The average spread between Iowa results and polls results is 0.395 which is not pretty good! In 1988 and 1992 US presidential elections, the market was really better. It should be interesting to figure out why. Which parameters changed between these two dates and the rest ? We see that there is a general tendency for the market to be closer and more steady than the polls. It is maybe due to the fact that the population mind is more easy to change thanks to media. Furthermore, traders bank their judgement on relevant facts, contrary to the majority of people. That could be an explanation. On the graph, it is surprising to see how bad the market was on the eve prediction, during the 96 Elections. (more than 4.5%). this is not due to chance, and there should be a good reason. the two necessary ingredients that make the market work properly are 1)a big n, where n is the number of trader, in order to have good average results. This is intuitive, (in all polls, the more people we have, the better the result is, provided to process is good.) 2)The market mechanism must facilitate aggregation of their disparate information. This point is less intuitive, but it is logical. In the text the speak about the mistakes done by the market makers. It should be great to have statistics about these occurring mistakes, and to see if it is rational to bank on these mistake, to make profits. Michael Aubourg
zhenming.liu	--_542d62eb-ea61-4029-84b5-7866de74a5bd_ Content-Type: text/plain; charset="Windows-1252" Content-Transfer-Encoding: quoted-printable Comments on the paper Zhenming Liu Both [BFNR00] and [CWZ08] described experimental results from prediction ma= rkets. The market participants in [BFNR00] can be general public while part= icipants in [CWZ08] are Google employees (and mostly engineers). It is inte= resting to contrast these markets=92 behavior.=20 While impressive amount of data are presented in both papers=2C most of the= ir results and conclusions are not surprising. Works of such type is clearl= y important and useful=3B nevertheless I don=92t have much to say regarding= on the data. After all=2C a significant portion of the data is generated f= rom statistical methods like regression=2C which infrequently appear in the= ory or system research in computer science.=20 On the other hand=2C I am interested in a relevant question regarding on de= fining the research methodologies in an emerging area coined as =93empirica= l computer science=94. A clear way in this new area is to mimic what econom= ists have been doing on the problems that tie with computer science. Both [= BFNR00] and [CWZ08] are examples of this approach. Perhaps with our domain = knowledge computer scientists can take one step further to process massive = volume of data which would otherwise be infeasible for economists to proces= s. However=2C I am feeling that this approach heavily addresses the applica= tion side of computer science. (Theoretical) computer science community sha= ll be able to contribute more than =93computational power=94 after its year= s of study on discrete objects=2C which have already led to the establishme= nt of new exciting areas like computational complexity theory or probabilis= tic methods.=20 =20 [BFNR00] J.Berg=2C R.Forsythe=2C F.Nelson=2C T.Rietz =93Results from a doze= n years of election futures market research=94.=20 [CWZ08] B.Cowgill=2C J.Wolfers=2C E.Zitzewitz =93Using Prediction Markets t= o Track Information Flows: Evidence from Google=94 = --_542d62eb-ea61-4029-84b5-7866de74a5bd_ Content-Type: text/html; charset="Windows-1252" Content-Transfer-Encoding: quoted-printable Comments on the paper Zhenming Liu Both [BFNR00] and [CWZ08] described experimental results from prediction m= arkets. The market participants in [BFNR00] can be general public while par= ticipants in [CWZ08] are Google employees (and mostly engineers). It is int= eresting to contrast these markets=92 behavior. While impressive amount of data are presented in bot= h papers=2C most of their results and conclusions are not surprising. Works= of such type is clearly important and useful=3B nevertheless I don=92t hav= e much to say regarding on the data. After all=2C a significant portion of = the data is generated from statistical methods like regression=2C which inf= requently appear in theory or system research in computer science. <= /P> On the other hand=2C I am interested in a relevant q= uestion regarding on defining the research methodologies in an emerging are= a coined as =93empirical computer science=94. A clear way =3Bin this ne= w area is to mimic what economists have been doing on the problems that tie= with computer science. Both [BFNR00] and [CWZ08] are examples of this appr= oach. Perhaps with our =3Bdomain knowledge computer scientists can take= one step further to process massive volume of data which would otherwise b= e infeasible for economists to process. However=2C I am feeling that this a= pproach heavily addresses the application side of computer science. (Theore= tical) computer science community shall be able to contribute more than =93= computational power=94 after its years of study on discrete objects=2C whic= h have already led to the establishment of new exciting areas like computat= ional complexity theory or probabilistic methods. =3B [BFNR00] J.Berg=2C R.Forsythe=2C F.Nelson=2C T.Rietz= =93Results from a dozen years of election futures market research=94. [CWZ08] B.Cowgill=2C J.Wolfers=2C E.Zitzewitz =93Usi= ng Prediction Markets to Track Information Flows: Evidence from Google=94 =3B = --_542d62eb-ea61-4029-84b5-7866de74a5bd_--