Wolfgang Pacher

Centralized- versus Decentralized Prediction Markets for Financial Assets

Are blockchain-based prediction market applications simply the better solution to forecasting financial assets?

MASTER THESIS

submitted in fulfilment of the requirements for the degree of

Master of Science

Programme: Master's programme Applied Business Administration Branch of study: General Management

Alpen-Adria-Universität Klagenfurt

Evaluator Assoc.Prof.Mag.Dr. Alexander Brauneis Alpen-Adria-Universität Klagenfurt Institut für Finanzmanagement

Klagenfurt, May 2019 Affidavit

I hereby declare in lieu of an oath that

- the submitted academic paper is entirely my own work and that no auxiliary materials have been used other than those indicated, - I have fully disclosed all assistance received from third parties during the process of writing the thesis, including any significant advice from supervisors, - any contents taken from the works of third parties or my own works that have been included either literally or in spirit have been appropriately marked and the respective source of the information has been clearly identified with precise bibliographical references (e.g. in footnotes), - to date, I have not submitted this paper to an examining authority either in Austria or abroad and that - when passing on copies of the academic thesis (e.g. in bound, printed or digital form), I will ensure that each copy is fully consistent with the submitted digital version.

I understand that the digital version of the academic thesis submitted will be used for the purpose of conducting a plagiarism assessment.

I am aware that a declaration contrary to the facts will have legal consequences.

Wolfgang Pacher m.p. Klagenfurt, May 2019

ii Abstract

This thesis intends to examine if currently emerging decentralized prediction markets for financial assets are “better” than their incumbent centrally administered counterparts. After an extensive literature discussion, introducing prediction markets and proving why they work, the main focus shifts onto an in-depth investigation and subsequent comparison of Vetr and Augur. Two prediction market platforms, where it is feasible to collect information regarding financial assets, primarily differing by their process of finding consensus. This core difference is made possible by an innovative technology called the blockchain, which will be discussed in chapter 8. However, the “highlight” of this paper will be a utility value analysis which aims to compare both service offerings from the practical standpoint of an average investor based on several predetermined criteria. The work continues with an interpretation of the freshly gathered insights and a brief glimpse into the possible future of prediction markets, before concluding with a section where personal takeaways as well as recommendations for further research are presented.

iii Table of Contents

Affidavit ...... ii

Abstract ...... iii

List of Figures ...... vi

List of Tables ...... vii

List of Equations ...... viii

Abbreviations ...... ix

1. Introduction ...... 1

2. A Definition of Prediction Markets ...... 4

3. The History of Prediction Markets ...... 6

4. Prediction Markets in Theory ...... 12

4.1. Information Aggregation ...... 12

4.2. The Wisdom of the Crowds ...... 15

4.3. Types of Contracts ...... 18

4.4. Administering Formats of Prediction Markets ...... 20

5. Prediction Markets in Practice ...... 22

5.1. What Makes a Prediction Market? ...... 22

5.2. How Prediction Markets Operate ...... 23

5.3. Why They (sometimes) Fail ...... 28

5.4. Cutting-Edge Design ...... 32

5.5. Evidence on Forecast Accuracy ...... 34

5.5.1. Macro Derivatives ...... 34 5.5.2. Politics...... 38 5.5.3. Business ...... 41 6. Outlining the Research Methodology ...... 43

iv 7. Centralized Prediction Markets for Financial Assets ...... 45

7.1. General Overview ...... 45

7.2. Vetr ...... 45

7.2.1. History of the Company ...... 46 7.2.2. Service Description ...... 48 7.2.3. User Interface ...... 50 8. The Blockchain Technology ...... 53

8.1. Description...... 53

8.2. Brief Historical Background ...... 55

8.3. Types of Blockchains ...... 56

8.4. Defining Cryptocurrencies: Coins versus Tokens ...... 57

9. Decentralized Prediction Markets for Financial Assets ...... 60

9.1. General Overview ...... 60

9.2. Augur ...... 62

9.2.1. History...... 63 9.2.2. Service Description ...... 65 9.2.3. Potential Issues and Risks ...... 70 9.2.4. User Interface ...... 72 10. Utility Value Analysis...... 77

10.1. Augur versus Vetr ...... 77

10.2. Interpretation of Results ...... 83

11. The Future of Prediction Markets ...... 85

12. Conclusion ...... 89

Bibliography ...... 94

v List of Figures

Figure 1: The Gartner Hype Cycle for Emerging Technologies of 2005 ...... 9 Figure 2: The Ideal Balance of Information Diversity and Network Connectivity ...... 17 Figure 3: Operation Principle of Prediction Markets ...... 23 Figure 4: Information becomes quickly incorporated...... 24 Figure 5: New information is continuously aggregated across time...... 25 Figure 6: Prediction markets show very little arbitrage potential...... 26 Figure 7: The Favorite-Longshot Bias Pricing Anomaly in Sports Betting Markets ...... 30 Figure 8: The Favorite-Longshot Bias in Prediction Markets ...... 31 Figure 9: Macro derivatives are slightly more accurate than survey-based forecasts...... 35 Figure 10: Prediction markets perform better even the night preceding an election...... 39 Figure 11: Prediction markets are usually more accurate than polls, even after excluding known biases...... 40 Figure 12: Vetr’s User Growth and Development of Aggregated Ratings ...... 48 Figure 13: The Upper Half of Vetr’s Information Screen about Apple Inc...... 50 Figure 14: Vetr’s “Special Sauce” ...... 51 Figure 15: The Lower Half of Vetr’s Information Screen on Apple ...... 52 Figure 16: A Blockchain Visualization ...... 54 Figure 17: Augur’s Reputation Token ...... 63 Figure 18: The Reporting Flowchart ...... 68 Figure 19: Augur’s User Interface (Web Version) ...... 73 Figure 20: Apple Inc. related Markets on Augur ...... 74 Figure 21: The Most Frequented Prediction Market featuring Financial Assets on Augur ... 75

vi List of Tables

Table 1: The Main Contract Types ...... 19 Table 2: Economic derivatives show somewhat smaller errors than forecaster...... 36 Table 3: Macro-economic derivatives exhibit slightly smaller errors than experts...... 37 Table 4: A Comparison of Information from Prediction Markets and Polls ...... 41 Table 5: A Utility Value Comparison of Augur and Vetr ...... 78

vii List of Equations

Equation 1: The Trader’s Utility Maximization Problem ...... 13 Equation 2: The Prediction Market Equilibrium ...... 13 Equation 3: The market price equals the mean belief...... 13

viii Abbreviations

AI… Artificial Intelligence

AML… Anti-Money Laundering

BIS… Bank for International Settlements

BTC… Bitcoin

CAGR… Compound Annual Growth Rate

CARA… Constant Absolute Risk Aversion

CBOT… Chicago Board of Trade

CFTC… Commodity Futures Trading Commission

CVP… Customer Value Proposition

DARPA… Defense Advanced Research Projects Agency

EMH… Efficient Market Hypothesis

ETF(s)… Exchange-Traded Fund(s)

ETH… Ethereum

FOSS… Free and Open-Source Software

FX… Foresight Exchange

HSX… Hollywood Stock Exchange

ICO… Initial Coin Offering

IEM… Iowa Electronic Markets

IoT… Internet of Things

IPO… Initial Public Offering

ix ISM… Institute for Supply Management

KPSS… Kwiatkowski-Phillips-Schmidt-Shin

KYC… Know-Your-Customer

MIT… Massachusetts Institute of Technology

ML… Machine Learning

NYSE… New York Stock Exchange

OPEC… Organization of the Petroleum Exporting Countries

PM(s)… Prediction Market(s)

REP… Reputation Token

ROI… Return on Investment

SARB… South African Reserve Bank

WMD(s)… Weapon(s) of Mass Destruction

x 1. Introduction

While looking at the list of potential research topics for this master thesis, this one clearly stood out. It almost flawlessly combines my greatest passions and academic interests, namely technology and innovation, financial assets, and forecasting (i.e., predicting future outcomes). In addition, it incorporates an “excitement factor”, which can best be described as, a favorable time window that allows me to observe and explore potentially lifechanging cutting-edge technological innovation within the space of prediction markets (PMs) while still in progress.

This is true, given the fact that the concept of prediction markets has been known and extensively studied for more than a century, but there have only been two major disruptive waves that changed anything of significance. The first was the implementation of the internet, which enhanced network effects through an easier as well as faster access to PMs and information sharing in general. The second technology, the blockchain, has been around for decades but has just recently been hyped enough to gain mainstream attention, allowing certain entities to gather necessary funds to try and successfully launch their highly innovative ideas. On the one hand, a black box and an “unnecessary evil” for many people, but, on the other hand, a revolutionary tool for developers all over the globe to digitalize real-life applications and ultimately improve human-machine systems’ efficiencies.

One of such projects, foreseen to launch its product at around August of 2018, aiming to combine prediction markets and the blockchain technology, is called Augur. A project that will be a main subject of interest throughout this thesis, with special attention given to the consumer adaptation process and financial assets.

I will commence this work by giving a proper definition of prediction markets and specifying how they can be used to gather viable information to facilitate predicting uncertain future outcomes.

This is followed by a chapter concerning the origins of prediction markets. A timeline will be established, featuring major developments, from Galton’s “Vox populi” to examples where prediction markets failed, and prediction market’s current position in its development process.

1 In the next segment, I am going to elaborate on the basic underlying principles of prediction markets. The state of the current academic research will be discussed – Hayek’s marginal- trader hypothesis, Fama’s efficient market hypothesis, and James Surowiecki et al. – answering questions such as, why do they work (most of the time) and, under what conditions do they work best? Furthermore, the most important prediction market formats and contract types will be presented in this section.

Thereafter, in chapter 5, the focus of attention shifts onto analyzing the performance of prediction markets in numerous practical areas of implementation.

The main part of this paper begins with an outline of the research methodology, followed by a chapter describing the concept of traditional (i.e., centralized) prediction markets and their various current implementations as a commercialized market research tool for financial assets. Accordingly, I will perform the same analysis for the currently emerging blockchain-based decentralized prediction markets. However, in an effort to smoothly bridge the knowledge gap, the blockchain technology will be discussed in a chapter amid those aforementioned examinations.

In the comparison passage, I plan on focusing solely on publicly accessible prediction markets for financial assets, ultimately choosing only one centralized- and one decentralized application. Vetr, a platform where stocks and exchange-traded funds (ETFs) are valued by the crowd, or Estimize, a crowdsourced restricted database for earnings estimates, are decent examples of centralized PMs specialized in financial assets and accessible to the public. In the blockchain-based PMs category, there are several potential candidates, beta-testing their product right now, some of them estimated to be ready for market entry within months, but clearly the most ambitious and biggest, regarding market capitalization, is Augur.

Here, the main goal is to establish a list of similarities as well as unique value propositions and discuss pros and cons from a practical end-user’s point of view. Typically, in such a case, for a self-proclaimed “finance guy” who writes his master thesis at the Department of Finance and Accounting, the only reasonable choice to achieve an objective comparison, featuring unbiased and tangible results, is some variation of a quantitative analysis. Unfortunately, due to a lack of obtainable research data, such analysis remains unfeasible. Therefore, after careful consideration and examining several alternatives, I have decided to conduct a utility value

2 analysis, a qualitative research method which intends to aggregate one final value score based on predefined criteria and their weightings.

In the closing part of this thesis, I would like to personally glimpse into the future. Undoubtedly, the introduction of blockchain technology into the prediction market evolution process has been, among others, a recent disruptive and reenergizing force, but what else may be expected in the upcoming years?

There are other technologies, like the Internet of Things (IoT), which pair arguably well with prediction markets. Here, a probable future scenario involving sensors as permanent information providing participants in prediction markets, first comes to mind. Others may include technologies such as Artificial Intelligence (AI) and Machine Learning (ML) combined with a decentralized prediction market. Once those systems are properly established, crossover applications like automated trading bots or similar products can quite rapidly become a reality.

At last, I am going to conclude this paper by summarizing the new-found insights, alongside outlining my personal impressions regarding prediction markets and their technological evolution, as well as providing recommendations for further research.

3 2. A Definition of Prediction Markets

This thesis intends to focus on PMs for financial assets, therefore a broad definition of prediction markets, which includes forecasting of prices for different underlying assets such as futures markets or hedge funds, as well as the traditional PMs aiming to dynamically estimate the likelihood of an event outcome, will be given.1

A prediction market (also sometimes referred to as information market, electronic market, decision market, virtual market, election stock market, idea futures, artificial market, political stock market or game market) is a forecasting mechanism using predictive analytics2, designed to efficiently manage the dynamic aggregation of dispersed information among various agents3. In practice PMs, belonging to the more general concept of crowdsourcing, can be thought of as exchange-traded markets, structured to elicit and accumulate beliefs about states of the future. Participants with different information trade contracts whose payoffs are related to unknown future outcomes, consequently the market prices of these contracts represent a market-aggregated prediction regarding the likelihood of those events occurring4.

Broadly speaking, prediction markets may be categorized into two types. On one side there are enterprise PMs, which are usually restricted to a select group of participants but can be found in a wide variety of areas within a corporation. Examples range from revenue forecasting, capital budgeting and demand planning to risk-, innovation life cycle- and project management. In 2013, according to Melanie Swan, PMs have been employed in 100 to 200 organizations including economic powerhouses such as Microsoft, Ford, Hewlett Packard, Electronic Arts, PayPal, Bosch, Harvard, Boeing, and Amazon. Furthermore, prediction market vendors, such as Consensus Point, Inkling or Qmarkets, aim to monetize the concept of PMs by providing an enterprise idea management solution to retail- and institutional clients.5

Consumer prediction markets, which are designed to reach the general public either for the purpose of financial gain or scientific research, reside on the other side of the spectrum. Due

1 see. Shrier et al., 2016, p. 3. 2 Ibid. 3 see. Tziralis/Tatsiopoulos, 2012, p. 75. 4 see. Wolfers/Zitzewitz, 2006, p. 1. 5 see. Swan, 2013, p. 15.

4 to the fact that consumer PMs can be applied to any interest area where sufficient demand as well as a diversity of opinion regarding an uncertain and clearly definable future outcome exists, nowadays these markets seem almost pervasive in our daily lives. Examples vary from event prediction in sectors such as politics and economics (PredictIt, Iowa Electronic Markets), sports (Bwin, Betfair), science and technology (SciCast), healthcare (Iowa Electronic Health Markets), acting (Hollywood Stock Exchange) and online gaming (SimExchange) to markets specializing in various underlying assets, such as financials (Vetr, Stox, Estimize), the weather, or a fruitful global crop of tomatoes next season (Augur, Gnosis).

Rivalling alternatives to prediction markets include the Delphi method, as well as expert surveys, and polls.

While static surveys and polls objectively seem inferior to the dynamic nature of PMs, the Delphi method, in comparison, may potentially be a more favorable research tool. In 2011, Graefe and Armstrong conducted a laboratory experiment designed to compare several research methods. Even though no statistically significant differences in accuracy among tested methods could have been observed, participants concluded that they were least satisfied with the group interaction process of PMs and perceived them as the most difficult method to apply6.

6 see. Graefe/Armstrong, 2011, p. 183.

5 3. The History of Prediction Markets

Today’s modern concept of prediction markets as a forecasting method originated from a tradition as old as mankind, gambling. That is, people betting among each other on the outcome of an event. According to Rhode and Strumpf, such historical betting markets have been active for at least half a millennium. The first written records relating to the predecessors of contemporary prediction markets date back to 1503, when wagering on who will become the next pope was already considered to be an outdated practice. Around ninety years later in 1591, Pope Gregory XIV banned the betting on papal elections, consequently forcing those markets to shut down or illicitly operate in the shadows. In the 16th and 17th century, betting on the outcome of civic elections was a shared practice among people in Italian metropoles such as Venice or Genoa. In Great Britain, including its former colonies New Zealand, Australia, Singapore, South Africa, Canada and the United States, wagering on political subjects can be th traced back as far as the 18 century. Although bets on other events such as the outcome of foreign military missions were made, the overwhelming portion was placed on political events. This can be attributed to the fact that back then, politics was one of the few areas of interest where several people had sufficient information supply and could relate to the outcome.7

Meanwhile in the 1730s, Japanese Samurai, who at that time were exclusively compensated in rice, saw their fortunes tumbling when a fruitful rice harvest (i.e., an oversupply) caused the price of rice to sharply depreciate. In an effort to mitigate such high volatility risk scenarios, the Japanese invented the first rice bills, the predecessor of today’s futures markets.8 About a century later in 1848, the first official futures exchange, the Chicago Board of Trade (CBOT), opened its doors in the United States9.

Futures markets are designed to focus on prices of underlying assets, rather than outcome- related possibilities, practically making it possible to speculate on an immense amount of different assets. As it turns out, the prices of real underlying assets may also contain valuable information about probabilities of future incidents. In 1984, economists, examining the relationship between orange futures and the weather, revealed an occurrence where prices of

7 see. Rhode/Strumpf, 2014, pp. 560-586. 8 see. Shrier et al., 2016, p. 4. 9 see. CME Group, 2018.

6 orange futures at the close of the market, around 3:00 pm, predicted errors in the weather forecasts of the temperature later that evening. This observation impeccably exemplifies how the crowd, in this case the aggregation of orange buyers and suppliers, could end up exposing information about the weather that even climate experts sometimes miss10.

In 1906, Francis Galton, a renowned mathematician and half-cousin of Charles Darwin, drew interesting conclusions from a public weight judging competition in Plymouth, England. After thoroughly analyzing the results, he detected that the average participant of this weight judging contest, assumed to not be an expert, was as well-equipped to make a reasonable estimate as the average voter who decides on political subjects. His calculations revealed that the vox populi, or voice of the people, was surprisingly accurate. The median was within 0.8% of the true value and the average within 0.08%. These findings became eternalized as an early example of what we now regard to as “the wisdom of the crowds”, a vital pillar of modern prediction markets.11

In the period between 1888 and 1940, when markets on presidential elections were so popular that sporadically they had higher trading volumes than the New York Stock Exchange (NYSE), and major news outlets such as the New York Times and the Sun reported the betting odds as forecasts of election outcomes on a regular basis, election betting markets reached their initial peak in the United States12. Although these markets provided accurate predictions of election results in an era before scientific polling13, in the years following 1940 their employment seemingly disappeared. This can mainly be explained by rising moral concerns associated with wagering on political outcomes, opinion polls developing popularity as an alternative method, and other gambling opportunities such as horse race betting becoming legally available14.

In 1945, the Austrian economist Friedrich Hayek publishes an essay titled “The Use of Knowledge in Society”, which endorses the ability of the price system to aggregate dispersed information within an economy and discusses how this can only contribute to efficiency in an open market.15 This essay, coupled with the efficient market hypothesis (EMH) by Eugene

10 see. Roll, 1984, pp. 861-880. 11 see. Galton, 1907, pp. 450-451. 12 see. Rhode/Strumpf, 2004, p. 128. 13 see. Erikson/Wlezien, 2012, p. 532. 14 see. Rhode/Strumpf, 2004, p. 139. 15 see. Hayek, 1945, pp. 519-530.

7 Fama as well as experimental economics, e.g. by Plott and Sunder in 1982 and 1988, ignited a wider scientific interest for prediction markets as a forecasting method in the second half of the 20th century.16

In 1988, supported by advances in telecommunications, which allowed information to be effortlessly and rapidly shared worldwide17, the first modern online prediction market application, the Iowa Electronic Markets (IEM), was launched. Employed by the University of Iowa, to forecast the US presidential election outcome of that same year, it soon emerged that the accuracy of the markets’ predictions was outstanding18. Consequently, researchers began studying the validity of prediction markets in several other areas such as sports, the weather, or the economy.

Furthermore, it must be mentioned that two other events in the early-2000s contributed to the popularity of prediction markets. Firstly, a 2003 Defense Advanced Research Projects Agency (DARPA) announcement to abandon its “Policy Analysis Markets” (i.e., a PM experiment to gather intelligence for military purposes) due to ethical concerns regarding its terror attack- and assassination markets, brought wide public awareness to the potential of prediction markets, and secondly, the 2004 publication of James Surowiecki’s bestseller “The Wisdom of Crowds” in which he describes prediction markets as an efficient tool to harness collective intelligence. Shortly thereafter, in 2005, prediction markets became registered on the renowned Gartner Hype Cycle (Figure 1).19

The Gartner Hype Cycle separates the evolution process of promising emerging technologies into five stages. It all starts with a technological breakthrough, which generates media attention and publicity. This positive news coverage, featuring a few minor success stories, creates excitement among consumers and researchers alike, leading to the peak of inflated expectations in the second stage. The third stage, the trough of disillusionment, is reached because on the cutting edge of technology setbacks are inevitable, and when they happen, it diminishes prematurely overstated expectations of initial clients, who will only continue to support remaining providers if they actively engage in improving products to meet their constantly

16 see. Snowberg et al., 2012, p. 1. 17 see. Snowberg et al., 2008, pp. 385-402. 18 see. Forsythe et al., 1992, p. 1142. 19 see. Graefe, 2017, p. 6.

8 evolving demands. The next stage, the slope of enlightenment, is characterized by a technology maturing process, where further use cases and benefits are detected as well as second- and third-generation product updates are realized. In the final stage, the technology reaches its plateau of productivity through mainstream adoption, potentially becoming a mass consumer product.20

Figure 1: The Gartner Hype Cycle for Emerging Technologies of 2005

Notes: Reproduced from Gartner's Hype Cycle Special Report for 2005, (Source: https://www.gartner.com/doc/484424/gartners-hype-cycle-special-report).

Based on this information, it can be concluded that in late 2007, when PMs were used to accurately project events from sales of computer printers and pre-clinical drug approvals to the spread of infectious diseases21, election results, and the Federal Reserve’s interest rate decisions, the technology had entered stage two and quite possibly already reached its peak of

20 see. Gartner, 2018. 21 see. Polgreen et al., 2007, pp. 272-279.

9 inflated expectations. In 2008, when Intrade, a PM platform established in 1999, failed to correctly predict a couple of political forecasts, first signs of disapproval started to emerge. Although it was evident that Intrade’s performance was objectively more accurate than any single poll or pundit22, critics proclaimed that PMs were too small, in terms of participants and stakes, as well as too slow to react to new events23. Accordingly, this incident can be interpreted as the prelude to the trough of disillusionment, the third phase in Gartner’s Hype Cycle.

From that point on, things deteriorated for Intrade, at that time the only real-money PM focused on predicting a diversity of events besides sports and politics, and subsequently for the concept of PMs as a whole. First, the Dodd-Frank financial reform of 2010 led to an immediate prohibition of all futures markets related to terrorism, war, assassinations, gaming, and anything “contrary to the public interest”, advising in 2012 to include election markets. Followed by allegations of price manipulation at Intrade, grounded on an incident involving the ominous “Romney Whale” in 2012.24 In the same year, the Commodity Futures Trading Commission (CFTC) sued Intrade for filing false forms and violating the CFTC’s Off- Exchange Options Trading Ban by offering markets on economic indicators like the future unemployment rate, eventually leading to the demise of Intrade in March 2013. The fact that after the shutdown around $4.2 million from Intrade and another related company simply vanished, only reinforced the suspicion of wrongdoing or illicit behavior within the firm, and negatively impacted the image of prediction markets.25

However, since 2013, when people started to adjust their expectations downward to more realistic levels, it appears that PMs began to transition into the fourth phase of the Gartner Hype Cycle. New adaptations, such as quadratic voting, which ensures efficient outcomes in large populations26, or, assigning a prediction quality score to individual users, as well as social information gathering through iterated predictions, represent only a fraction of prediction markets’ looming potential.27

22 see. Leonhardt, 2008. 23 see. Ledbetter, 2008. 24 see. Rice, 2014. 25 Ibid. 26 see. Lalley, 2015. 27 see. Shrier et al., 2016, p. 8.

10 In the upcoming chapters of this thesis, I am going to explore the current and future prediction market landscape in further detail, specifically focusing on the blockchain technology and financial assets prediction markets.

11 4. Prediction Markets in Theory

This chapter begins by exploring the scientific foundations from which modern-day prediction markets have emerged and concludes with a presentation of main PM contract types as well as administering formats.

4.1. Information Aggregation

By presenting the efficient market hypothesis, Fama laid out the theoretical basis for the claim that prediction markets perform well, maybe even fully efficiently, in aggregating dispersed information. It states that a sufficient number of marginal traders with rational expectations, aiming to maximize utility through maximizing profits, drive prices in capital market in such a way that there is no opportunity for arbitrage28. This implies that in theory there will always be individuals looking for opportunities where the crowd is wrong, which tends to push market prices back towards a sensible value. This claim inspired a wave of data scientists as well as other researchers to further investigate and try to define conditions that may allow for such an “efficient” prediction market, where prices “at any time ‘fully reflect’ all available information”29.

In 1976, Grossman formulates a set of sufficient conditions for the equilibrium price of index futures to summarize private information flawlessly: In a market where traders with constant absolute risk averse (CARA) utility functions each receive independent draws from a normal distribution about the true value of an asset, the market price fully summarizes their information.30

In 2004, Manski, while analyzing binary options, realized that many researchers simply assumed prices to reveal a market-based probability estimate but no appropriate theoretical results existed that could support this assumption. He exemplified the importance of this matter by creating a scenario where prediction market prices fail to correctly collect information. In his model all participants are prepared to risk exactly $100. Therefore, if a contract paying $1 when an event occurs, is selling for $0.667, then buyers each purchase 150 contracts, while

28 see. Shvarts/Green, 2007, p. 12. 29 see. Malkiel/Fama, 1970, p. 383. 30 see. Wolfers/Zitzewitz, 2006, p. 2.

12 sellers, at a price of $0.333, have the funds to sell 300 contracts. Accordingly, this market can only be in equilibrium if there are exactly two times more buyers than sellers, inferring that the market price must decrease at the 33rd percentile of the belief distribution, rather than the mean. The same reasoning suggests that a prediction market price of π implies that 1-π% of the participants think that the outcome has less than a π% chance of taking place. It is evident that the assumption all traders are willing to risk a fixed amount, is the driving force in this example.31

In 2005, Wolfers and Zitzewitz presented a prediction market setup where prices matched the average beliefs among participants. They considered traders with a log utility function and initial wealth, y, who must decide how many prediction market contracts, x, to acquire at a price, π, provided that they believe the probability of winning their bet is q:

Equation 1: The Trader’s Utility Maximization Problem

The prediction market reaches equilibrium if supply equals demand:

Equation 2: The Prediction Market Equilibrium

If beliefs (q) and wealth (y) are independent, then this implies:

Equation 3: The market price equals the mean belief.

31 see. Manski, 2006, pp. 425-429.

13 Showing that under log utility the prediction market price equals the mean belief among participants, or, in case of a correlation between wealth and beliefs, the prediction market price represents a wealth-weighted average belief. Although this discovery only holds under log utility, it carries the generalization advantage that no assumptions about the distribution of opinions is required. A sensitivity analysis regarding a range of alternative utility functions as well as distributions of beliefs generally yields prediction market prices that only insignificantly deviate from the mean of beliefs.32

Furthermore, it should be noted that in both models, Manski’s and Wolfers/Zitzewitz’s, the origins of beliefs are undefined, giving either the advantage to circumvent a theoretical problem introduced by Milgrom and Stokey in 1982, that under mutual beliefs no trade will develop. The main implication of the “no trade theorem” is that if a participant is eager to trade, his counterpart should be aware of the fact that the initiator must have an information advantage and therefore should adapt his own belief system accordingly. To this day, solving the riddle of, why any trades in prediction markets happen, remains an important theoretical open quest. In 2006, Wolfers and Zitzewitz observed prediction markets driven by uninformed outsiders, discovering that participant’s motivation either derived from hedging- or entertainment demands, or the desire to distort market prices.

An additional key feature of prediction markets is that possible reward incentivizes traders to continually discover new information. Back in 1976, Grossman and Stiglitz presented a scenario, which featured high costs for collecting information, while pointing out that contract prices can never be fully efficient otherwise no trade would emerge. Hence, their model made sure that in equilibrium just enough information was unknown so that a portion of participants would become interested in seeking out new information.

Moreover, prediction markets not only incentivize constant information gathering, they also encourage the truthful revelation of information. Imagine a prediction market as source of an upcoming decision. In that case, there could be participants motivated to manipulate prices in an attempt to influence the decision’s outcome in their favor. Usually, the manipulator

32 see. Wolfers/Zitzewitz, 2006, p. 3.

14 executing those “contrary trades” loses money in this market, which in turn raises the rewards for informed trading33 and indirectly may lead to a higher accuracy of market prices.34

4.2. The Wisdom of the Crowds

Prediction markets basically need two things to operate sufficiently. They need the technology, of which the underlying scientific principles have just been discussed, combined with motivated individuals interacting amongst each other.

Since Galton’s observation in 1906, extensive research, regarding social interaction and how to improve the process of tapping into the wisdom of the crowds, has been conducted, suggesting that understanding, enabling and incentivizing the human behavior in prediction markets is as important as the technological pillar. One of those researchers, James Surowiecki, a scientist fascinated by collective intelligence and a proponent of prediction markets, laid out a set of characteristics that would describe a “wise” crowd. He states three conditions; diversity, independence, and a “particular kind of decentralization” that a group of ordinary people should possess in order to solve problems of cognition more accurately than a group of experts35.

Firstly, diversity is important to warrant that the group has a lot of different information. If a crowd almost exclusively contains likeminded people it is unlikely to be wise, simply because the group will not know more than the individuals of whom it is composed. Secondly, independence is necessary to ensure that people express what they know, rather than hide it. Surowiecki is aware of the fact that groups are frequently mistaken if members merely follow one another without pooling individually held information. Thus, he infers that organizations often do best if everyone behaves independently and does not pay an unhealthy portion of their attention to the acts or statements of others. “The smartest groups” he declares, “are made up of people with diverse perspectives who are able to stay independent of each other.” As an example, he mentions investment clubs in the United States whose worst performers consist of people who like one another, socialize together, and display a high level of agreement. The best performers involve people who welcome dissent but rarely see each other. Accordingly, it

33 see. Hanson/Oprea, 2004, p. 3. 34 see. Wolfers/Zitzewitz, 2006, p. 4. 35 see. Sunstein, 2004, p. 2.

15 can be assumed that small but highly independent crowds may still be able to produce valuable predictive power, whereas large and highly connected groups most likely lack such a substantial attribute. Additionally, independence aids decreasing the stern risks associated with information cascades, which occur when people disregard their own knowledge and rather pay attention to signals provided by others. Social scientists have been observing such cascades not only among ordinary people choosing restaurants, sneakers or political candidates, but also among physicians performing diagnoses, and even federal judges’ ruling on cases. The underlying problem, concerning information cascades, is that the individuals within the group tend to perform far worse than they would if everyone just disclosed their private information. Therefore, in an effort to minimize the effects of bad cascades Surowiecki suggests a “wide array of options and information” as well as having at least a few people who are willing “to put their own judgment ahead of the group's, even when it's not sensible to do so.” In other words, he suggests a diversity of options and information as well as adding several persons to the group that act as “safeguards” (or market makers) to ensure the integrity of the market.36

In practice, markets are dynamic, hence the degree of information diversity and network connectivity (i.e., the independence of market participants) varies from market to market and changes over time. However, as presented in Figure 2, research suggests that there can be an optimal balance, able to enhance decision making skills among contributors.

In 2013, a team of MIT researchers focused on the process of social learning in an attempt to obtain further insight on how people make decisions and follow the idea flow through a human network. Their study featured a transparent social trading platform implemented to analyze financial decision making among traders. As it turns out, those best at decision making are social explorers, characterized by an open mind, endlessly seeking out new people and ideas without any prejudice regarding the “best” people or ideas. Social explorers intend to gain exposure to a large variety of perspectives, therefore interacting with a great diversity of individuals37. The results showed that traders who had the right balance and diversity of ideas in their network generated 30% higher returns than either isolated traders or those following the masses. Moreover, the study disclosed that comparatively minor adaptations to a person’s social learning strategy, to match the current state of the flow of ideas, can help ameliorate their

36 see. Sunstein, 2004, pp. 2-3. 37 see. Shrier et al., 2016, p. 12.

16 status quo. Researchers used little incentives to encourage isolated participants to get more involved, and conversely, traders who were too attached to one group were promoted to explore outside their immediate social network.

Figure 2: The Ideal Balance of Information Diversity and Network Connectivity

Notes: Reproduced from Shrier et al., 2016, p. 13.

In addition, the team discovered that with the help of computers and deep learning techniques, they were able to adjust the social network so that it steadily remained in the healthy “wisdom of the crowds” balance. Results of a subsequent large-scale experiment displayed improved ROIs for all social traders by more than 6%, confirming that managing idea flows through a computer generated “idea flow” prediction model leads to significant enhancements to human behaviors, and that markets can be structured with adaptive incentives to produce more accurate “wisdom of crowds” results.38

Finally, the last and potentially least intuitive of Surowiecki's three conditions, decentralization, must be elaborated. Surowiecki is critical of the widespread idea that what is needed is more centralization. He argues that desired solutions are far more likely to occur “if you set a crowd of self-interested, independent people to work in a decentralized way on the

38 see. Pentland, 2013.

17 same problem”, essentially looking for processes in which autonomous individuals, all equipped with their own knowledge, can attend to problems “while also being able to aggregate that local knowledge and private information into a collective whole.” In 2004, Surowiecki tried to back up his claims regarding decentralization, with an example of communication- and implementation procedures of the American military during the Iraq war, but ultimately was not able to fully support his conclusions. However, this does not imply that his statements were incorrect.39

Coincidentally, the currently launching Augur project is trying to create such decentralized conditions, through a blockchain-based prediction market platform open to the public. In theory, offering anyone with an internet connection the opportunity to trade and aggregate information on any topic imaginable. In due time, when enough data to allow for a meaningful analysis is available, maybe Augur will provide us with the ultimate proof that decentralizing the information aggregation process enhances accuracy of crowds’ predictions.

4.3. Types of Contracts

As previously mentioned, a prediction market is commonly implemented as a simple contract (or wager) which pays off if a particular outcome, such as crude oil reaching a specific target value y, takes place. In theory, provided that both the efficient market hypothesis holds and that the market acts as a risk-neutral representative trader, the contract’s price will be the best estimate of multiple parameters tied to the probability of that outcome. Table 1 shows three standard types of contracts designed to discover either the probability of a certain occurrence, or, in case the outcome takes on numerical values, the expected mean or median.40

Firstly, the “winner-takes-all” contract, which costs a certain amount $p and only pays $1 if event y, for example the initial unemployment claims are announced to be between 330,000 and 340,000, occurs. In this case, the contract’s price represents the market's belief of the likelihood that initial unemployment claims will be within this range. Moreover, if one considers a market that offers all possible range variations, it is possible to illustrate the entire distribution of the market's expectations regarding the true value. For instance, a set of contracts that wins if initial unemployment claims are between 310,000 and 320,000, another that pays

39 see. Sunstein, 2004, p. 3. 40 see. Snowberg et al., 2012, p. 2.

18 if they are between 320,000 and 330,000, and so on, reflects the complete market’s probability density function of initial unemployment claims, integrated over bins of 10,000.41

Table 1: The Main Contract Types

Notes: Reproduced from Wolfers/Zitzewitz, 2004, p. 110.

The second contract type is called an “index” contract. These contracts are designed to capture moments of the market’s distribution of opinions. Their payoff is tied to a specific outcome value, thus initially unknown. Following above’s example, y can be defined as the number of initial unemployment claims, consequently the price of the contract will be the market’s prospect of the true unemployment number: E[y]. In addition, by combining several index contracts, for instance by adding a contract that pays off the square of initial unemployment claims y2, it is possible to aggregate higher order moments, such as in this case the variance of the market’s expectations: E[y2]- E[y]2.42

The last type, the “spread” contract, is a combination of the aforementioned two contracts. Familiar to sports bettors around the globe, these contracts are employed by researchers to elicit the central tendency of a market’s expectations. They are characterized by a predetermined price and payoff if an indicator is below the spread value, but the spread’s value y* fluctuates

41 see. Snowberg et al., 2012, pp. 2-3. 42 see. Snowberg et al., 2012, p. 3.

19 as trade occurs. Participants bid on the cutoff which determines whether an event happens. For instance, in sports, one could bet that their team will win by a point spread of at least y* points43, or, in an economic context, there may be a contract that pays off if initial unemployment claims turn out to be y* percent over 320,000.

In either case, the predefined cost/payoff structure determines which percentile of distributed opinions will ultimately be identified. For example, if a contract costs $1 and pays $2 if y > y*, then the spread size y* will keep changing until an equal quantity of buyers and sellers is reached. At that point, y* represents the median of the market’s distribution of beliefs. However, a contract that costs $4 and pays $5 if y > y*, will produce a value of y* that the market believes to be a four-fifths probability of y < y*, hence identifying the 80th percentile of the distribution.44

In concluding this segment, it should be addressed that although all these aforementioned contract types have been employed in some form or the other, the “winner-takes-all” contracts on a single event are by a wide margin utilized the most. This can be explained by their simplicity, which makes it easy for the general population to understand and apply them.

4.4. Administering Formats of Prediction Markets

Although present-day prediction markets are predominantly employed as equity markets, meaning interested buyers and sellers trade via a perpetual double auction process45, a few other design variations have been suggested throughout the years. In this section, the most notable alternatives, such as play-money markets, pari-mutuel markets, and market scoring rules, will be discussed.

In 2006, the CFTC started to highly regulate prediction markets in the U.S.46, leading to the creation of play-money markets, a system were successful traders get rewarded by a virtual currency, which can be exchanged for certain goods and services, or collected for prestige. Besides numerous internal corporate prediction markets, this format is utilized by the Hollywood Stock Exchange and Lumenlogic, a leading prediction market vendor. Current

43 see. Snowberg et al., 2005, p. 367. 44 see. Snowberg et al., 2012, pp. 3-4. 45 see. Snowberg et al., 2012, p. 4. 46 see. Arrow et al., 2008, pp. 877-878.

20 research shows that these play-money markets can function as accurate as real-money markets. However, further observation and examination is needed to legitimize these conclusions47.

The next form, pari-mutuel markets, have been occasionally run as large-scale economic derivatives markets by investment banks such as Deutsche Bank and Goldman Sachs. Participants are able to bet on specific ranges of values regarding macro-economic figures, for instance, the initial unemployment claims or non-farm payrolls, but the price remains undecided until the end of the auction process. During the auction, theoretical end price estimates are steadily publicized and accessible to traders. Winning participants, those who picked the right range, split the total amount wagered on all offered bets among them. One advantage of pari-mutuel markets is that they capture the full probability density function of the market’s beliefs with one single market, rather than combining multiple “index” contracts to obtain the same result.48

The last design, market scoring rules49, have already revealed great potential under lab conditions and have been applied at Microsoft as well as other commercial companies for their internal markets50. The concept begins with a simple scoring rule, rewarding participants for the accuracy of their forecasts. If other entrants think they have a better prediction, they can acquire the right to the reward and potentially profit themselves. This specific market form decreases speculation and works well for low frequency markets but requires a market maker to fund the reward for accurate predictions51.

47 see. Servan-Schreiber et al., 2004, pp. 243-251. 48 see. Snowberg et al., 2012, p. 5. 49 see. Hanson, 2003, p. 110. 50 see. Abramowicz, 2007, p. 1. 51 see. Snowberg et al., 2012, p. 5.

21 5. Prediction Markets in Practice

In this chapter, I am going to thoroughly examine prediction markets in several different real- world scenarios. I will start by recapitulating why prediction markets work. This is followed by a presentation of numerous examples on how these aspects lead to favorable characteristics, such as the fast incorporation of information, an absence of arbitrage opportunities, and a resistance to manipulation, in practice. The next part gives an overview of design flaws which may cause prediction markets to fail and presents design recommendations to avoid such incidents, before concluding this chapter with an examination of prediction market accuracy among a selection of different application areas.

5.1. What Makes a Prediction Market?

Present research suggests that prediction markets are not always efficient. This means that in some cases the simplified underlying math simply does not fully reflect reality. However, in many practical settings the impact from violations against these modelling rules is quite manageable and the forecasts obtained have shown to be fairly reliable, indicating that in most cases prediction markets possess at least a weak form of efficiency52. In addition, they are known to be scalable, relatively cost-efficient, and able to incorporate multiple sources of information. Furthermore, it must be mentioned that prediction markets employ the price system of a market not for speculation, but rather for the primary purpose of collecting dispersed information from individuals. By incorporating human judgment and translating it into a numerical estimate, prediction markets can be classified as a judgmental approach (see. The Methodology Tree of Forecasting53). Moreover, it can be stated that the price system of the market aggregates qualitative information, or “knowledge of the kind which by its nature cannot enter into statistics.”54

Basically, there are three, already alluded, intertwined aspects that lead to a prediction market’s capability of yielding accurate and reliable predictions. First, the market mechanism is essentially an algorithm for aggregating information. Second, since the accuracy of information

52 see. Snowberg et al., 2005, p. 370. 53 see. Armstrong, 2001. 54 see. Hayek, 1945, p. 524.

22 is directly related to financial gain, there exists an incentive for truthful revelation. Third, and lastly, because of the existence of a market, traders have long-term incentives for specializing in discovering novel information and profiting from it. While it is true that any market adheres to these characteristics, other forecasting methods, such as polling or utilizing professional forecasters, lack at least one of them.55 For instance, polling lacks incentives for truthful revelation, and professional forecasters could have other motivations than just forecasting accuracy56.

5.2. How Prediction Markets Operate

I will start this section with a very basic illustration of how prediction markets operate in practice.

Figure 3 shows a rudimentary example of a prediction market in a practical setup. This market intends to forecast car sales for 2008.

Figure 3: Operation Principle of Prediction Markets

Notes: Reproduced from Luckner, 2008, p. 238.

55 see. Snowberg et al., 2012, p. 6. 56 see. Marinovic et al., 2013.

23 It features a widespread “winner-takes-all” contract which pays off $100 if the sales are between 500 and 600 cars in 2008, otherwise the pay-off is $0. The current market price for this contract, which can be understood as the probability of car sales totaling within 500 to 600 units in 2008, is $45.

Now, if a trader assumes with a likelihood of around 70% that sales in 2008 will (not) end up in that range, he or she should be willing to buy (sell) this contract at the current price level. Hence, he or she will be motivated to get involved if the current forecast differs from his or her own prediction.

Finally, it is presumed that the participant bought 20 shares of this contract at a price of $45, and that the actual sales for 2008 totaled 583 cars. Then, after contract settlement, this trader would have made a profit of $1,100.

As an example of how rapidly prediction markets manage to incorporate new information, I am going to have a look at occurrences surrounding the death of Osama bin Laden.

It was 10:25 p.m. Eastern time on May 1st, 2011, when Keith Urbahn, former chief of staff to Defense Secretary Donald Rumsfeld, tweeted “So I'm told by a reputable person they have killed Osama Bin Laden. Hot Damn.”

Figure 4: Information becomes quickly incorporated.

Notes: Reproduced from Snowberg et al., 2012, p. 7.

24 Figure 4 shows the reaction of a prediction market tracing the chance that Osama bin Laden would be captured or killed by December 31st, 2011. Within instants of Urbahn's announcement, the price of this former rarely traded contract started to increase. Within 25 minutes of Urbahn’s initial message, the probability implied by this contract had risen from 7% to almost 99%. In addition, it is noteworthy that this final quote happened eight minutes before any mainstream media outlet started reporting on bin Laden's death.

Furthermore, prediction markets have the ability to continuously accumulate new information. Figure 5 exhibits the accuracy of election forecasts at the Iowa Electronic Markets over time. The graph nicely demonstrates that as the election day approaches, and more information regarding expected outcomes gets exposed, the forecast error gradually decreases.

Figure 5: New information is continuously aggregated across time.

Notes: Reproduced from Snowberg et al., 2012, p. 7. (Data from the IEM, 1988-2000, available at http://www.biz.uiowa.edu/iem), adapted from Wolfers and Zitzewitz, 2004.

Moreover, prediction markets display little evidence of arbitrage opportunities.

Generally speaking, there are various approaches to discovering arbitrage openings. Firstly, an arbitrageur could look for price disparities among similar contracts across different exchanges

25 or different securities. Secondly, he or she could examine whether foreseeable patterns of price movements allow for arbitrage. Thirdly, and lastly, arbitrageurs might be able to profit by trying to exploit predictable deviations from rationality.

As it turns out, contracts traded on a single exchange are usually arbitrage-linked. For example, a rapid surge in the price of a contract tied to the victory of a certain political candidate commonly incites declines in the prices of contracts related to other candidates, indicating that the implied probability of someone winning the election stays relatively close to 100%. Figure 6 illustrates various contract prices tied to Arnold Schwarzenegger winning the Californian gubernatorial race of 2003. Here, a significant variation of prices at any given point in time is clearly visible, however, overall prices followed the same path.57

Figure 6: Prediction markets show very little arbitrage potential.

Notes: Reproduced from Snowberg et al., 2012, p. 8. (Data underlying lower line for each data source are bids, data underlying upper line for each data source are asks. Prices collected electronically every four hours by David Pennock), adapted from Wolfers and Zitzewitz, 2004.

57 see. Snowberg et al., 2012, pp. 8-9.

26 Additionally, prediction markets have proven to be fairly difficult to manipulate.

At the turn of the century, several influential political party members tried to manipulate the price of their candidates’ contracts on various gambling markets, but mostly failed to do so. The same outcome could have been observed in more recent attempts conducted by political candidates themselves. At last, it has been found that such manipulatory efforts aggravate only brief transitory effects in prices.58

In 1998, Camerer placed large bets in pari-mutuel horse racing markets only to cancel them moments before the race. His goal was to explore if this would generate a bandwagon effect of follow-on bets. It did not59. Ten years later, in 2008, Rhode and Strumpf placed random $500 wagers (the largest allowed) on the IEM and found that prices rapidly returned to pre- manipulation levels60. Moreover, observations from experimental prediction markets, employed under lab conditions, harvested similar results. In a first experiment, conducted by Hanson et al., some participants were incentivized to try to manipulate prices, but little evidence was found that these contributors succeeded61. A subsequent second experiment, which based incentives on whether observers of the market price could be manipulated, likewise saw little evidence that manipulators influenced the perceptions of observers62. In fact, by providing more liquidity, manipulators might even increase the accuracy of prediction markets63.

Besides these aforementioned observations and numerous other reports demonstrating that manipulators are generally not successful, there is one exception current research is aware of. The manipulation of a contract linked to Hillary Clinton's chance of winning the presidency in 2008, conditional on winning the Democratic nomination. A single whale (i.e., a large quantity trader) acquired contracts on Intrade.com at prices that implied Clinton was much more electable by the general electorate than her Democratic opponents. Although the manipulator achieved to keep prices high for a significant period of time, his actions eventually were detected and discussed by other participants, resulting in massive losses for the whale.

58 see. Rhode/Strumpf, 2008a. 59 see. Camerer, 1998. 60 see. Rhode/Strumpf, 2008b. 61 see. Hanson et al., 2006. 62 see. Oprea et al., 2011. 63 see. Hanson/Oprea, 2009.

27 Furthermore, it appears to be interesting that only one mainstream media station reported on this story.64

I am going to conclude this part by returning to the beginning of this chapter where I initially claimed that in most cases prediction markets seem to satisfy at least a weak form of efficiency, thereby implying that there is no evidence which might suggest that trading on past prices results in creating a profit. This has been explicitly demonstrated for prediction markets by Leigh, Wolfers and Zitzewitz in 2003, Tetlock in 2004, as well as Berg, Forrest and Rietz in 2006.

Leigh, Wolfers and Zitzewitz examined prediction markets related to the death of Saddam Hussein, discovering that an augmented Dickey-Fuller test cannot reject the null hypothesis that those markets follow a random walk. They also found that a Kwiatkowski-Phillips- Schmidt-Shin (KPSS) test rejects the null hypothesis that prices are trend-stationary.65 In addition, Berg, Forrest and Rietz similarly exposed that prices in the Iowa Electronic Markets follow a random walk66. Lastly, Tetlock revealed some evidence of mispricing in prediction markets concerning sporting events, including signs of over-reaction to news. Nevertheless, this mispricing was not large enough to allow for profitable trading strategies. Furthermore, Tetlock was not able to detect any proof of mispricing in prediction markets about financial events on the same exchange.67

More evidence regarding the efficiency of prediction markets can be found in the vast research literature on gambling markets. This literature, e.g. several chapters of Williams (2005) as well as Hausch and Ziemba (2007), confirms that betting markets are, in fact, for the most part weakly efficient.

5.3. Why They (sometimes) Fail

While it has been established by now that prediction markets generally perform quite well, there are certain design flaws which sometimes may prevent reliable predictions from occurring. These defects usually cause an unintentional lack of noise traders (or thin markets),

64 see. Zitzewitz, 2007. 65 see. Leigh et al., 2003. 66 see. Berg et al., 2006. 67 see. Tetlock, 2004.

28 which diminishes incentives for discovering and trading based on private information.68 But, what is a noise trader? A noise trader is simply an individual who may behave irrationally or sometimes erratic, which causes prices and risk levels to diverge from expected levels, even if all other traders are rational.

In an effort to entice desirable noise traders, the subject of a prediction market must be interesting, and information must be widely spread. Furthermore, prediction market contracts should be clearly specified, to prevent confusion regarding the payoff conditions. However, this specificity may conflict with offering an interesting contract to traders. For instance, in 2003, Intrade operated markets that asked, “Will there be a UN Resolution on Iraq (beyond #1441)?” and ”Will Saddam be out of office by June 30th?” The first question seems to be better specified, but the latter had much higher trading volume. Moreover, even the former question shows ambiguity: what does it mean for a UN Resolution to be “on” Iraq? and what does “(beyond #1441)” imply?69

Additionally, it is quite possible that noise traders may, rather rationally, elect not to trade in markets which contain a high degree of insider information. For example, despite the widespread intrinsic interest in who will be a Supreme Court nominee, markets on this decision have regularly turned out to be unsuccessful. This may be because most traders realize that there can realistically only be a few individuals with genuine information on who the President will select. Also, this story underlines the importance of prohibiting insider trading. For instance, consider a market forecasting the Institute for Supply Management's (ISM) business confidence measure, which would most likely not work if it was public knowledge that ISM employees were taking part in it.70

As an example of an extreme form of information not being broadly dispersed, I will present a case involving prediction markets tied to whether weapons of mass destruction (WMDs) would be discovered in Iraq or not. These markets projected that it is very likely such WMDs would be found. However, the untruthful confidence that could be encouraged by such an estimate, ignores the fact that there was no information at all being aggregated by these markets.

68 see. Snowberg et al., 2005, p. 372. 69 see. Snowberg et al., 2012, p. 11. 70 Ibid.

29 Meaning, it was improbable that anybody stationed in Iraq, who might actually have some valuable information about Iraq's WMD program, was trading in these markets.71

Finally, it remains uncertain to what degree prices in prediction markets may be affected by behavioral biases. The most popular, and well-understood, pricing anomaly in sports betting is the favorite-longshot bias. This pattern, the divergence between risk-neutral implied probabilities and actual probabilities, illustrated in Figure 7, derives from long-shots being overbet and favorites being underbet relative to their risk-neutral probabilities of success.72 Hence, interpreting the prices of wagers on horses as probabilities will tend to underestimate a horse's probability of victory when that probability is very high, and tend to overestimate a horse's probability of victory when that probability is very low.

Figure 7: The Favorite-Longshot Bias Pricing Anomaly in Sports Betting Markets

Notes: Reproduced from Snowberg et al., 2012, p. 12, adapted from Snowberg and Wolfers, 2010.

71 see. Snowberg et al., 2012, pp. 11-12. 72 see. Tompkins et al., 2008.

30 In Figure 8, recorded from Intrade’s U.S. election markets of 2004, a similar pattern can be observed. Nearly all the contracts that traded above a price of 50 won, and almost all of those that traded below 50 lost. This implies that those markets which forecasted a high probability of one candidate winning underpredicted the actual probability of victory, while those that projected a low probability of success overpredicted the actual probability.

Figure 8: The Favorite-Longshot Bias in Prediction Markets

Notes: Reproduced from Snowberg et al., 2012, p. 13. (Data pools 50 markets on the state-by-state outcome of the electoral college, and 34 markets on U.S. Senate races on Intrade.com, then: Tradesports.com.)

Current research has attributed the favorite-longshot bias occurrence in gambling markets to behavioral phenomena, with Jullien and Salanié ascribing it to asymmetries in the way traders value gains and losses73, and Snowberg and Wolfers assigning it to misperceptions of probabilities74.

Regardless of the real underlying reason, it is advised to use extreme caution when interpreting results based on contracts that imply a risk-neutral probability between 0 and 10%, or 90% and 100%.

73 see. Jullien/Salanié, 2000. 74 see. Snowberg/Wolfers, 2010.

31 5.4. Cutting-Edge Design

These anecdotes of failure, presented above, lead to a couple of simple rules for designing prediction markets. Firstly, the question needs to be well-defined. Secondly, there must be sufficient interest in the question to ensure liquidity. Thirdly, and lastly, the information regarding the subject of interest must be dispersed. Yet, it turns out that such basic guidelines leave much to be wanted. For instance, what if the question is intrinsically too difficult to describe, or there is not enough interest in the question? These problems and numerous other similar questions have been challenged by corporations employing prediction markets, and even though there are few academic documentations on their practices, I am going to try my best and summarize these experiences in this passage.

Businesses are paying attention to prediction markets because they have the potential to transfer unbiased information from a company's front-line employees to the top management75. Nevertheless, many inquiries that executives express cannot be considered widely interesting, nor are there many employees who may possess relevant information, which generates a lack of liquidity in markets and possibly leads to no trading or worse, erroneous forecasts. In an effort to solve this issue, Microsoft began using a market-making algorithm, which essentially is a variation of the market scoring rules described above76. The purpose of this market maker system in a prediction market is similar to that in any other equity market. It buys and sells contracts when other traders are not interested. However, unlike most market-makers, the primary goal of market-making algorithms is informational efficiency, and not to yield a profit, often leading the system to lose money.

Unconnectedly, Hewlett-Packard has experimented with a design, comparable to a pari-mutuel market, in which a few participants were required to take bets. This implementation is exceptionally interesting, since the market creators incorporated extra information of market participants, such as their social connections and risk-attitudes, to improve the market's efficiency. For instance, if three traders who the system identified as friends each wager massively on an identical potential result, the market interprets these bets as being based on

75 see. Strumpf, 2012, p. 3. 76 see. Berg/Proebsting, 2009.

32 redundant information and accordingly lowers the impact of those wagers on the overall price.77

Evidently, one of the most prevalent corporate use cases for prediction markets has been managing R&D portfolios. Typically, this is executed via so-called idea markets, which allow employees to purchase contracts of particular ideas in an effort to identify those ideas that are projected to be the most lucrative to the company. But, in this case only supply and demand determine the price of a contract. The actual outcome, such as profit if the product eventually happens to reach consumers, is not relevant. In essence, these markets, sometimes labelled preference markets, are very similar to pure beauty contests78, implying a looming risk of divergence between the actual question of interest and what the market is able to deliver. Indeed, General Electric has revealed that in their idea markets generally the creators of ideas trade aggressively to increase the price of their idea and decrease the price of ideas originating from competitors79.

In spite of the theoretical difficulties associated with such markets, as well as the absence of a coherent way to evaluate their performance, they are enjoying increasing popularity, which has induced a series of applications featuring highly diverse designs. Such experiments aim to inspect multiple fine details of a market’s structure, thus making it very difficult to properly summarize them. However, further details can be found in Lavoie (2009), Spears et al. (2009) as well as Dahan et al. (2010). Moreover, a remarkably detailed description of the design and evaluation of a prediction market for technology assessments can be found in Gaspoz (2010).

Finally, the last design restriction, that must be mentioned, is the uncertain legal and regulatory environment surrounding prediction markets80. Regulation (i.e., the law concerning prediction markets) generally tends to be ever-changing, sometimes ambiguously worded or undefined, and different for each jurisdiction. Therefore, corporations are usually well advised to consult with a professional purveyor of prediction market services, a list of which has been published in Berg and Proebsting (2009), before launching a new prediction market project. Additionally, a summary of the recent prediction markets’ legal situation can be found in Bell (2009).

77 see. Chen/Krakovsky, 2010. 78 see. Marinovic et al., 2010. 79 see. Spears et al., 2009. 80 see. Arrow et al., 2008.

33 5.5. Evidence on Forecast Accuracy

The most noteworthy attribute of prediction markets is undeniably the accuracy of their forecasts.

Over the recent past, the performance of prediction markets has been analyzed and compared to alternative methods for various fields of application. From areas such as sports, where Spann and Skiera (2009) examined the forecast accuracy of prediction markets for predicting the outcomes of German premier soccer league games, and similarly Luckner et al. (2008) analyzed data from the FIFA World Cup of 2006, with both studies concluding that prediction markets are more accurate than experts and perform equally well to betting odds, to other more exotic areas, such as the film industry and events that lie far in the future, with Pennock et al. (2001) studying data from the Hollywood Stock Exchange (HSX) and the Foresight Exchange (FX) respectively. For a complete overview and categorization of research in the field see Tziralis and Tatsiopoulos (2012), or MIT’s Handbook of Collective Intelligence edited by Bernstein and Malone in 2015.

In this segment, I will discuss the performance of prediction markets by illustrating various case studies of corporations as well as the political arena. However, I will begin with an analysis of the only, large-scale, prediction market tied to macro-economic outcomes, the earlier briefly addressed Economic Derivatives markets.

5.5.1. Macro Derivatives In October 2002, Goldman Sachs and Deutsche Bank began offering markets directly related to macro-economic events81. These markets, usually referred to as “Economic Derivatives”, allowed investors to acquire contracts with payoffs linked to economic indicators such as non- farm payrolls, retail sales, the levels of the ISM's manufacturing diffusion index (a measure of business confidence), and initial unemployment claims, among others. Since these payoffs were tied to certain data release dates, investors had the opportunity to practically hedge against specific event-based future risks. Furthermore, it should be mentioned that these markets only lasted for a couple of years, presumably due to a lack of sufficient trading volumes. Nevertheless, while they were running, Gürkaynak and Wolfers (2006) were able to study the

81 see. The Economist, 2002.

34 performance of these markets. A summary of their main findings will be given in this subsection, as described in Snowberg et al. (2012)82.

The results, gathered from their research, have been compared to a survey-based forecast released by Money Market Services, which typically averages predictions from a pool of roughly 30 forecasters, visualized in Figure 9. Here, it is noticeable that the market-based forecast is slightly more accurate than the survey-based forecast, and this can be verified by a numerical analysis.

Figure 9: Macro derivatives are slightly more accurate than survey-based forecasts.

Notes: Reproduced from Snowberg et al., 2012, p. 17, adapted from Gürkaynak and Wolfers, 2006.

Table 2 shows the mean absolute error and mean squared error of both forecast methods, normalized by the average forecast error from past surveys, thus allowing data sets across all four indicators to be sufficiently comparable to sum them all together in the fifth column.

82 see. Snowberg et al, 2012, pp. 16-19.

35 Table 2: Economic derivatives show somewhat smaller errors than forecaster.

Notes: Reproduced from Snowberg et al., 2012, p. 18, (**, * represent statistically significant Diebold and Mariano (1995) / West (1996) tests, implemented as recommended by West (2006), at the 1%, 5% and 10% level. Standard errors are in parenthesis. Forecast errors normalized by historical standard error of survey-based forecasts.), adapted from Table 1 in Gürkaynak and Wolfers, 2006.

It is recognizable that for each series the market-based forecast is more accurate than the survey-based prediction. Furthermore, across all sets, solely depending on the market-based forecast would have diminished forecast errors by around 5.5% of the average forecast error over the past decade. Due to the small amount of observations in each series, these differences seem to be barely statistically significant. However, the differences in the combined data sets are statistically significant at the 5% level.

Table 3 inspects the forecasting strength of every single predictor on his own. The first panel displays the correlation of each prediction with the real outcomes. We can detect quite high

36 correlations across the board. Hence, it is evident that both sources possess considerable unconditional forecasting power.

Table 3: Macro-economic derivatives exhibit slightly smaller errors than experts.

Notes: Reproduced from Snowberg et al., 2012, p. 20, (***, **, * denote statistically significant coefficients at the 1%, 5% and 10% level with standard errors in parenthesis. Forecast errors are normalized by historical standard error of survey-based forecasts.), adapted from Table 1 in Gürkaynak and Wolfers, 2006.

Panel B implements Harvey, Leybourne and Newbold (1998) tests of forecast encompassing and discovers that the market does not encompass the survey for Business confidence as well as initial unemployment claims. Moreover, it can be dismissed that the survey encompasses the market for business confidence. Additionally, the p-values on the test regarding the null hypotheses that the survey encompasses the market for retail sales as well as initial unemployment claims are 0.11 and 0.16 respectively. Therefore, it can be concluded that both the market and the survey provide some degree of exclusive information.

37 The final panel follows Fair and Shiller (1990) in performing a regression-based test of the information content for each forecast. The results are astounding. It turns out that for all four indicators the coefficient on the market-based prediction is statistically indistinguishable from one, and that the coefficient on the survey-based forecast is only statistically dissimilar to zero in a single data set, where it is (abnormally) negative, which implies that conditioning on the market-based forecast renders the survey-based prediction de facto useless. Furthermore, analyzing the combination of all four data sequences only supports these observations.

While it has been revealed that the market-based predictions fare better than those of surveyed forecasters, according to most of the statistical criteria inspected above, it could very well be that these criteria may not be the most desirable. To this day, it remains unanswered whether a specific trading strategy, similar to Leitch and Tanner (1991), would perform better or worse using information from forecasters or prediction markets.

5.5.2. Politics Prediction markets first gained global recognition for their capability to accurately forecast political events.

In 1988, academics at the University of Iowa introduced markets following numerous candidates' vote shares and winning chances. This research project, called the Iowa Electronic Market (IEM), rapidly demonstrated to be exceptionally accurate.

Figure 10 compares gathered information of the IEM with predictions of polls from Gallup. It illustrates that markets performed only marginally better than polls the day before the election. However, in the run-up to the election, it is statistically verified that markets are more accurate than polls. In fact, during this period, markets have half the forecast error of polls 83.

Regardless of these observations it needs to be mentioned, as Erikson and Wlezien (2008) did, that there are well-known biases in polls which can and should be accounted for.

For instance, candidates typically see their poll numbers spike immediately after their party's convention.

83 see. Berg et al., 2008.

38 Figure 10: Prediction markets perform better even the night preceding an election.

Notes: Reproduced from Snowberg et al., 2012, p. 21, (Market forecast is closing price on election eve; Gallup forecast is final pre-election projection).

Figure 11 outlines Rothschild’s (2009) comparison of de-biased forecasts from polls with prediction market prices, adjusted for the favorite-longshot bias, over the 2008 Presidential election.

To de-bias polls, Rothschild (2009) follows Erikson and Wlezien’s (2008) method, which finds the optimal projection from polls at the same point in previous election cycles and eventual elections, and then applies that projection to polls from the current electoral cycle.

Although during the first 70 days of the election cycle, the mean squared error of polls is noticeably higher, sometimes statistically significantly so, in the second half of the observation period polls and markets switch places (twice), inferring that one method is not undoubtedly superior to the other.84

84 see. Rothschild, 2009.

39 Figure 11: Prediction markets are usually more accurate than polls, even after excluding known biases.

Notes: Reproduced from Snowberg et al., 2012, p. 22, (Each data point is the difference in mean-squared error between two different types of forecast over the 50 electoral college races in 2008. Each forecast is produced by taking raw data from a poll or a prediction market and utilizing the most efficient transformation from raw data into forecasts, as outlined in Rothschild, 2009).

However, as revealed in Table 4, in Fair and Shiller (1990) type horse-race regressions, prediction markets capture all the information of polls when contracts that quote a likelihood of 90% of one or the other candidate succeeding are removed from the sample. If these markets are included, the coefficient on the estimate of the prediction market is still statistically significant and near one, but the coefficient on de-biased polls is statistically different from zero.

Taking into consideration all the evidence presented here, it can be concluded that raw prediction market prices deliver better forecasts than raw poll data, but polls may hold supplementary information, specifically in races where one candidate profoundly leads.

40 Table 4: A Comparison of Information from Prediction Markets and Polls

Notes: Reproduced from Snowberg et al., 2012, p. 22, (*** denotes statistical significance at the 1%, 5% and 10% level with robust standard errors in parenthesis. Table uses the data from Rothschild (2009), with prediction market prices converted into expected vote shares using past trends. Less-certain races are those where neither candidate has a (projected) 90% chance of winning. More details of the procedure for de-biasing polls and prediction market prices can be found in Rothschild, 2009).

5.5.3. Business It has already been exposed that corporations have frequently employed prediction markets to ameliorate their internal forecasts, and yet there is only sparse academic record of their ultimate performance.

In this section, I will discuss a few exceptional use cases of corporate prediction markets which have been recognized on an academic level.

In 1998, Ortner described an internal prediction market at Siemens, where participants correctly projected the delay of a software development three months ahead of the scheduled target date85. Furthermore, an article published in IEEE Spectrum reported similar discoveries for an internal market at Microsoft in 200786.

In 2002, Chen and Plott reported on eight prediction markets within Hewlett-Packard, which tracked important indicators such as the quarterly printer sales. They found that the market performed better than the official forecasts of the company in 6 out of 8 times, even though at

85 see. Ortner, 1998. 86 see. Cherry, 2007.

41 the time the official forecast was announced, the markets had already been closed, rendering prices transparent and unchangeable to all participants.87

Separately, Cowgill, Wolfers and Zitzewitz (2009) examined data from 270 internal markets run by Google. Their research revealed that in general those markets performed quite accurately, in a few market prices, however, perceptible biases could have been observed. Specifically, when markets revolved around Google’s performance as a business, participants appeared to overestimate optimistic outcomes. Additionally, they found that this optimistic bias was more distinct among traders who were recently hired, and on days when the company's stock price soared.88

Lastly, Berg, Neumann and Rietz (2009) instigated numerous prediction markets to foresee Google's market capitalization at the closing of its first trading day. The results have been compared to the auction process that Google exerted in setting its IPO price. Prediction markets did fairly well. Their aggregated forecast was 4% above the actual market capitalization, while the IPO price quoted 15% below the true value. Therefore, it can be theorized that if the corporation had set its IPO price based on the prediction market estimate, they would have accumulated around $225 million more in their IPO.89

87 see. Chen/Plott. 2002. 88 see. Cowgill et al., 2009. 89 see. Berg et al., 2009.

42 6. Outlining the Research Methodology

As previously mentioned, the core of this work will be a comparison between an example of a centralized- and a decentralized prediction market for the purpose of determining if the completely new, and potentially revolutionary, concept of incorporating the blockchain technology into prediction markets, performs better than a similar centralized counterpart. To accomplish this task, academics traditionally choose to gather sufficient data sets, normalize them (i.e., make them comparable), and try to find significant differences in regard to accuracy or other measurable indicators. However, in this case, mainly due to a lack of obtainable data, such an empirical analysis will not be feasible.

In this part of the paper, I am going to outline the research methodology for the main analysis and list explanations why I decided to do so.

Although prediction markets have been around for decades, it is a fact that as a result of constant technological innovation, they continue to improve and evolve. In today’s information age, it is widely known that crowdsourced data is a very valuable asset. Therefore, it is no surprise that rather than transform it into public knowledge, corporations and hedge funds keep their gathered information undisclosed, to analyze or monetize it themselves. This infers that even though a handful of online crowdsourcing-based investment research platforms exist, it is very difficult to obtain suitable data from those centralized prediction markets specialized in financial assets.

The same issue applies to decentralized markets, since at the start of my research in April 2018, a fully functioning decentralized prediction market was not to be found. Due to these complications, I realized early on that a quantitative analysis will not be achievable, thus forcing me to look out for alternative methods to compare both concepts.

In such a position, the best solution remains an investigation of the literature, where theoretical advantages as well as disadvantages are presented and discussed. In addition, several qualitative methods, designed to compare multi-attributes projects, can be applied to complement the theoretical examination’s results.

43 The research method I have ultimately chosen is called a utility value analysis90. This qualitative approach seeks to analyze a number of complex alternatives for the purpose of ranking them, according to the preferences of the decision-maker, in a multidimensional target system. The ordering is carried out by calculating so-called utility values for each of the different alternatives.

In practice, this subjective research method consists of the following five steps. Firstly, each target criterion needs to be specified. Secondly, each criterion obtains a weighting, between 0 and 1, which correlates to the perceived importance of that criterion to the overall value. Moreover, it needs to be added that the sum of the combined weightings should total 100 percent. In the third step, the ability of every alternative to fulfil each attribute is evaluated and a corresponding partial utility value, usually ranging from 1 to 100, is assigned. Subsequently, these partial utility values are weighted and summed to receive one total value for each alternative, the utility value. The main reason for this procedure is that it allows unfavorable results on one target measure to be compensated by better results on others, leading to a harmonization of end results. Finally, these end scores can be ranked and interpreted.91

Based on this information, my assignment became clear. I am going to conduct a literature research, where I will try to spot similarities as well as differences and discuss main features in greater detail. This will be accompanied by a utility value analysis of Vetr, one example of a centralized prediction market for financial assets, and Augur, currently the biggest decentralized prediction market project available, conducted from the perspective of a consumer.

The reason why I elected a user’s perspective was to further objectify the rather subjective results that will be gathered in the comparison part. This is true, since I am convinced to be qualified enough to represent an average individual, who is interested in forecasts of financial assets, which ensures similar preferences.

Finally, I plan on concluding this research by interpreting the new-found insights and discuss further implications.

90 see. Götze et al., 2008, pp. 175-179. 91 see. Götze et al., 2008, pp. 175-176.

44 7. Centralized Prediction Markets for Financial Assets

This chapter commences with a brief investigation of currently available centralized prediction markets aimed at satisfying the investment community, and proceeds with an in-depth analysis of Vetr92.

7.1. General Overview

Principally, it can be said that within the globally accessible online arena, a variety of all slightly differing centralized crowdsourcing-based investment platforms do exist, but most of them operate on a give-to-get model, which makes them unsuitable for my upcoming comparison.

For example, Estimize93, a financial and economic data provider, specialized in accommodating institutional investors and quantitative researchers, where you can either pay for its services or contribute proprietary research data sets to gain access. Furthermore, there is ClosingBell94, another centralized prediction market provider that grants access to its buy-side research data, if you connect an active Robinhood95 brokerage account to its application.

For a full list of centralized applications please consult “Business Opportunities in Crowdsourced Stock Market Analysis”, a master thesis by Lauri Kolmonen written in 2017. (Unfortunately, this interesting paper is only physically accessible at the Aalto University in Finland, therefore I am not able to properly verify this suggestion.) Moreover, the TechCrunch article “Why Has Social Failed in Fintech?” by Shane Leonard96 contains some valuable information.

7.2. Vetr

Taking into consideration all the restrictive attributes, I have decided to choose Vetr, a financial technology start-up that harnesses the wisdom of the crowds to help improve investors’

92 see. Vetr, 2018: https://www.vetr.com/. 93 see. Estimize, 2018: https://www.estimize.com/. 94 see. ClosingBell, 2018: https://closingbell.co/. 95 see. Robinhood, 2018: https://robinhood.com/. 96 see. Leonard, 2015.

45 financial literacy, as the universal representative for all centralized prediction markets focusing on financial assets.

The reasoning for my choice is twofold. Firstly, it wouldn’t have made much sense, nor does it seem practically possible, to put all the diverse centralized applications into one basket and aim to derive veritable general statements about their performance from this. Secondly, I was specifically looking for an innovative centralized application that is accessible to potentially anybody free of charge, for the primary goals of achieving comparability with Augur, and being able to inspect the application from a user’s point of view.

Vetr fits all these minimum criteria quite nicely. It features, the innovative component by incorporating gathered social information into the process of vetting stocks and exchange- traded funds, as well as a free access to its service after signing up, ultimately providing the highest achievable comparability with Augur, where theoretically anybody with an internet connection can create a market for basically any asset imaginable.

The logic behind this is to seemingly eliminate all differences between centralized- and decentralized prediction markets for financial assets except for one, the blockchain technology. This implies that if I find any performance differences, in my case from a participant’s perspective, it can be assumed they derived from incorporating the new technology into the “old” concept of prediction markets.

7.2.1. History of the Company In 2013, baby-boomer Jan van Eck and millennial Savneet Singh (unconfirmed97) shared the same frustration with the apparent information overload and complexity of navigating current financial markets. The underlying problem, concerning this situation, is that usually it appears to be very difficult and/or time-consuming for the self-directed investor to cut through the noise (i.e., to identify and separate valuable- from redundant information) in order to make an educated decision afterwards.

It is common knowledge within the investment community that sell-side (i.e., Wallstreet experts) Buy-, Sell- and Hold recommendations generally tend to be biased and slightly too optimistic, ultimately leading to inaccurate price targets for the buy-side end consumer.

97 see. Gust, 2018: https://gust.com/companies/vetr0/.

46 Furthermore, analysts stock coverage appears to be very narrow and limited to a few main stocks at the large end of the market, seemingly ignoring independently generated and dispersed information of numerous other large- and midcap companies. Given all these features, one can see how investors sometimes struggle to identify a delta and transform it into a positive alpha.98

Fed up with the status quo, the two prospective founders decided to collaborate on a unique value proposition of offering easy to understand, crowdsourced stock- and ETF ratings to the general investor, in order to “VET” financial assets togethe”R”. The result was the creation of a private and profit-oriented financial start-up named “Vetr”, headquartered in New York City.

Soon thereafter, in its first and only seed funding round on the 28th of July 2013, an unnamed business angel invested 1 million USD, and the initial idea was able to be transformed into a monetizable business model.99

In 2016, Vetr collaborated with the Massachusetts Institute of Technology (MIT) Media Lab to analyze its data and try to refine its existing approach. In some cases, they managed to achieve an accuracy of more than 99%. According to Mike Vien, CEO of Vetr from April 2016 to June 2018, “the results from that research were amazingly accurate, as low as 0.3% of an error”100. Unfortunately, this claim is unverifiable, since I could not find any raw data from that study, in fact, I was not able to find the study at all, which leads me to the assumption that a nondisclosure obligation was part of the cooperation agreement.

Moreover, on April 20th in 2017, Vetr gave a presentation on the growth of its business at the Demo Day 4.0 hosted by FinTech Sandbox in Boston101. Figure 12, adapted from this presentation, reveals the firm’s performance over the prior two years. Here, it should be mentioned that information provided in a pitch presentation is usually biased and numbers tend to be slightly inflated or sugar-coated, hence researchers must always be cautious when relying on those figures. As it turns out, user growth has significantly slowed down and Vetr’s projections appear to be somewhat overstated, because according to Web Traffic, an information service provided by SimilarWeb, Vetr had around 21,028 monthly web visitors in

98 see. Vien, 2017. 99 see. Crunchbase, 2018. 100 see. Vien, 2017. 101 see. Vetr Blog, 2017.

47 November of 2018, which infers a compound annual growth rate (CAGR) for registered participants of around 42% since the beginning of April 2017102.

Figure 12: Vetr’s User Growth and Development of Aggregated Ratings

Notes: Figure 12 is a combination of 3 slides from Vetr’s pitch deck in Boston on April 20th of 2017. Retrieved on the 12th of December 2018, via http://blog.vetr.com/.

Nevertheless, these are still respectable growth figures, and it must be acknowledged that this recognition has been accomplished using very little marketing efforts, thus almost entirely through organic growth. Furthermore, at the 2017 meeting in Boston, Mike Vien revealed that the company just acquired its first paying customer, as well as being in discussion with 5 more prospective clients. Therefore, it can be concluded that during the last three years a noteworthy expansion has been visible, and Vetr’s business model can be characterized as legitimate and worthwhile.

7.2.2. Service Description Generally, a successful business model consists of 4 properly interacting constituent parts; the customer value proposition (CVP), key resources as well as key processes, and the profit formula.103 In this segment, I will discuss each element from Vetr’s point of view.

Vetr’s CVP becomes apparent when reading its mission statement; “Our mission is to create a community for people to easily share insights and get meaningful information about the stock market”104. As a result, the firm offers a community-driven rating application specialized in

102 see. Crunchbase, 2018. 103 see. Johnson et al., 2008, p. 62. 104 see. Crunchbase, 2018.

48 financial assets, which is comparable to services such as Yelp (small enterprises), TripAdvisor (travel), Glassdoor (job search), or Amazon’s product rating feature. On Vetr, users are able to make predictions about future asset values, look up specific stocks, add those to their watch- lists to receive alerts, and review ratings from other participants who have shared insights on assets they are interested in.105

When a user submits a prediction as well as a specific time horizon, the system calculates a crowd target price, and attaches a rating, based on the current- and past performance of each individual, to each forecast, to let participants know how insightful a certain estimate is. Users, who perform remarkably well, are rewarded through social recognition and eventually their reputations on the site become featured as “Top Rater”106. Additionally, after providing a histogram of all the forecasts each member posted, Vetr offers the customer the ability to change their price target. Essentially, what Vetr does, is identifying the best predictors and people who are better at learning than others from the crowd, and subsequently filters this information to improve the overall accuracy of predictions. The result of this process is what Vetr refers to as an “intelligent star rating”, which is created through submitted predictions of participants that are improved via Vetr’s prediction algorithms.107

This dynamic interaction of people on the platform, combined with incorporating their status information into Vetr’s algorithms, can be described as the company’s main key processes, because it allows the centralized institution not only to obtain dynamic real-time asset predictions, but also to produce massive amounts of potentially monetizable meta data.

As a consequence, this proprietary meta data can be regarded as the main key resource of Vetr’s business model, since it provides the firm with raw data ready to be explored while at the same time offering a way of generating income.

More precisely, Vetr’s profit formula intends to sell the aggregated data to asset management firms and hedge funds, who aspire to integrate those data sets into their own quantitative models and trading strategies, for a small monthly fee.

105 see. Vetr Story, 2018. 106 see. Shrier et al., 2016, p. 9. 107 see. Vien, 2017.

49 7.2.3. User Interface Now, after it has been established that Vetr legitimately serves a customer’s need, I am going to become familiar with Vetr’s user interface by simulating one basic interaction process between a customer and the platform.

I have already discussed the options a new participant attains after signing up to Vetr’s service via email-address, Twitter, StockTwits, Facebook, or LinkedIn. But, how does it all work in practice?

Vetr’s service relies on a connection to the internet and, besides them offering the service in an explorer window, it can be consumed via several available mobile applications. In my scenario, I am interested in information about Apple Inc.’s stock (AAPL), and I plan on making a prediction regarding its target price on the webpage.

The first step, I take, is to connect to the internet and go to https://www.vetr.com/. There, I type “AAPL” in the search bar at Vetr’s home screen. After pressing “ENTER” I get transferred to the Apple specific information screen, illustrated in Figure 13. Here, on the top left, I can see Apple’s actual stock price as well as the date.

Figure 13: The Upper Half of Vetr’s Information Screen about Apple Inc.

Notes: Retrieved from https://www.vetr.com/, on the 13th of December 2018.

50 Right below, there is the crowdsourced star rating, plus recommendation, and the aggregated target price. Next to the crowd’s target price on the right, I can find the analysts target price, which basically is an average of all the sell-side experts who cover the stock on Wall Street. This is an important number, because it allows me to potentially identify that aforementioned delta by comparing the crowdsourced price (buy-side) to the analyst’s target price (sell-side) and subsequently to my expectations about the future value of the asset. Further to the right, I can view the overall accuracy of predictions from 1-10, and the corresponding overall rank, as well as a field to enter my personal estimates. The two charts in the center show Apple’s historical price on the left, and a projected price path, which incorporates all the green dots (bullish forecasts) as well as red dots (bearish estimates), on the right side.

If I now enter a price target and a time horizon of either 1-, 3-, or 6 months into the system, I will receive an instant star rating for my prediction. If I continue to post it, I get transferred to a screen that resembles Figure 14. Here, I can see a histogram of all registered Apple stock forecasts and get asked if I would like to change my recently submitted quote. Furthermore, I am requested to rate my confidence in the prediction made on a scale from 0 to 10.

Figure 14: Vetr’s “Special Sauce”

Notes: Retrieved from Vetr’s 2017 company pitch in Boston.

51 I could change my price target or not, in either case, the application returns me back to the information screen, of which the bottom half is visualized in Figure 15.

At this point, mainly the social aspect of Vetr’s service is being featured. I can browse members and see their ratings (i.e., reputation). Additionally, I can reveal what they are interested in, read and write comments, sort and filter users by various parameters such as accuracy or success rate, and follow each participant individually. Furthermore, I may read current headlines and see who is currently buzzing about the stock, look up information of fundamentals and earnings for AAPL, as well as read research reports, if I have proven my willingness to contribute to the community by sharing my own insights beforehand.

Figure 15: The Lower Half of Vetr’s Information Screen on Apple

Notes: Retrieved from Vetr.com on the 13th of December 2018, (The member account features next to the “Headlines” (center) two more options, named “Earnings” and “Fundamentals”, which are not visible in this Figure).

In summary, it can be stated that using Vetr’s service is simple and intuitive, meaning that it is easy to navigate. Furthermore, the service offers a lot of valuable information to individual investors, who can’t afford the monthly Bloomberg terminal service fee108 of around $2000, even if you are a nonsubscriber, therefore essentially being free of charge.

108 see. Seward, 2013.

52 8. The Blockchain Technology

In this thesis the blockchain technology represents the main difference between Vetr and Augur, therefore, as intended, metaphorically connecting centralized prediction markets for financial assets with their decentralized adversaries. The following chapter features a closer look at this innovative concept, while trying to shift the focus onto the blockchain technology that runs Augur and financial asset transactions.

8.1. Description

Although the practical implementation of this technology is very complex and the adaptation process, regarding people’s mindsets, the physical infrastructure, and the regulatory environment, demands a lot of time, the idea behind the concept is as simple as it is ingenious.

A blockchain is a distributed and constantly growing digital list of immutable transaction records, called blocks, which are chronologically connected using cryptography. Once a group of transactions is verified and securely recorded, a block is closed and added to the chain of existing blocks in a linear order. Each block contains a cryptographic hash (i.e., an image) of the previous block for reference, a timestamp, the block’s own hash, and the difficulty statement.109 With the help of the right protocols the blockchain software then automatically determines when each unit of value was transferred, preventing the long-standing problem of double spending in an environment without central oversight. Moreover, once recorded, the data in any given block cannot be retroactively changed without alteration of all succeeding blocks.

In other words, the blockchain is a distributed ledger that can efficiently record transactions between two parties in a verifiable and permanent way, without the need for an intermediary or a centralized authority to update and maintain the information generated by those transactions.110

Distribution is typically achieved through a peer-to-peer network of computers (nodes), which collectively adhere to a single inter-node communication protocol, allowing participants to

109 see. Raval, 2016, pp. 1-2. 110 see. Iansiti/Lakhani, 2017, p. 4.

53 validate as well as audit transactions independently and relatively inexpensively. This verification process is conducted either by miners or alternative solutions, aimed at reaching consensus on the latest blockchain version or ledger balance. In exchange for this proof of work, miners usually receive a reward and possibly transaction fees.111

In general, it can be stated that such a design facilitates robust workflow in cases where participants' uncertainty concerning data security is marginal. Figure 16 illustrates a basic implementation process of how the blockchain technology can help to approve and execute a transaction of funds.

Figure 16: A Blockchain Visualization

Notes: Adapted from Rennock et al., 2018, p. 37. (Source: The Financial Times)

Additionally, a blockchain can be described as a value-exchange protocol, that, through eliminating the middleman and enabling a direct contractual interface between the two parties involved in a transaction, aims to be faster, safer and cheaper than any traditional alternative112.

111 see. Catalini/Gans, 2016, p. A-2. 112 see. Twesige, 2015.

54 Furthermore, a blockchain can maintain title rights, because when correctly set up, to detail the exchange agreement, it provides a record that compels offer and acceptance113.

Finally, probably the most significant characteristic of the blockchain technology, its integrated model of deflation, must be discussed.

The use of a blockchain removes the aspect of infinite reproducibility from a digital asset, since the number of computable blocks is limited to a certain threshold. For example, if we consider the mother of all cryptocurrencies, the Bitcoin (BTC), it can be observed that as the size of each block increases the number of Bitcoins being produced decreases over time, eventually being condemned to stop working once supply peaks at 21,000,000 coins, rendering its value de facto deflationary.114 This element opens up numerous discussable questions about how to properly behave in an environment that represents exactly the opposite of the incumbent inflation-based economic model of today’s western societies.

8.2. Brief Historical Background

Although the first description of a cryptographically secured chain of blocks was laid out by Stuart Haber and W. Scott Stornetta in 1991, the real story of the blockchain begins with the conceptualization of the first cryptocurrency by a person (or a group of people) referred to as Satoshi Nakamoto in 2008. By using a hashcash-like method to add blocks to the chain without requiring them to be signed via a trusted third party, Nakamoto managed to improve the existing design in an important way115. One year later, his design was implemented as a core component of the cryptocurrency Bitcoin, where it continues to serve as the public ledger for all transactions on the network to this day116. Since then, Bitcoin’s blockchain file size, which contains the records of all transactions that have occurred on the network, grew to around 210 gigabytes (GB) at the end of March 2019117.

Nowadays, besides Bitcoin, thousands of other blockchain applications are available. Basically, it is a complex jungle of technical terminologies, a media hype, false promises and misconceptions. This obscure situation may be one of the main reasons why the current reality

113 see. World Blockchain Foundation, 2018. 114 see. Medium, 2018c. 115 see. Narayanan, 2016. 116 see. Nakamoto, 2009. 117 see. Statista, 2019.

55 of the blockchain technology ecosystem appears to be rather humbling. In a recent survey, conducted by the global research company Gartner in 2018, only 1% of CIOs indicated any kind of blockchain adoption within their organizations, and only 8% of CIOs were in the short- term “planning or [looking at] active experimentation with blockchain.” Furthermore, an astonishing 77% of CIOs said their company “has no interest in the technology and/or no action planned to investigate or develop it.”118 If we now consider an average Bitcoin dominance of around 54% regarding all daily transactions, and account for 2071 different cryptocurrencies with a combined market cap of around $112.5 billion on December 18th 2018119, it makes one wonder how valuable each of these different blockchain implementations really is.

8.3. Types of Blockchains

Presently, it can be differentiated between three types of distributed ledger technology networks. First, a private blockchain, which usually can be found within companies who wish to incorporate the blockchain into their own accounting- or record-keeping procedures without sacrificing autonomy and running the risk of exposing sensitive data to the public internet. This is possible because the private blockchain is permissioned, which infers that individuals can only join the “intranet” if they are invited by a network administrator.120

The second type is called a consortium blockchain. This network is also permissioned, but instead of a single organization controlling it, several companies may each operate a node on such a network. The administrators of a consortium chain can restrict users' reading rights and limit the number of trusted nodes that can execute a consensus protocol. Therefore, a consortium blockchain can be characterized as being a semi-decentralized network.121

The third type, the public blockchain, is the most common, hence it is the blockchain I am going to shift my focus on. It is “permissionless”122, which implies that there are no access restrictions, thus theoretically anybody with an internet connection should be able to access the network, send transactions, and participate in the execution of a consensus protocol (i.e., act as

118 see. Artificial Lawyer, 2018. 119 see. CoinMarketCap, 2018. 120 see. Marvin, 2017. 121 see. Zheng et al., 2017, p. 559. 122 Ibid.

56 a validator). Currently, the two largest public blockchains are the Bitcoin and the Ethereum (ETH) blockchain.

This section will be concluded by listing key characteristics which all commercial transactions using blockchain technology have in common.

Besides the aforementioned real-time record keeping and the immutability of records, the blockchain technology also makes it easier for network users to be pseudonymous, which has direct ramifications for operators of networks, who are subject to anti-money laundering (AML) and know-your-customer (KYC) regulations.

Moreover, blockchain applications share a constantly looming cybersecurity risk. While no blockchain has been successfully hacked or manipulated yet, the companies and the technological infrastructure surrounding them certainly have been. Even though the decentralized structure of blockchain networks makes them more resilient against network- wide attacks or tampering, incidents involving security breaches range from banal service disruptions to more serious crimes such as thefts of sensitive data or valuable cryptocurrencies.

Finally, it should be mentioned that dealing with blockchain transactions which involve virtual currencies can lead to unexpected tax consequences, depending on how the appropriate tax authority treats a certain virtual asset. Some jurisdictions, like the US, treat virtual currencies as property, which means that a transaction may create the need to recognize a gain or loss on the exchanged cryptocurrency in an individual’s tax form (see. IRS’s Tax Treatment of Virtual Currencies).123

8.4. Defining Cryptocurrencies: Coins versus Tokens

In theory the blockchain can be used for any type of transaction, including data sets or other types of values, assets, smart contracts, agreements and, of course, cryptocurrencies.

Cryptocurrencies are virtual currencies typically powered via a blockchain and secured using cryptography. Although several cryptocurrencies have existed prior to Bitcoin, the creation of Bitcoin, the first decentralized cryptocurrency, precipitated the expansion of a diverse ecosystem of numerous coins and tokens which are altogether often regarded as

123 see. Rennock et al., 2018, pp. 36-37.

57 cryptocurrencies. However, the majority of them does not fall under the academic definition of a currency, since technically a currency represents a unit of account, a store of value, and a medium of exchange. Yet, all these features are inherent within Bitcoin, and because the cryptocurrency environment was practically kickstarted by Bitcoin’s creation, any other coins conceived after Bitcoin are frequently wrongfully considered as a cryptocurrency.124

Generally, cryptocurrencies can be categorized into two groups: Alternative Cryptocurrency Coins (Altcoins) and Tokens.

Altcoins, or just coins, are simply alternatives to the Bitcoin. They can further be subdivided into altcoins, which are built using Bitcoin’s open-sourced original protocol and changing its underlying code (i.e., forks of Bitcoin), as well as coins, which derive from independently created blockchains and protocols, supporting their native currency. Examples for variants of Bitcoin’s code are the very first altcoin, the Namecoin, as well as Peercoins, Litecoins, Dogecoins, and Auroracoins. While examples of coins using their own blockchain include Ethereum, Ripple, Omni, Bitshares, NEO, Waves, and Counterparty. Hence, it can be summarized that each altcoin possesses its own unique and independent public blockchain, which solely processes transactions relating to its native coin. Additionally, alternative cryptocurrency coins may be sent, received or mined, but they are not meant to perform any other functions beyond acting as a currency.125

The second type, commonly wrongfully referred to as cryptocurrency, are tokens. Basically, tokens are a representation of a certain asset (security tokens) or utility (utility tokens) which typically resides on top of another blockchain. Theoretically, tokens can represent any asset that is fungible and tradeable. Examples reach from commodities to loyalty points to other cryptocurrencies.

The creation of tokens is made possible by employing smart contracts, which are programmable computer codes that are self-executing and do not need any third parties to fully operate. Project developers can follow a standard template on a certain blockchain, such as on the Ethereum or Waves platform, and easily create and distribute their own tokens, which then either could be used as a method of payment inside a project’s ecosystem (utility token) or as

124 see. Aziz, 2018. 125 see. Medium, 2018a.

58 a share of the issuing company (security token). Moreover, a template for token creation provides a standard interface for interoperability between tokens, making it much easier for individuals to store different types of coins within a single digital wallet. One widespread example of such a template is the ERC-20 standard on the Ethereum blockchain, which is used by over 1000 tokens, including, a token I am going to become more familiar with in the upcoming chapters, the Reputation token (REP) of the decentralized prediction market platform Augur126.

Furthermore, after tokens are created, they may be distributed to the public via an Initial Coin Offering (ICO), which is a method of crowdfunding through the release of a new cryptocurrency or token to fund a project’s development. This process is comparable to an Initial Public Offering (IPO) for stocks, although less regulated, and with a few other critical distinctions that have to be made127. Nowadays, numerous, especially tech-savvy and young, individuals prefer to invest in ICOs rather than IPOs, as they offer a rapid and internationally accessible way to invest any amount of money into interesting projects that could possibly yield tremendous financial returns.

In conclusion, it can be stated that the main difference between altcoins and tokens lies in their structure. While altcoins are independent virtual currencies with their own separate blockchain, tokens operate on top of a blockchain and may represent a company’s share, give access to products and/or services, or perform several other functions within a project’s network. It seems that the phrase “You can buy a token with a coin, but not vice versa!” legitimately summarizes this section quite nicely.

126 see. Eidoo, 2019. 127 see. Cointelegraph, 2019.

59 9. Decentralized Prediction Markets for Financial Assets

Finally, a solid knowledge foundation to combine prediction markets with the blockchain technology has been established, thereby enabling me to properly analyze decentralized prediction markets for financial assets.

First of all, since decentralized prediction markets theoretically allow anyone to create a market for any imaginable question regarding the future state of basically anything, all of them appear to be equally “specialized” in financial assets and, therefore, each qualifies as a potential comparison candidate. However, the Augur project was the first, is the largest by market cap, and represents the purest as well as the most transparent approach of a decentralized prediction market platform, hence erasing any doubt about what project to ultimately choose.

9.1. General Overview

In this section, I am going to take a brief glance at the current decentralized prediction market landscape, before continuing with an in-depth examination of Augur in the next segment.

Nowadays, the German project Gnosis128, which can be regarded as Augur’s number one competitor, positions itself as an entire platform for building decentralized prediction market applications, focusing on the creation of actionable information that can be utilized by human and AI decision-making agents. It appears that after losing the first-mover race to Augur, the project decided to expand its current business model into new products and services to complement its existing Ethereum-based prediction market platform. Furthermore, Gnosis provides third party developers with tools to build new prediction market applications on top of Gnosis, since its team believes that different categories of prediction markets should have different trading interfaces as well as specialized marketing- and regulatory strategies. Ultimately, this unique focus legitimizes Gnosis’ subsistence, and offers the opportunity to unlock a completely new and diverse world of indirect applications, such as insurance or asset hedging, for prediction markets.129

128 see. Gnosis, 2018: https://gnosis.io/. 129 see. Stox Whitepaper, 2018, pp. 33-34.

60 The second project, Stox, is also hosted on the Ethereum blockchain, and its team, the individuals behind invest.com, arguably has the best track record of understanding financial markets. As a result, Stox was specifically designed to serve as a practical framework for mainstream investments in prediction markets. However, for Stox to offer a competitive and reliable service, the trade engine must be practical along with being able to operate at a velocity that satisfies its customers. Thus, the oracle mechanism130, within the decentralized application, is usually assumed to be centralized.131 But, this fact essentially contradicts the decentralized ideals of a truly trustless and fair system, and therefore provides the main reason why I do not consider the Stox project to be the best overall representative of a decentralized prediction market.

Next up, is a project called Delphy132. Delphy is very similar to Gnosis and Augur, in that it employs a distributed oracle, is built on the Ethereum blockchain, and its proprietary DPY tokens, which are based on Ethereum smart contracts that comply with ERC20 standards, are used to trade contracts and incentivize participants to accurately communicate their wisdom. Though, Delphy’s whitepaper clearly states that after a market matures, there will be no exchange of any kinds of funds, neither fiat nor DPY tokens, instead users will be incentivized through a pre-fixed amount of DPY tokens minus transaction fees. Besides this distinction, there are two other main differentiation criteria which need to be addressed. Firstly, the Delphy application is an open-source light Ethereum node that exclusively runs on mobile devices. Secondly, and lastly, it was founded and launched in Singapore133, which infers a completely different political and regulatory environment than in the US or Europe.134

Another decentralized prediction market project, which is being developed and established in the Far East, is called Bodhi135. Alike all other existing decentralized applications, Bodhi’s network is Ethereum-based and the company’s proprietary digital tokens (BOT), which serve to enhance the reliability of the prediction market’s decision-making process, comply with ERC20 standards. Yet, the most significant difference, compared to its peers, is that Bodhi

130 An oracle is an external actor or entity that feeds information from the real world into the blockchain. 131 see. Stox Whitepaper, 2018, p. 34. 132 see. Delphy, 2018: https://delphy.org/. 133 see. Gao, 2018. 134 see. Delphy Whitepaper, 2018, p. 1. 135 see. Bodhi, 2018: https://www.bodhi.network/.

61 employs a “replaceable oracle” mechanism. This means that Bodhi uses a third-party oracle, which automatically judges forecast results, to guarantee an efficient decision-making process, but in case the oracle happens to fail, participants can take over the voting process for a particular outcome and through arbitration find consensus among themselves.136

The last decentralized prediction market project, which should be mentioned here, is BlitzPredict137. BlitzPredict is a global platform under development that applies fintech solutions for providing functionality and liquidity to users of blockchain prediction markets and sportsbooks. The main two unique selling propositions of its service are BlitzPredict’s aggregator, which ensures that its customers will always get the best odds available in the market at any given time, and BlitzPredict’s liquidity reserve, which assures immediate payment of funds at the end of an event.138

9.2. Augur

Here, I am going to briefly summarize what has already been disclosed about Augur139, before continuing with the history of the project in the next passage.

As previously revealed, Augur is based on the Ethereum blockchain and its proprietary tokens, which are used to report and dispute outcomes on the platform, comply with ERC20 standards and are called Reputation (REP) tokens. However, trading is conducted with one of the most widely used cryptocurrencies, the aforementioned Ethereum coins, which helps to make trading on Augur accessible to everyone. Furthermore, this setup allows a high compatibility and gives developers around the world a chance to build applications on top of Augur’s platform.

Moreover, it has already been uncovered that Augur is the sole truly decentralized prediction market project, since its oracle is legitimately decentralized. This implies that the decision- making process is always performed by its decentralized participants, hence never by a single central authority. Additionally, it can be said that such a unique and uncensorable structure, besides having several interesting effects on the human-machine interaction process, certainly introduces a gigantic universe of new opportunities for visionaries and developers alike.

136 see. Bodhi Whitepaper, 2018, p. 10. 137 see. BlitzPredict, 2018: https://www.blitzpredict.io/. 138 see. BlitzPredict Whitepaper, 2018, p. 3. 139 see. Augur, 2018: https://www.augur.net/.

62 In the upcoming chapters, I will discuss all this and more, while focusing on the consumer experience.

9.2.1. History Augur is headquartered in the San Francisco Bay Area. It was founded in 2014, by Joseph Charles “Joey” Krug and Jack “John” Peterson, with support from the Forecast Foundation, a not-for-profit group of software developers, dedicated to decentralized applications, which includes advisors such as Intrade founder Ron Bernstein alongside Ethereum founder Vitalik Buterin140. In 2015, between the 17th of August and October 1st, Augur had its product crowdfunding REP token ICO, where they managed to accumulate roughly $5.3 million via selling 8.8 million tokens, out of a total of 11 million, for $0.60 each141. According to crunchbase.com, there has been a second venture funding round on October 1st of 2017, but no specific amounts have been publicized. Figure 17 shows the price in USD, the market cap, and the daily volume of Augur’s Reputation token. Moreover, the prices of Bitcoin and Ethereum have been added for comparison.

Figure 17: Augur’s Reputation Token

Notes: Graph assembled from Coinmarketcap.com on December 31, 2018.

140 see. Bloomberg, 2018. 141 see. ICObench, 2018.

63 Here, we can clearly observe a strong uphill linear relationship between the Bitcoin, Ethereum, and the REP token. In fact, a 180-days correlation matrix from September 2018 states a positive correlation for REP token of 67% with BTC and 69% with ETH, plus a positive correlation between BTC and ETH of around 88%142. This is one significant indicator showing how incredibly depended the entire blockchain industry on the valuation of Bitcoin really is.

In the second week of July 2018, the project finally went live. At that point, the REP token traded for around $34, implying a market cap of about $377.7 million, which signifies a discount of roughly 68% from its peak price of $107 per token registered in mid-January of 2018.143 Shortly thereafter, Augur’s team announced, via Twitter, that they had successfully destroyed their access to the fail-safe kill switch, completing the full decentralization of the application.144

Not only did this act transfer any type of control over the network to all its participants, theoretically, it also should have erased any kind of legal accountability, regarding possible illicit activities on the platform, for the developers. This is important, because right after launch, people started to create unlawful assassination markets featuring famous individuals. Additionally, the Commodity Futures Trading Commission began to investigate whether Augur was selling binary options without registering them, which would violate US law. This investigation is still ongoing, with one CFTC spokeswoman leaning towards a compliance breach on Augur’s part.145 Nevertheless, Augur is up and running and for now it seems that nobody will ever be able to single-handedly shut it down.

Lastly, I will take a look at Augur’s daily users. Here, it needs to be distinguished between, REP token holders, who basically power the platform but must not necessarily be daily users, and real daily participants of prediction markets on the Augur platform (i.e., traders).

One day after its launch, on July 10th of 2018, Augur had 56,338 unique REP holder addresses, yet only 265 active traders, averaging a weekly trading volume of around 1000 ETHs (150,000 USD). As it turns out, this happened to be the provisional peak of user activity, since on August 8th the number of daily users already had dropped to 35, and the weekly trading volume

142 see. Hauge, 2018. 143 see. Medium, 2018b. 144 see. Peterson, 2018. 145 see. Bloomberg, 2018.

64 amounted to around 404 ETHs (approx. 60,000 USD). Moreover, during the last 5 months of 2018 these numbers barely changed, practically stabilizing at this aforementioned level.146

The latest figures, from January of 2019, reveal a total of 1708 different prediction markets with a combined $2.04 million that has been at stake so far. However, only 52 out of those 1708, barely 3%, are liquid markets, the overwhelming majority are “ghost markets”, where no trading activity takes place.147 For a project that possesses almost unlimited potential, and today is valued at around 98 million USD, those numbers appear to be shockingly small.

In the upcoming chapters, I will gradually expose why this revolutionary project receives so little to almost no attention from a potential global audience of nearly 3.3 billion people.

9.2.2. Service Description In reality, Augur is nothing more than a decentralized oracle and peer to peer protocol, specifically designed for prediction markets. A set of smart contracts, written in Solidity, which can be deployed on the Ethereum blockchain.

In other words, it is a free and open-source software (FOSS), portions of which are licensed under the General Public License (GPL), and portions of which are licensed under the Massachusetts Institute of Technology (MIT) license, enabling the decentralization of prediction markets. People can freely use the Augur protocol in whichever way they please, but they are doing so at their own risk, and they must themselves ensure that the actions they are performing comply with the laws in all applicable jurisdictions, as well as acknowledge that others’ use of the Augur protocol may not be compliant with these laws.148 Consequently, this is shifting the responsibility of the platform’s activity from the creators of Augur to the individuals using it.

Augur markets follow a four-stage progression: creation, trading, reporting, and settlement, with Augur’s Reputation token playing a central role in all these operations, except the trading part, which is exclusively conducted through Ethereum.149

146 see. DappRadar, 2019. 147 see. Predictions.Global, 2019. 148 see. Augur FAQs, 2018. 149 see. Augur Whitepaper, 2018, p. 2.

65 Basically, anyone with the required equipment (ETH and REP) can create a market based on any upcoming real-world affair. The market creator sets the event end time and chooses a designated reporter to report the outcome. Moreover, he or she elects a resolution source, which reporters should consult to determine the outcome. This could simply be “common knowledge”, or it may be a specific source, such as the New York Times, the World Bank, nasdaq.com, or the OPEC. Furthermore, the market creator appoints a creator fee, which is a fee paid to the creator by traders who settle with the market contract. To finalize the market creation process, the market creator has to post two bonds: the validity bond, and the designated report no-show bond.

The validity bond is paid in ETH and returned to the market creator if the market resolves to any outcome other than invalid. The size of the validity bond is set dynamically, based on the proportion of invalid outcomes in recent markets. Thus, the validity bond incentivizes market creators to create markets based on well-defined events with objective and unambiguous outcomes.

The no-show bond, paid in REP, is only returned to the market creator if the market's designated reporter successfully reports during the first 72 hours after the market's event end time. If the designated reporter does not submit a report during the fixed 3-day window, then the market creator forfeits the no-show bond and it is ascribed to the first public reporter who manages to correctly report on the market. This incentivizes the market creator to select a reliable designated reporter, which should help to resolve markets faster. Similar to the validity bond, the size of the no-show bond is adjusted dynamically based on the proportion of designated reporters who failed to report on schedule during the prior fee window. Lastly, the market creator establishes the market and posts all required bonds via a single Ethereum transaction. Once the transaction has been confirmed, the market goes live.

Immediately thereafter, the trading of shares is open and, all users are free to submit orders on any market they like. Augur’s on-contract matching engine matches all outstanding orders in the most effective way, creating as many complete sets of shares as needed, for the purpose of concluding the highest achievable number of transactions. A complete set of shares is a collection of shares that consists of one share of each possible valid outcome of an event.150

150 see. Clark et al., 2014.

66 For instance, let us consider a market that has two possible outcomes, A and B. Furthermore, the order book shows that Sarah is willing to pay 0.6 ETH for a share of A, and Paul wants to invest 0.4 ETH for a share of B. Then, Augur matches these orders and collects a total of 1 ETH from Sarah and Paul. Subsequently, the system automatically creates a complete set of shares, awarding Sarah the share of A and Paul the share of B. Essentially, this is how shares of outcomes come into existence on Augur. Once the shares are created, they can be traded freely across the entire application at any time. In fact, all Augur assets, including shares in market outcomes, participation tokens, shares in dispute bonds, and even, ownership of the markets themselves, are always transferable among participants.

Additionally, Augur’s trading contracts maintain an order book for every market created on the platform. As already established, practically anybody can submit a new order or fill an existing order at any time. Requests, to buy or sell shares, are fulfilled immediately if there is a matching order already on the order book, however, if there is no such matching order available, or the request can only be partially filled, the remainder gets placed on the order book as a new order. In any case, the order’s creator always has the option to remove unfilled or partially filled orders from the order book. Moreover, orders are never executed at a worse quote than the limit price set by the trader but may be completed at a better price. Lastly, it should be mentioned that traders are only obligated to pay fees when complete sets of shares are sold.151

After the event on which the market is based has occurred, the outcome of that event is determined by Augur's oracle. The Augur oracle allows information to be migrated from the real world to the blockchain without relying on a trusted intermediary, which is why many (including me) refer to Augur as the first and only trustless prediction market application available today. If I get into further detail, I come to realize that the oracle is simply the process of finding a valid consensus among profit-motivated REP token holders i.e., the reporting phase.152

Reporters can dispute or report on a market by staking their REP on one of the market's possible outcomes. By doing this, the reporter declares that the outcome on which the stake was placed matches the real-world outcome of the market's underlying event. For the purpose of

151 see. Augur Whitepaper, 2018, p. 2. 152 see. Augur Whitepaper, 2018, p. 3.

67 determining the market's outcome, the consensus of a market's reporters is considered the ultimate “truth”. If a reporter's statement of a market's outcome does not match the consensus reached by the other reporters, Augur redistributes the REP staked on the non-consensus outcome by this reporter to the contributors that reported with the consensus. By owning REP and participating in the accurate reporting on the outcomes of events, token holders are entitled to a portion of the fees on the platform. Each staked REP token entitles its holder to an equal portion of Augur's market fees, but the more REP a reporter owns and reports correctly with, the more fees he or she will earn for his or her effort in keeping the platform secure.

Principally, Augur’s markets can be in seven different states, or phases, after creation. The relationship between these phases is illustrated in Figure 18.

Figure 18: The Reporting Flowchart

Notes: Reproduced from page 4 of Augur’s whitepaper.

68 At this point, I am facing a minor dilemma. Since, on the one hand, it is scientifically important to outline the precise inner workings of Augur’s reporting phase as well as the security and integrity of Augur’s forking protocol for the purpose of transparency and peer-reviewed consistency testing, but, on the other hand, this thesis focuses on the perspective of an average consumer who is interested in prediction markets for financial assets, and the only thing that really counts for such a user is that the service works dependably. However, this stable functionality has been proven by the fact that Augur’s decentralized oracle has been live and reliably running for the past 5 months, which is the result of multiple years of intense testing, bug-hunting, and improving, aimed at ultimately developing the best practical solution possible. Therefore, I have decided to skip the precise discussion of this part and continue to conclude this section by presenting the last lifetime stage of a market on Augur, while inviting readers who are eager to find out more to consult Augur’s whitepaper, where further technical details are accessible.

The reporting process, outlined in Figure 18, ends with a market entering the finalized stage. Finalization occurs, if the market passes through a 7-day dispute round without having its tentative outcome successfully challenged, or, after completion of a fork. Forking is the market resolution method of last resort. It is a very disruptive process, intended to be a rare occurrence, since it lasts up to 60 days and puts all other non-finalized markets on hold until resolved. Furthermore, the outcome of a fork cannot be disputed and is always considered final at the end of the forking period.153

Once a market is finalized, traders can settle their positions directly with the market, or sell their shares to another trader in exchange for funds. If kept in mind that every share comes into existence as part of a complete set when a total of 1 ETH has been escrowed with Augur, then settling with the market contract refers to the exchange that takes place when traders give Augur either a complete set, or, if the market has finalized, a share of the winning outcome, to transfer that 1 ETH out of Augur’s escrow into their accounts.

If a market resolves as invalid, traders who settle with the market contract receive an equal amount of ETH for shares of each outcome minus fees. Augur levies two types of fees, which are proportional to the amount being paid out, during settlement. Alongside the creator fee,

153 see. Augur Whitepaper, 2018, pp. 5-6.

69 which I have already discussed, the reporting fee, which is set dynamically and paid to reporters who participate in the reporting process, should be mentioned.

Due to technical limitations, trades cannot simply be unwound if a market resolves as invalid, implying that if a trader happens to own shares of an invalid market, he or she is assured to lose money and time. This fact appears to be another risk contributing aspect which may explain Augur’s current humble popularity.

9.2.3. Potential Issues and Risks Before I take a closer look at Augur’s user interface by simulating a trading process, it is necessary to address several risks and potential issues associated with this project, which, if ignored, could lead to devastating failures along the way. This section attempts to summarize all these latent threats.154

The first is the threat of parasitic markets. Such markets do not pay reporting fees to Augur but resolve in accordance with the resolution of a native Augur market. This means that parasitic markets do not have any reporters to pay and consequently are able to offer the same service as Augur, but with lower fees. Now, if parasitic markets attract trading interest away from Augur, then Augur's reporters will receive less in reporting fees, which in turn places downward pressure on the market cap of REP. Though, this market cap of REP token is directly correlated with the integrity of the forking protocol and if it descends too low, the long-term viability of Augur is put into serious jeopardy. Thus, the best protection against parasitic markets is to make trading on the Augur platform as inexpensive as possible (while still maintaining the integrity of the oracle), which diminishes the reward for running a parasitic market.

Next up, are large, sudden, and unexpected increases in open interest, like those that may occur during a popular sporting event. Such increases provoke a rapid surge in the market cap requirement of REP token to assure forking protocol integrity, and if the market cap requirement exceeds the market cap, there is a risk of economically rational attackers causing a fork to resolve incorrectly. To address this issue, Augur does attempt to adapt the market cap upwards throughout such situations, but these nudges are reactionary and are only adjusted

154 see. Augur Whitepaper, 2018, pp. 10-11.

70 once per 7-day fee window. Therefore, the best chance to avoid such problems are traders who witness the sudden increase in open interest and buy REP token in anticipation of the reactionary market cap nudge. This elevates the market cap of REP, perhaps to a point where the integrity of the forking protocol is no longer threatened, or, at the very least, it reduces the length of time during which the oracle may be vulnerable.

The third potential problem arises when a market creator chooses an inconsistent or malicious resolution source. Since there may ensue several scenarios in which honest reporters lose money, or, worse, such a market could initiate an undesirable and time-consuming fork. For this reason, reporters should always remain vigilant against markets with dubious resolution sources. The best defense, concerning this issue, is to publicly identify such markets, so all reporters can coordinate to make sure these markets finalize as invalid.

The fourth inconvenience, that may occur, are self-referential oracle queries. To explain this potential problem, I am going to consider a market which trades on the question, “Will any designated reporter fail to submit a report during their three-day forking period before December 31, 2018?” Bets placed on the No outcome of this market may act as a perverse incentive for designated reporters to intentionally fail to report, because if a designated reporter can buy up enough Yes shares at a low enough price to compensate for a loss of the no-show bond, then it makes economic sense for him or her to not submit a report.

These self-referential oracle queries will not threaten the integrity of the forking protocol if the market cap of REP token is large enough, yet they may negatively affect the performance of Augur via causing significant delays in the finalization of markets.

Another risk, that could potentially harm Augur’s platform, is the uncertainty of fork participation. Augur forks diverge from blockchain forks in one important aspect: after a blockchain fork, a user who owned a coin on the parent chain will now own a coin on both forks. As a result, ignoring replay attacks, blockchain forks pose little risk to users. Though, after an Augur fork, a user who owns a REP token in the parent universe can migrate that coin to only one of the child universes. If the user migrates the token to any universe other than the consensus universe, this token may lose all of its value. Hence, migrating REP during the forking period of a fork, before it is clear which child universe has achieved consensus, exposes a user to risk, which may discourage participation during the forking period of contentious

71 forks. To compensate for this risk and encourage participation during forking periods, Augur rewards all token holders who migrate their REP within 60 days of the start of a fork with an additional 5% of REP in the child universe to which they migrated.155

The last potential issue, which should be mentioned here, emerges when dealing with ambiguous or subjective markets. In general, solely events that have objectively knowable outcomes are suitable for use in Augur markets, however, it is possible to envision markets where some reporters are certain that the correct outcome is A, and others are certain that it is B. In the worst case, such markets may result in an undesirable fork where REP in more than one child universe maintains a non-zero market value. Therefore, reporters in the Augur network should always report a market as invalid if they believe that a market is not suitable for resolution by the platform. When a market resolves as invalid, traders are paid out at equal values for all possible outcomes, or, for scalar markets, they are compensated with a price that quotes exactly halfway between the market's minimum- and maximum price.

9.2.4. User Interface To retain objectivity, I am going to analyze Augur’s user interface in the same way as I did with Vetr. As a remainder, in this scenario I am interested in information about the future price of Apple Inc.’s stock (AAPL), and I plan on making a prediction regarding its target price on the application.

Augur’s service can be reached either through a web-based application or by downloading the Augur software to any device available (Windows, MacOS or Linux). Yet, in both cases, an Ethereum-enabled wallet, which can store cryptocurrencies and connects Augur to the Ethereum blockchain, is needed to fully engage in the prediction markets application. The Augur platform currently supports four different wallets, namely Metamask, Trezor, Ledger, and Edge. But to start, I am going to ignore the prerequisite of possessing a digital wallet and explore how much information about my target company can be gathered without it.

By clicking the “Get Started” button on www.augur.net and continuing to select the web-based version, I get redirected to a screen that resembles Figure 19156. The layout seems very clean

155 Augur cannot know in advance whether this 5% bonus will be enough to compensate for the risk and incentivize participation during a forking period. 156 This is a temporary website therefore a link cannot be provided.

72 and effectively sparse. On the left side, I can see, a “Markets” and a “Reporting” option window, and on the right, a search field.

Figure 19: Augur’s User Interface (Web Version)

Notes: Retrieved from Augur.net on January 8, 2019.

If I now type “Apple” in the search bar and press “ENTER”, I am able to obtain all currently open Apple-related prediction markets available on this platform.

The result, illustrated in Figure 20, shows a total of four different markets of which only two have recorded any trading activity, and none that features information about the future stock price of Apple. In fact, if I look closer, I come to realize that on this entire application there are only 360 open markets, 404 markets that are in reporting, and 974 markets which have been resolved so far. If I add up these numbers and compare them to the 1708 prediction markets from January 3rd of 2019, mentioned above, it can be deduced that 30 new markets have been created during the past 5 days. However, if we keep in mind that this includes markets from all imaginable areas, not just financial assets, and consider the troublingly low overall activity rate of around 3%, then this growth rapidly disappears into plain insignificance. Furthermore, the fact that there are more markets in a state of reporting than open markets, seems irritating to me and may point to a slow resolution process of markets within the Augur network.

73 Figure 20: Apple Inc. related Markets on Augur

Notes: Retrieved from Augur.net on January 8, 2019.

Additionally, excluding cryptocurrencies and tokens, I found exactly 14 open markets which are somehow related to financial assets. But, only two of those record any trading activity at all, and solely one question, “How will the S&P 500 change in the first 5 months of 2019?”, relates to traditional stocks.

Given all this information, it becomes clear that the most popular markets on Augur are cryptocurrency related. A total of 58 different markets can be found on the platform, with 13 of those showing active trading volumes. Among them, represented in Figure 21, the market presently exhibiting the highest trading activity on Augur, which deals with the market cap of ETH and Bitcoin.

Here, I can observe further details regarding the contracts traded in this market, such as the resolution source, the volume, estimated fees, the market depth, the order book, and the current best bid- and ask prices as well as quantities. The best asking price, and quote of the last transaction, therefore the current consensus price, is 0.2495 ETH, for a quantity of 0.16 ETH (roughly 24 USD). This implies that there is one participant who believes that there is a higher than 24.95% chance that Ethereum’s market cap will exceed the market cap of Bitcoin by the end of 2019. However, the best bid for this Yes contract is 0.0621 ETH, inferring an opinion

74 spread of 18.74% and a more rational mid-price of 0.1558 ETH, or, a current probability of 15.58% that Ethereum’s market cap will be higher than Bitcoin’s. Moreover, an open interest level of 73.5650 ETH, compared to the already traded 70.4278 ETH, indicates sufficient interest as well as a diverse distribution of opinions in this market. Nevertheless, it must be concluded, that besides a handful of markets, the overall trading volumes are far too low to assume any statistically significant information from Augur’s markets, yet. In addition, it should be noted that no new or valuable information about Apple’s stock price could have been gathered.

Figure 21: The Most Frequented Prediction Market featuring Financial Assets on Augur

Notes: Retrieved from Augur.net on January 8, 2019.

As it turns out, I was able to gather information about everything that is and has happened on Augur without utilizing a wallet. Though, if I wanted information about Apple’s stock price, I would have to engage with the platform, and thus open a wallet beforehand.

The process of creating a wallet or buying and installing a “relatively secure” hardware wallet like Trezor, is, summed up in one word, “overwhelming”. Despite what experts in step by step guides postulate, if you are an average tech-savvy investor who has never been exposed to the blockchain industry before, it takes a lot longer than 10 to 15 minutes157 to set up a wallet, and

157 see. Hongkiat, 2018.

75 quite frankly, I wouldn’t know what to do with it after completion. This entails at least another 3-4 days of research on how to open an account and transfer funds, while at the same time totally ignoring the ever-looming multiple threats of losing up to 100% of invested capital when dealing with digital assets. Consequently, this procedure represents a major entry barrier for potential users and may be a substantial reason why the participation rate of Augur is currently that low.

In conclusion, if I really wanted to find out more about the future price target of Apple’s stock, I would have to create a wallet, then connect the wallet to Augur, before buying REP and ETH to create a market. After that, one can only hope that enough people are interested and show up to actually trade on the market.

As a consequence, I have decided to not investigate any further, since the costs of setting up an account clearly outweigh the risk-adjusted possible benefits of continuing this examination. In fact, at the current stage, I believe that dealing with Augur for the first time can be quite a confusing endeavor, which involves a high degree of time- and capital consuming learning processes, accompanied by an exposure to stressful risks, for potentially no or very little reward.

76 10. Utility Value Analysis

In chapter 6, it has already been outlined who the subjects of this examination are and how this analysis is going to be conducted, hence I am almost prepared to start immediately. However, before I begin this comparison, a few basic assumptions, which had been laid out earlier, need to be discussed.

First of all, the premise of 100% functionality of both services, which had not been satisfied at the start of writing this thesis, is now a reality, and therefore no further adaptations to objectify this comparison are necessary.

Secondly, it needs to be clarified that this UVA is conducted from the perspective of a privacy- aware, risk neutral, tech-savvy investor, who is interested in exploring prediction markets as an alternative research technique for financial assets but has never been exposed to the blockchain technology before.

Lastly, I would like to elaborate on how the criteria catalog for this comparison has been established.

Essentially, the list of criteria for this investigation is based on three simple questions. Firstly, “What do I need to participate?”- where I identify potential prerequisites and barriers to entry. Followed by, “How is the overall usability and use experience?”- which contains criteria such as the user interface, the system’s security and flexibility, the reward program, eventual costs, resolution effort and -speed, privacy concerns, and the support arrangements. Completing the trio, is the question, “How good is the service offer, especially regarding financial assets?”, where I try to assess how accurate and viable the freshly gathered insights are.

The result is a list of 17 aspects which I deemed essential for properly evaluating my two target prediction market services.

10.1. Augur versus Vetr

Table 5 illustrates the final version of the complete utility value analysis. Augur scored 38.13%, and Vetr reached 84.39%, of the maximum UV-Score. Although these results seem to display

77 a clear favorite, it must be reiterated that the grading and weighting of criteria is a subjective task, and thus could substantially deviate from person to person or perspective to perspective.

Table 5: A Utility Value Comparison of Augur and Vetr

Notes: Created by author on January 19, 2019.

Before continuing to interpret these results in the next section, I am going to explain the reasoning behind my ratings and weights in this part.

The first two criteria are technological- and financial knowledge requirements, where Vetr managed high scores in both, whereas Augur scored poorly. That is because Vetr is a simple rating application, designed to accommodate a mainstream audience, where an average education level of technology and financials will easily suffice to participate. Augur, on the other hand, demands a vast knowledge base, not only of the technology sector but also from a financial perspective, to even be able to start participating. Presumably, the average user, besides understanding how certain contracts work, will have difficulties figuring out how to handle cryptos and wallets.

In terms of equipment requirements, the low rating for Augur’s service grounds in the fact that without a wallet to store digital assets, participation is simply impossible. By contrast, Vetr’s service only needs to be coupled with either Facebook, StockTwits, an email-address, Twitter, or LinkedIn to engage with its forecasting services, thus justifying its rating of 78.

78 Moreover, Vetr is a legitimate business, therefore its legal framework is clear. Customers can utilize Vetr’s prediction markets for free, but they indirectly pay with the centrally gathered meta data about them and their behavior on the platform. Consequently, this data is sold to business clients for a monthly fee. Conversely, Augur’s legal environment, at best, can be described as foggy. Although virtual currencies are regulated, decentralized prediction markets are not. However, they are under constant observation by the CFTC, and are currently being investigated for offering binary options without a permit. Furthermore, the possibility for users to create various kinds of illicit markets, including assassination markets, and the general lack of a responsible authority, negatively influence Augur’s score within this criterion.

Transparency is important, because it allows a potential user to determine in advance how a particular service works and if those terms are acceptable to him or her. The blockchain technology allows Augur to attain the highest possible score in this category. Simply because, it reliably traces and stores every activity within the application. Additionally, the Forecast Foundation and Augur’s whitepaper flawlessly outline the inner workings of the project, all publicly accessible online. In Vetr’s case, the inner workings appear to be clear at first glance. However, we know that meta data about user activity on the platform is a source of income for Vetr, and nobody can convince me that this data, coupled with account information from Facebook, is not a perfect starting point for some form of virtual crawler to search the entire web for additional matching and perfectly identifiable activity data, which then gets added to its database, almost instantaneously ready to be exploited. In my humble opinion, this is absolutely a possibility, but not mentioned in Vetr’s service description, which is why I elected to give them a 44 in this section.

Furthermore, both platforms attained high ratings in regard to the user interface. Since, in Augur’s case, I like the effectively sparse and futuristic layout which makes it easy to comprehend and control, and Vetr manages to display a lot of information embedded in an effortlessly maneuverable layout surface. The only minor point of criticism, I could claim, is that several markets on Augur are difficult to figure out, or quite simply nonsense. Though, this is hardly the fault of Augur’s user interface.

The next criterion, which needs to be elaborated, is the flexibility of these applications. In general, flexibility in the context of software development may refer to multiple characteristics. Here, it encompasses attributes such as error tolerance, installability, portability and

79 upgradeability, but mainly it represents the possibility for the user to customize and/or personalize the service offer within the application beyond basic features.

On Vetr, this offer seems to be very narrow and focused on the main purpose of forecasting financial assets without any real money involved, which is the reason why it only achieved a low score. On Augur, on the other hand, participants can, besides the basic trading for real money (ETH), create as well as report on markets, and generally engage in the process of maintaining the platform to earn valuable digital assets, which practically provides a maximum amount of freedom to the customer. Therefore, Augur receives an almost perfect score of 98 in this category.

With regard to the current state of my targets’ system security, the riddle I had to solve was how to measure security, as I believe that at any given point in time, a network can either be secure or not. However, risks of corrupting a system will always be an issue, and those are quantifiable, hence the system showing the least amount of potential risk factors can be considered as relatively more secure. Even though there are several potential issues and risks which may arise in a system featuring a central authority like Vetr’s, they tend to be relatively minor compared to those that may occur within Augur’s highly innovative decentralized application. Nevertheless, currently both platforms can be considered as secure, which is the reason why both acquired a high rating in this part.

Next, privacy concerns have to be discussed. As mentioned above, nowadays the right to privacy for customers solely relies on trust in the institutions operating the various platforms. Unfortunately, due to the evolution of the internet, where free of charge does not necessarily mean free of charge, but rather implies that users pay with their activity data, the right to privacy has slowly but surely been eradicated throughout almost the entire open internet. However, the dark- and deep web, as well as cryptocurrencies allow users to remain at least pseudonymous, maybe even anonymous. As a result, Augur gets an almost perfect rating in this section, and Vetr only attains a weak 17% of the maximum score.

I am now moving on to the incentives structure, which I elected to weight at only 2%, because, for obtaining insights about the future price of financial assets, a reward system is not a necessity. However, some kind of incentives are needed in order to motivate users to predict truthfully, hence it remains a vital part of these systems. Augur’s participants can reward

80 themselves through successful trading and can get rewarded by creating as well as reporting on markets, thus a high UV-score of 90 is justified.

Vetr incentivizes users by endorsing their social status within the network. If they happen to be exceptionally accurate forecasters, users may receive promotional gifts or exclusive rights to additional features on the platform, but at no point during this interaction any real money changes hands. Therefore, I have decided to give Vetr a 68 for this criterion.

Furthermore, there are no monetary costs for the general user on Vetr. Though, business customers, such as hedge funds and peers, may obtain Vetr’s datasets for a monthly fee. In contrast, the fee structure on Augur is very complex and dependent on many factors, including fees to acquire ETH, and the daily volatility of ETH, as well as the Ethereum gas price158. According to one of the founders, fees for one trade vary from approximately 5.5% in the best- case scenario, up to around 14% in the worst case.159 Compared to current competitors such as Betfair, who charges about 7.5%, these numbers seem uncomfortably high and signify another main reason why participation on Augur remains low. But all these issues were expected to occur during the early months of Augur’s existence, and they can be improved if given time, therefore Augur receives a generous 44 in this category, while Vetr reaches a solid 58.

When addressing resolution effort and -time, it can be stated that Vetr’s application is almost perfect for an average investor. Vetr’s prediction markets are tailored for financial assets and never end, hence users can effortlessly find desired information in less than 10 seconds. On the contrary, Augur’s decentralized markets must expire at some point, which only starts their resolution process of reporting and validating an outcome, to eventually come to a stage of finalization. Although traders can sell their shares to anybody at any time, a settlement with the market is only possible if a market is finalized, which takes a minimum of 10 days.

It is becoming evident that this is Augur’s primary weak spot. Countless things could go wrong, if a couple of people intentionally or unintentionally act irrationally, or try to harm the system on purpose, and each little issue contributes to ultimately slowing down the entire network. The good news is that to every problem there is a solution, and that the system continues to adapt as well as improve on a daily basis. But for now, the resolution effort and -time on Augur

158 see. Frankenfield, 2018. 159 see. Krug, 2018.

81 can be described as almost unbearable to participants and may involve losing money through no fault of their own, which is exceptionally painful. Therefore, the ratings in these two categories are almost perfect scores for Vetr, and Augur attains a weak 26 and 17 respectively.

Since a significant portion of Vetr’s platform relies on the social connections and the interaction among participants, I cannot imagine that anybody would really need a support to utilize Vetr’s application, though customers are able to contact the company via email at any time they please. Conversely, Augur’s support community is concentrated on a handful of people hidden in specialized communities like GitHub, Medium or Reddit, where it is often not an easy task to gather meaningful and viable information.

Moreover, at first glance the sphere of Augur with 56,384 individual REP holder addresses160 seems higher than Vetr’s estimated customer base of around 22,000. However, nobody knows how many of Augur’s individual addresses belong to only one institution or person, thus making it de facto impossible for me to identify an exact amount. The present reality is that, with around 30 daily active users, Augur’s participation rate, which directly relates to my definition of reach, is nowhere near sufficient levels capable of producing reliable prediction market results, whereas Vetr’s daily activity is reasonably high and provides dynamic as well as reliable information in a decent portion of its prediction markets, totaling around 115,000 individually crowdsourced ratings161 during its existence.

Additionally, when addressing the financial product offering, it is already known that Vetr’s service has been designed with a focus on publicly listed stocks, exclusively providing prediction markets for financial assets. However, Augur’s markets cover all imaginable areas featuring uncertain future events and only about a third of markets refer to cryptocurrencies, which can be considered financial assets. If we now recall that a great portion of those markets is practically without any trading activity, the score of only 11% becomes apparent.

This brings me directly to the overall quality of information, the last criterion which has been chosen for this comparison. As the current reliability of discoverable information obviously accounts for a major part of this analysis, the final section is weighted at 18%.

160 see. Etherscan, 2019. 161 There is a ratings counter on https://www.vetr.com/.

82 On Vetr, the user base clearly reached a critical mass, where network effects are able to positively influence the quality of data. For popular stocks, the accuracy can reach up to 99.7%, proving, once again, that prediction markets are able to dynamically produce tremendously accurate results, if set up correctly. On the other end of the scale, lies Augur’s current information validity. Throughout this comparison, it became apparent that besides a few minor bugs the platform works, but, due to a lack of participants, the results of nearly all prediction markets cannot be regarded as reliable, and therefore Augur must receive a very low rating in this last category.

Finally, it should be briefly added that the weights have been distributed, almost evenly among all listed criteria, with a distinct focus on the user interface and the quality of the product, especially regarding financial assets.

10.2. Interpretation of Results

The UVA was designed so that the maximum score would resemble a perfect prediction market application embedded in a perfect environment. This setup allows me to easily interpret its results in a digestible manner, in that, Vetr acquired around 84% and Augur reached around 38% of a perfect prediction market application for an average privacy-aware investor.

But, to be completely honest, this has never been a fair “fight”. Vetr is a legitimate business with a sound execution model, featuring a sufficiently large user base, solely specializing in financial assets. Moreover, the centralized nature of its service allows Vetr to fully navigate all network activity and quickly intervene if necessary. Therefore, Vetr is able to offer a rapid, easy-to-use, and reliable prediction market service exclusively focused on financial assets.

The only downside to using its platform is that participants must be aware any data they produce becomes Vetr’s property and probably will be commercialized in various ways, though how remains elusive. The most recent example of what could happen is the Facebook- Cambridge Analytica data scandal which unfolded in early 2018.162 Nonetheless, anybody who is slightly familiar with the internet must have realized by now that this practically applies to any service on the open web.

162 see. Cadwalladr/Graham-Harrison, 2018.

83 On the other hand, Augur’s main goal was never to generate a profit per se. Primarily, Augur has been a research project, featuring the revolutionary goal of combining the blockchain technology with prediction markets, and in doing so, decentralizing the traditionally centralized oracle, or process of reaching agreement.

It took several years to develop a functioning system, ready to be released to the public, indicating a responsibility swap of the entire application from the founders and developers to the people that use it. Now, and that was the general idea, it is the users’ responsibility to properly maintain the platform, mainly by creating popular markets, which attract new participants in order to raise liquidity and decrease costs, but also by not trying to actively harm the system and rather act rationally to accelerate the overall interaction process.

So far, it can be concluded that the community is doing a subpar job. The good news, though, is that combining prediction markets with the blockchain technology works, and the fundament to building better as well as faster decentralized applications on top of Augur has irreversibly been laid out. The bad news, however, is that the system is far away from reaching the critical mass of participants that it desperately needs to generate viable prediction market results.

Then, there is always the possibility of building a decentralized oracle specialized in financial assets on top of Augur. However, in that case we need to start thinking about regulation, besides scalability and the computational limitations of the Ethereum blockchain, as well as issues which may solely surface if Augur has been stress-tested to its limits. Nevertheless, the current situation does not necessarily infer a positive outlook on this happening any time soon, as the activity on the platform continues to stagnate at alarmingly low levels during recent months, and unresolved legal issues keep persisting in the background.

84 11. The Future of Prediction Markets

So, what about the future of prediction markets can be deducted from this thesis, and what else lies ahead for prediction markets as a forecasting tool beyond the currently ongoing blockchain integration?

At the very latest, since 2005, when chess amateurs Steven Cramton and Zackary Stephen, whose world rankings stood at around 1,400 to 1,700, won the freestyle chess tournament beating Hydra, the most powerful chess computer at that time, by utilizing regular Dell and Hewlett-Packard computers and software which they acquired for sixty dollars, the power of human-computer collaboration had undeniably been revealed, and businesses began to take notice.163 Consequently, we have seen AI and prediction markets converge during the past ten years.

Quantitative hedge funds were among the first financial adopters of prediction markets and are still experimenting with new ways to attain purer models of machine learning, thereby inevitably pushing the combination of both technologies forward. Examples include Aidyia’s AI hedge fund, in which all trades are executed entirely via machine. Their system identifies and executes its own trading strategies entirely autonomously using various forms of AI which analyzes basically everything, from prices to macro-economic data, social media posts and corporate accounting documents.164 Another example is Sentient, a distributed artificial intelligence platform that has initially been used for trading but is now being extended to other areas such as e-commerce and healthcare.165 Lastly, Numerai is an example of a start-up that is combining crowdsourcing and AI. Numerai is a hedge fund which harnesses and aggregates promising machine learning models from the masses, in the form of a global artificial intelligence tournament, to predict the stock market.166

Additionally, as discussed in chapter 4, AI is being used to enable and enhance human decision making. Part of this new human-machine approach is to employ AI models that can

163 see. Thompson, 2013. 164 see. Metz, 2016. 165 see. Adjodat, 2016. 166 see. Craib, 2016.

85 automatically implement adaptive incentive structures to correct for negative human behaviors therewith augmenting the operating abilities of participants.

Along with continuing to evolve better machine learning models, the push for greater amounts of unique data is a big focus of innovators. Data has become an increasingly important competitive advantage, thus there continues to be an ever-greater demand for faster and novel data. In response to this kind of client demand, incumbent professional information providers, such as Thomson Reuters, started experimenting with new business models and data collection techniques.167

If we now visualize to combine the corresponding integration of this new-found data with machine learning, prediction markets, and other modeling technologies, then the future potential of application areas appears to be mindboggling.

This is where the Internet of Things (IoT), the third innovative technology that theoretically couples nicely with prediction markets, comes into play. Sensors, cameras, essentially any electronic device with a Wi-Fi connection and a basic operating software can be used to automatically gather information. Imagine sensors, located wherever needed, which could either predict themselves or sell their information autonomously.

Finally, the fact that Augur’s trustless and censorship-resistant application is open source, means developers can build bots which predict and execute trades based on any model desired. Programs can absorb much more and make much better predictions about the future than slow humans. Therefore, potentially any program could start predicting, much faster and more effective than any person, ultimately providing the liquidity that Augur desperately needs to be useful. Here, the difference compared to a traditional market maker is subtle, nevertheless, the primary purpose of those bots is not to facilitate trading but rather to automate predictions.

This setup allows me to take a brief glance into the possible future of prediction markets.

The first scenario, which comes to mind, involves the aforementioned information markets. Information markets could develop around trading information to help predict and report outcomes. A sensor could produce information and log it into the public blockchain. If used for reporting, it may charge for that service, however, it could also result in markets where the

167 see. Shrier et al., 2016, p. 14.

86 information is not made public, but encrypted, upon which it may then be sold to others. It is not farfetched to imagine that this won’t just exist for sensory data, but it might eventually become the norm to report any kind of data into the blockchain. This does not only imply that one can have audited and transparent information in the blockchain, but it also means one immediately provides a trust source for information to be bet on at very little cost to the producer. For example, visualize a business that decides to log its monthly revenue to the blockchain.168

The second thing, that may appear, are new forms of organizations. With the rise of social media, which allows near instant communication and important news to immediately be filtered (based on retweets), we already saw new forms of headless movements like the Arab Spring, or more recently, a movement in South Africa that rallied around the simple hashtag “#feesmustfall” to reduce tertiary tuition fees. These organizations are unlike anything we have encountered ever before. There is no permission to be a part of it, and affiliation is as effortless as using a hashtag. In fact, it can be described as the network’s version of an organization. Prediction markets could help these new establishments to incentivize people to make desirable decisions. As a group goes about its goals various outcomes are constantly generated upon which the people in the organization and those outside of it wager on, leading it inevitably towards outcomes which serve the goals of the organization.169

Since these networked organizations move at the speed of social media, people may start to employ bots to bet on their behalf. This brings me directly to the third potential application area, which is the sharing of wealth. If prediction markets will be so useful in creating wealth, then people may develop automated personalized prediction bots to minimize effort and maximize reward. These bots automatically bet on events based on who you are and what you do. For example, if you join a hashtag organization, then the bot automatically detects this and assumes it as being a marginal indicator that this movement might succeed, thus bets accordingly on those outcomes, which should result in an automatic financial gain.

This scenario offers a potentially whole new perspective on organizations. Perhaps in the future, we won’t even need things like shares anymore and funds might simply flow towards

168 see. Medium, 2015. 169 Ibid.

87 some locus upon which they are utilized to improve a metric that can be predicted upon. Essentially, what this envisions is a new model of “investment”. For example, if you donate $5,000 to an organization or a group, you know that its metric of success will ameliorate, and therefore you can benefit from that financial gain. There are numerous parameters which could be influenced in this manner, simply by being alive one affects them.

Eventually, this opportunity will become available to anyone, at any scale, and it doesn’t even have to be of financial nature. For instance, imagine futarchies, which are governance systems where elected officials define measures of national wellbeing and prediction markets are used to determine which policies will have the most positive effect.

Or, what about preserving the habitability of our planet? Global warming is certainly a very real and blatant threat to the continued existence of the entire human race. Hence, another likely future application scenario could be to protect the environment. Picture a prediction market which trades based on a metric such as growth of new trees in an area. If you are interested in seeing the ecosystem flourish, you can get financially rewarded by simply protecting and fostering it. First, you predict and bet on it, then you enact the change (i.e., you make sure that a certain number of trees are planted). If we now envision measuring pollution, sensors can provide much more nuanced feedback and reporting than the rather concrete outcome of trees planted. To set up these sensors, they could be crowdfunded by a group of environmentalists, who in turn earn fees via selling the gathered information to interested parties, which subsequently would result in further creating a sustainable ecosystem. The end scenario would be a decentralized network of automated and interconnected sensors continuously collecting information for predictions.

I am going to conclude this segment, which clearly must be treated with some reservation, by summarizing that a decentralized prediction market can be a powerful tool, and once we incorporate the capability to automate predictions combined with AI, Machine Learning, and the Internet of Things, it becomes something that possesses the ability to change just about everything. In essence, it will result in being capable of modeling externalities in a much better fashion.170

170 see. Medium, 2015.

88 12. Conclusion

The main goal of this thesis was to compare traditional centralized prediction markets with the highly innovative and not yet completed idea of a decentralized prediction market platform, which is owned and maintained by the people that use it. It rapidly became apparent that properly evaluating projects on the cutting edge of technology cannot be conducted using traditional valuation methods. Therefore, I had to adapt my approach to accommodate the fact that a functioning blockchain-based prediction market practically did not exist. Ultimately, I elected a qualitative analysis, coupled with a literature research, through which I was able to shed some light on the current state of the prediction market environment.

The results were plenty and culminated in the utility value analysis scores of Vetr and Augur, which can directly be translated as percentage of completion to a, obviously unachievable, perfect prediction market platform for an ordinary investor. Although these results were attained in a subjective manner, and I am certain that other perspectives would have produced different results, I believe an emerging overall tendency is visible.

This leads me directly to the things I have learned throughout writing this paper and the personal conclusions I was able to draw from them. As someone with no prior exposure to the blockchain technology nor a traditional prediction market, therefore undoubtedly representing the majority of people, I was able to gather a vast amount of new information, among which the most significant will be discussed in this final section.

The first observation is that every prediction market mainly has two parties which need to interact so that the platform can deliver reliable results. On one side, there is the technological component, and on the other side, we can find humans. Over the course of this work, it became clear that both components need to, and can, be improved. The chapter above, discussing the future of prediction markets, shows how fast the technological component is improving over time and that it even can be used to guide the human learning process. However, the most interesting realization, here, is the perceived velocity gap between the technological innovation and the evolution of the human adaptation process.

89 Let me explain. I know, that according to Kondratieff and Schumpeter, technological innovation occurs in waves, or cycles, and that these cycles are becoming shorter, implying that the speed of technological innovation is accelerating171. I am also aware of the human adaptation process and that it takes a certain amount of time for the mainstream to reach acceptance. The 5 stages, represented in Everett Rogers’ diffusion of innovations theory172, illustrate that an innovation must be widely adopted to self-sustain and that within the rate of adoption there is a point at which an innovation reaches critical mass. However, the time of reaching that critical level is different from case to case, and therefore has not been further defined. Now, I claim that the alacrity of this human adaptation is observably becoming slower relative to the velocity of technological innovation over time.

In the same theory, Geoffrey A. Moore introduced a visible chasm between innovators and early adopters on one side, and pragmatists, who represent the early majority of people, on the other side173. Without having read the book, I can confidently disclose that I was able to observe this occurrence within the blockchain industry during my research.

There is a massive divide between the overly confident and positive narrative of developers, innovators, and a handful of early adopters, and the average majority of individuals, who either have never used a blockchain-based service before or are currently experiencing the fairly inscrutable and sometimes frustrating consumer reality of today’s cryptocurrency ecosystem. The reality is that, besides a vast amount of illicit activities, Augur is one of an elite group of blockchain-based applications which may actually serve a legitimate purpose to potentially everyone.

Maybe, this anecdote can illustrate what I am trying to express. In 2017, during the cryptocurrency hype, as you can imagine, several German television stations reported on Bitcoin and others. Over the course of this year, I happened to watch exactly four reports about cryptocurrencies. Three out of those four featured the same store, where they interviewed individuals purchasing their goods with Bitcoin. As it turned out, this was not a coincidence,

171 see. Schumpeter, 1939. 172 see. Rogers, 2003. 173 see. Moore, 2002.

90 they had to film this exact site because it was the only physical location in Germany where it was possible to pay in person with a virtual currency.

To elaborate further on this point, it is necessary to mentally go back to section 8.4. of this paper, where I stated that Bitcoin possesses all the relevant characteristics of a currency. This was purposefully taken from someone of whom I believe to be decently knowledgeable regarding cryptos in a technical sense, but the important part here is that his overly positive assumption appears to be based on a desire to accumulate wealth (i.e., to promote cryptocurrencies), rather than providing an objective and realistic view on the cryptocurrency ecosystem. Since, in his definition he deliberately left out one very important component that practically changes, in my opinion, everything. Not once, did he mention the stability of a currency. But, the stability (i.e., low volatility) of a currency is so vital to the whole concept of fiat currencies that almost every nation in the western world has numerous specialized institutions and policies in store to prevent high volatility scenarios from occurring, and even guarantees the value of a currency up to a certain amount.

In an attempt to prove that I am not the only one who has reservations concerning the entire cryptocurrency industry and that there is a significant gap among perspectives, I am going to provide a fairly recent statement of South African Reserve Bank (SARB) Deputy Governor Francois Groepe coupled with a few other declarations that validate my point.

On May 24 of 2018, the central bank official told reporters in Pretoria: “We don’t use the term ‘cryptocurrency’ because it doesn’t meet the requirements of money in the economic sense of the stable means of exchange, a unit of measure and a stable unit of value... We prefer to use the word ‘cyber-token’.” Furthermore, in January of 2018, Nigerian central bank Governor Godwin Emefiele proclaimed that investing in Bitcoin is a “gamble”, while Lesotho’s central bank announced a month later, that it will not offer any recourse to investors who lose money on virtual currency trades. Additionally, in May of 2018, Bank for International Settlements (BIS) General Manager Agustin Carstens told the German newspaper Börsen-Zeitung that he prefers to call these currencies “cryptoassets.”174

174 see. Amogelang et al., 2018.

91 All this information subsequently leads me to the conclusion that it is essential to make a distinction between fiat currencies and digital currencies. Moreover, I think it is necessary to separate Bitcoin and Ethereum.

Based on my observations, I can confidently conclude that Bitcoin is not a cheaper and more effective alternative to traditional fiat currencies. It is, however, an anonymous, more complex, riskier, and more volatile option to those traditional currencies. While Bitcoin’s sole purpose is to serve as a virtual currency, Ethereum, on the other hand, is the developers’ blockchain, hence deeply intertwined with innovation, which in my opinion boosts its implied intrinsic value beyond levels that are comprehendible to the human mind, yet.

Although I may have concluded that a differentiation between fiat currencies and virtual currencies as well as Bitcoin and Ethereum has to be made, the majority of people is not in this position, and the reason for that is very simple; because they don’t have to be.

With all that in mind, it becomes evident that an average rational human is never going to actively consider taking part in the blockchain environment if he or she is not being properly persuaded by offering a globally accessible, secure, and flawless service which is cheaper as well as more effortless to handle than existing alternatives.

Therefore, I suggest further research aimed at examining if this aforementioned chasm is gradually getting wider as time moves on, and if so, what possible implications can be deduced from this. Because one thing is certain, it doesn’t matter how good your product is, if you cannot attract a sufficiently high number of consumers, who are ready to accept it, then you won’t succeed.

This brings me to the second important conclusion which I was able to extract during my research. In my opinion, a successful prediction market platform needs to meet three main conditions. First, there must be a smooth-running and regularly maintained platform which is able to constantly attract new users. Second, the system should be uncensorable and users should be anonymous. Simply because, I believe that trust is not something that is given but rather has to be earned. Third, and finally, the platform must have a high participation as well as liquidity in order to produce viable results.

92 Centralized applications, such as Vetr, are censorable and providers can easily identify their participants, which is violating condition number two. Augur, on the other hand, was particularly designed to fulfill the second condition. Nevertheless, I observed that the application is nowhere near operating flawlessly and participation as well as liquidity presently stagnates at worryingly low levels.

This current low user rate can be explained by numerous barriers to entry for the average individual, as well as the costs for using Augur’s platform, which remain too high to offer a competitive alternative to existing comparable services. Thus, it can be concluded that Augur’s primary goal should be to acquire a critical mass of users as quickly as possible. However, this process cannot be forced too much, because the system and people, including the regulatory environment, need time to mature and adapt.

Although this slow integration was expected by developers and insiders, forecasted to take between 2-3 years after launch, my research suggests rather unexpected alarmingly low levels of adoption since the official launch date, implying a slower acceptance process than anticipated.

With that being said, even though it takes longer than expected, I still believe in the tremendous potential of the blockchain technology, especially Ethereum, and in the combination with prediction markets as a forecasting tool. They have the potential to change the world and fortunately both are here to stay, providing us with the prime opportunity of being able to witness what will happen next. Because the truth is, nobody knows what the future holds, but with decentralized prediction markets everyone gets the chance to pseudonymously make an educated guess.

Since I am convinced that it is just a matter of time until Augur is ready to deliver reliable results, further observation as well as a subsequent examination of performance, regarding volumes and accuracy, is highly recommended.

93 Bibliography

Abramowicz, 2007 Abramowicz, Michael. "The hidden beauty of the quadratic market scoring rule: A uniform liquidity market maker, with variations." GWU Law School Public Law Research Paper 251 (2007).

Adjodat, 2016 Adjodat, Babak. “Interview with Chief Scientist and Co-founder of Sentient, Babak Adjodat.” Telephone interview conducted by Shrier et al., March 14, 2016. (Found in Shrier et al., 2016, p. 11).

Amogelang et al., 2018 Amogelang Mbatha, Ntando Thukwana and Odwa Mjo. “Crypto Money Is a Token, Not a Currency, South Africa’s Central Bank Says.” Bloomberg.com, May 25, 2018: https://www.bloomberg.com/news/articles/2018-05-24/crypto-money-is-a-token-not-a- currency-s-africa-s-sarb-says (Accessed on May 3, 2019).

Armstrong, 2001 Armstrong, Jon Scott, ed. Principles of forecasting: a handbook for researchers and practitioners. Vol. 30. Springer Science & Business Media, 2001.

Arrow et al., 2008 Arrow, Kenneth J., et al. "The promise of prediction markets." (2008): 877-878.

Artificial Lawyer, 2018 Artificial Lawyer. “Hype Killer - Only 1% of Companies Are Using Blockchain, Gartner Reports.” ArtificialLawyer.com, May 4, 2018: https://www.artificiallawyer.com/2018/05/04/hype-killer-only-1-of-companies-are-using- blockchain-gartner-reports/ (Accessed April 22, 2019).

Augur FAQs, 2018 Augur. “Frequently Asked Questions.” Augur.net, 2018: https://www.augur.net/faq/ (Accessed November 21, 2018).

Augur Whitepaper, 2018 Krug et al. "Augur: a Decentralized Oracle and Prediction Market Platform." July 12, 2018: https://github.com/AugurProject/whitepaper/blob/52b5c9c54d763479ea651fece67b9cb11c21 4f06/english/whitepaper.pdf (Accessed July 22, 2018).

Aziz, 2018 Aziz. “Coins, Tokens & Altcoins: What’s the Difference?” Masterthecrypto.com, 2018: https://masterthecrypto.com/differences-between-cryptocurrency-coins-and-tokens/ (Accessed November 13, 2018).

94 Bell, 2009 Bell, Tom W. "Private prediction markets and the law." (2009).

Berg/Proebsting, 2009 Berg, Henry, and Todd A. Proebsting. "Hanson's automated market maker." Journal of Prediction Markets 3.1 (2009): 45-59.

Berg et al., 2006 Berg, Joyce, Forrest Nelson, and Thomas Rietz. "Accuracy and forecast standard error of prediction markets." Tippie College of Business Administration, University of Iowa, mimeo (2006).

Berg et al., 2008 Berg, Joyce, et al. "Results from a dozen years of election futures markets research." Handbook of experimental economics results 1 (2008): 742-751.

Berg et al., 2009 Berg, Joyce E., George R. Neumann, and Thomas A. Rietz. "Searching for Google's value: Using prediction markets to forecast market capitalization prior to an initial public offering." Management Science 55.3 (2009): 348-361.

BlitzPredict Whitepaper, 2018 BlitzPredict. “BlitzPredict Whitepaper.” 2018: https://icorating.com/upload/whitepaper/FNNnxJCZbAr0ltu3kMByuZMdpw5K4gPpV9mAY 9JZ.pdf (Accessed September 12, 2018).

Bloomberg, 2018 Leising, Matthew. “As Crypto Meets Prediction Markets, Regulators Take Notice.“ Bloomberg.com, July 26, 2018: https://www.bloomberg.com/news/articles/2018-07-26/as- crypto-meets-prediction-markets-u-s-regulators-take-notice (Accessed September 13, 2018).

Bodhi Whitepaper, 2018 Lin, Xiahong. “Bodhi: Decentralized Predictive Market at Scale.” Whitepaper V1.0, 2018: https://www.bodhi.network/file/Bodhi_English_White_Paper.pdf (Accessed July 25, 2018).

Bonneau et al., 2014 J. Clark, J. Bonneau, E.W. Felten, J.A. Kroll, A. Miller, and A. Narayanan. On decentralizing prediction markets and order books. WEIS '14: Proceedings of the 10th Workshop on the Economics of Information Security, June 2014. Cadwalladr/Graham-Harrison, 2018 Cadwalladr, Carole and Emma, Graham-Harrison. “Revealed: 50 million Facebook profiles harvested for Cambridge Analytica in major data breach.” Theguardian.com, March 17, 2018: https://www.theguardian.com/news/2018/mar/17/cambridge-analytica-facebook-influence- us-election (Accessed January 22, 2019).

95 Camerer, 1998 Camerer, Colin F. "Can asset markets be manipulated? A field experiment with racetrack betting." Journal of Political Economy 106.3 (1998): 457-482.

Catalini/Gans, 2016 Catalini, Christian, and Joshua S. Gans. Some simple economics of the blockchain. No. w22952. National Bureau of Economic Research, 2016.

Chen/Plott, 2002 Chen, Kay-Yut, and Charles R. Plott. "Information aggregation mechanisms: Concept, design and implementation for a sales forecasting problem." (2002).

Chen/Krakovsky, 2010 Chen, Kay-Yut, and Marina Krakovsky. Secrets of the moneylab: How behavioral economics can improve your business. Penguin, 2010.

Cherry, 2007 Cherry, Steven. "Bet on it." IEEE Spectrum 44.9 (2007): 48-53.

Clark et al., 2014 Clark, Jeremy, et al. "On decentralizing prediction markets and order books." Workshop on the Economics of Information Security, State College, Pennsylvania. 2014.

CME Group, 2018 CME Group. "Timeline of CME Achievements." CME Group: How the World Advances, 2018: http://www.cmegroup.com/company/history/timeline-of-achievements.html (Accessed May 25, 2018).

CoinMarketCap, 2018 CoinMarketCap. “All Cryptocurrencies.” Coinmarketcap.com, 2018: https://coinmarketcap.com/all/views/all/ (Accessed on December 18, 2018).

Cointelegraph, 2019 Cointelegraph. “ICO vs IPO: Key Differences.” Cointelegraph.com, 2019: https://cointelegraph.com/ico-101/ico-vs-ipo-key-differences (Accessed April 22, 2019).

Cowgill et al., 2009 Cowgill, Bo, Justin Wolfers, and Eric Zitzewitz. "Using Prediction Markets to Track Information Flows: Evidence from Google." AMMA. 2009.

Craib, 2016 Craib, Richard. “Interview with Founder of Numerai, Richard Craib.” Telephone interview conducted by Shrier et al., March 14, 2016. (Found in Shrier et al., 2016, p. 11).

Crunchbase, 2018 Crunchbase. “Vetr: Overview.” Crunchbase.com, 2018: https://www.crunchbase.com/organization/vetr#section-overview (Accessed July 25, 2018).

96 Dahan et al., 2010 Dahan, Ely, Arina Soukhoroukova, and Martin Spann. "New product development 2.0: Preference markets—how scalable securities markets identify winning product concepts and attributes." Journal of Product Innovation Management 27.7 (2010): 937-954.

DappRadar, 2019 DappRadar. “Augur: Users and Volume comparison.” Dappradar.com, 2019: https://dappradar.com/app/673/augur (Accessed January 3, 2019).

Delphy Whitepaper, 2018 Wang, Bo. “Delphy Foundation: Revolutionary Platform for Predicting Future Events.” Delphy.org, 2018: https://delphy.org/src/papers/Delphy_Whitepaper_EN.pdf (Accessed July 22, 2018).

Eidoo, 2019 Eidoo. “ERC20 Tokens list.” Eidoo.io, 2019: https://eidoo.io/erc20-tokens-list/ (Accessed January 5, 2019).

Erikson/Wlezien, 2008 Erikson, Robert S., and Christopher Wlezien. "Are political markets really superior to polls as election predictors?" Public Opinion Quarterly 72.2 (2008): 190-215.

Erikson/Wlezien, 2012 Erikson, Robert S., and Christopher Wlezien. "Markets vs. polls as election predictors: An historical assessment." Electoral Studies 31.3 (2012): 532-539.

Etherscan, 2019 Etherscan. “REP Token Holder Addresses.” Etherscan.io, 2019: https://etherscan.io/token/0xe94327d07fc17907b4db788e5adf2ed424addff6 (Accessed January 18, 2019).

Fair/Shiller, 1990 Fair, Ray C., and Robert J. Shiller. "Comparing information in forecasts from econometric models." American Economic Review 80.3 (1990): 375-389.

Forsythe et al., 1992 Forsythe, Robert, et al. "Anatomy of an experimental political stock market." The American Economic Review (1992): 1142-1161.

Frankenfield, 2018 Frankenfield, Jake. “Definition of Gas (Ethereum).” Investopedia.com, March 13, 2018: https://www.investopedia.com/terms/g/gas-price-ethereum.asp (Accessed January 14, 2019).

Galton, 1907 Galton, Francis. "Vox populi (the wisdom of crowds)." Nature 75.7 (1907): 450-451.

97 Gao, 2018 Gao, Ada. “Summary of Bo Wang, Founder of Delphy’s Interview with Hecaijing.” Medium.com, May 21, 2018: https://medium.com/delphy/summary-of-bo-wang-founder-of- delphys-interview-with-hecaijing-3e96aa682b48 (Accessed January 23, 2019).

Gartner, 2018 Gartner. "Gartner Hype Cycle." Gartner.com, 2018: http://www.gartner.com/technology/research/methodologies/hype-cycle.jsp (Accessed June 4, 2018). Gaspoz, 2010 Gaspoz, Cédric. Prediction Markets Supporting Technology Assessment. Diss. Université de Lausanne, Faculté des hautes études commerciales, 2010.

Götze et al., 2008 Götze, Uwe, Deryl Northcott, and Peter Schuster. "Investment appraisal." Methods and Models, Berlin, Heidelberg 2008 (2008).

Graefe/Armstrong, 2011 Graefe, Andreas, and J. Scott Armstrong. "Comparing face-to-face meetings, nominal groups, Delphi and prediction markets on an estimation task." International Journal of Forecasting 27.1 (2011): 183-195.

Graefe, 2017 Graefe, Andreas. "Political markets." The Sage Handbook of Electoral Behavior. Los Angeles: Sage. FC (2017).

Gürkaynak/Wolfers, 2006 Gürkaynak, Refet, and Justin Wolfers. Macroeconomic derivatives: An initial analysis of market-based macro forecasts, uncertainty, and risk. No. w11929. National Bureau of Economic Research, 2006.

Hanson, 2003 Hanson, Robin. "Combinatorial information market design." Information Systems Frontiers 5.1 (2003): 107-119.

Hanson/Oprea, 2004 Hanson, Robin, and Ryan Oprea. "Manipulators increase information market accuracy." George Mason University (2004).

Hanson/Oprea, 2009 Hanson, Robin, and Ryan Oprea. "A manipulator can aid prediction market accuracy." Economica 76.302 (2009): 304-314.

Hanson et al., 2006 Hanson, Robin, Ryan Oprea, and David Porter. "Information aggregation and manipulation in an experimental market." Journal of Economic Behavior & Organization 60.4 (2006): 449- 459.

98 Harvey et al., 1998 Harvey, David I., Stephen J. Leybourne, and Paul Newbold. "Tests for forecast encompassing." Journal of Business & Economic Statistics 16.2 (1998): 254-259.

Hauge, 2018 Hauge, Hans. “Modern Cryptocurrency Portfolio Theory.” Seekingalpha.com, September 12, 2018: https://seekingalpha.com/article/4205809-modern-cryptocurrency-portfolio-theory (Accessed September 16, 2018).

Hausch/Ziemba, 2007 Hausch, Donald and William Ziemba, eds. 2007. Handbook of Sports and Lottery Markets. Elsevier: Handbooks in Finance series.

Hayek, 1945 Hayek, Friedrich August. "The use of knowledge in society." The American economic review 35.4 (1945): 519-530.

Hongkiat, 2018 Hongkiat, Lim. “How to Setup TREZOR Wallet for the First Time.” Hongkiat.com, April 9, 2018: https://www.hongkiat.com/blog/setup-trezor-beginner/ (Accessed January 7, 2019).

Iansiti/Lakhani, 2017 Iansiti, Marco, and Karim R. Lakhani. "The truth about blockchain." Harvard Business Review 95.1 (2017): 118-127.

ICObench, 2018 ICObench. “Augur ICO.” ICObench.com, 2018: https://icobench.com/ico/augur (Accessed August 12, 2018).

Johnson et al., 2008 Johnson, Mark W., Clayton M. Christensen, and Henning Kagermann. "Reinventing your business model." Harvard business review 86.12 (2008): 57-68.

Jullien/Salanié, 2000 Jullien, Bruno, and Bernard Salanié. "Estimating preferences under risk: The case of racetrack bettors." Journal of Political Economy 108.3 (2000): 503-530.

Krug, 2018 Krug, Joey. “Fees, Fees, and… Fees?” Medium.com, August 8, 2018: https://medium.com/@joeykrug/fees-fees-and-fees-8939c2b5ecae (Accessed January 14, 2019).

Lalley, 2015 Lalley, Steven P., and E. Glen Weyl. "Quadratic Voting. SSRN Scholarly Paper ID 2003531." Social Science Research Network, Rochester, NY (2015).

99 Lavoie, 2009 Lavoie, Jim. "The innovation engine at rite-solutions: Lessons from the CEO." The Journal of Prediction Markets 3.1 (2009): 1.

Ledbetter, 2008 Ledbetter, James. "The Fizz-dom of Crowds." Slate.com, April 16, 2008: http://www.slate.com/articles/news_and_politics/hey_wait_a_minute/2008/04/the_fizzdom_o f_crowds.html (Accessed June 17, 2018).

Leigh et al., 2003 Leigh, Andrew, Justin Wolfers, and Eric Zitzewitz. What do financial markets think of war in Iraq? No. w9587. National Bureau of Economic Research, 2003.

Leitch/Tanner, 1991 Leitch, Gordon, and J. Ernest Tanner. "Economic forecast evaluation: profits versus the conventional error measures." The American Economic Review (1991): 580-590.

Leonard, 2015 Leonard, Shane. “Why has Social failed in Fintech?” Techcrunch.com, August 6, 2015: https://techcrunch.com/2015/06/08/why-has-social-failed-in-fintech/?guccounter=1 (Accessed April 19, 2019).

Leonhardt, 2008 Leonhardt, David. "Looking for Sure Political Bets at Online Prediction Market." Nytimes.com, February 13, 2008: http://www.nytimes.com/2008/02/13/business/13leonhardt.html (Accessed June 15, 2018). Luckner, 2008 Luckner, Stefan. "Prediction markets: Fundamentals, key design elements, and applications." BLED 2008 Proceedings (2008): 27.

Luckner et al., 2008 Luckner, Stefan, Jan Schröder, and Christian Slamka. "On the forecast accuracy of sports prediction markets." Negotiation, auctions, and market engineering. Springer, Berlin, Heidelberg, 2008. 227-234.

Malkiel/Fama, 1970 Malkiel, Burton G., and Eugene F. Fama. "Efficient capital markets: A review of theory and empirical work." The journal of Finance 25.2 (1970): 383-417.

Malone/Bernstein, 2015 Malone, Thomas W., and Michael S. Bernstein, eds. Handbook of collective intelligence. MIT Press, 2015.

Manski, 2006 Manski, Charles F. "Interpreting the predictions of prediction markets." economics letters 91.3 (2006): 425-429.

100 Marinovic et al., 2010 Marinovic, Iván, Marco Ottaviani, and Peter Norman Sørensen. "Modeling idea markets: Between beauty contests and prediction markets." (2010).

Marinovic et al., 2013 Marinovic, Iván, Marco Ottaviani, and Peter Sorensen. "Forecasters’ objectives and strategies." Handbook of economic forecasting. Vol. 2. Elsevier, 2013. 690-720.

Marvin, 2017 Marvin, Rob. "Blockchain: The invisible tech that’s changing the world." PC Magazine 2, August 29, 2017: https://uk.pcmag.com/features/87703/blockchain-the-invisible-technology- thats-changing-the-world (Accessed April 22, 2019).

Medium, 2015 ConsenSys. “Why & How Decentralized Prediction Markets Will Change Just About Everything.” Medium.com, December 16, 2015: https://medium.com/@ConsenSys/why-how- decentralized-prediction-markets-will-change-just-about-everything-15ff02c98f7c (Accessed January 18, 2019).

Medium, 2018a Bonpay. “What is the Difference Between Coins and Tokens?” Medium.com, March 13, 2018: https://medium.com/@bonpay/what-is-the-difference-between-coins-and-tokens- 6cedff311c31 (Accessed November 17, 2018).

Medium, 2018b Augur. “Augur Launches.” Medium.com, July 9, 2018: https://medium.com/@AugurProject/augur-launches-794fa7f88c6a (Accessed July 10, 2018).

Medium, 2018c Ecurrencyhodler. “Deflationary Currency, Bitcoin, and Just a Store of Value?” Medium.com, July 18, 2018: https://medium.com/@ecurrencyhodler/deflationary-economics-bitcoin-and- just-a-store-of-value-70c4078abd85 (Accessed November 17, 2018).

Metz, 2016 Metz, Cade. “The Rise of the Artificially Intelligent Hedge Fund.” Wired.com, January 25, 2016: http://www.wired.com/2016/01/the-rise-of-the-artificially-intelligent-hedge-fund/ (Accessed January 18, 2019). Moore, 2002 Moore, Geoffrey A. "Crossing the Chasm, revised edition." New York: HarperBusiness Essentials (2002).

Nakamoto, 2009 Nakamoto, Satoshi. “Bitcoin v0.1 released.” Mail-archive.com, January 9, 2009: https://www.mail-archive.com/[email protected]/msg10142.html (Accessed September 14, 2018).

101 Narayanan, 2016 Narayanan, Arvind, et al. Bitcoin and cryptocurrency technologies: A comprehensive introduction. Princeton University Press, 2016.

Oprea et al., 2011 Oprea, Ryan, et al. "Can Manipulators Mislead Prediction Market Observers?" George Manson University, mimeo (2011).

Ortner, 1998 Ortner, Gerhard. “Forecasting markets—An industrial application.” TU Vienna, mimeo (1998).

Pennock et al., 2001 Pennock, David M., et al. "The real power of artificial markets." Science 291.5506 (2001): 987- 988.

Pentland, 2013 Pentland, Alex Sandy. "Beyond the echo chamber." Harvard Business Review 91.11 (2013): 80-+.

Peterson, 2018 Peterson, Andy. ”No Escape: Augur Burns Key to Network ’Killswitch’.” CCN.com, July 25, 2018: https://www.ccn.com/no-escape-augur-burns-key-to-network-killswitch/ (Accessed July 28, 2018). Polgreen et al., 2007 Polgreen, Philip M., et al. "Use of prediction markets to forecast infectious disease activity." Clinical Infectious Diseases 44.2 (2007): 272-279.

Predictions.Global, 2019 Predictions.Global. “Prediction Markets powered by Augur.” Predictions.Global, 2019: https://predictions.global/ (Accessed January 3, 2019).

Raval, 2016 Raval, Siraj. "What Is a Decentralized Application?" Decentralized Applications: Harnessing Bitcoin's Blockchain Technology. O'Reilly Media, Inc., 2016. 1-2.

Rennock et al., 2018 Rennock, Michael J. W., Alan Cohn, and Jared R. Butcher. "Blockchain technology." The Journal (2018).

Rhode/Strumpf, 2004 Rhode, Paul W., and Koleman S. Strumpf. "Historical presidential betting markets." Journal of Economic Perspectives 18.2 (2004): 127-141.

Rhode/Strumpf, 2008a Rhode, Paul W., and Koleman S. Strumpf. "Historical Election Betting Markets: An International Perspective." University of Arizona, mimeo (2008).

102 Rhode/Strumpf, 2008b Rhode, Paul W., and Koleman S. Strumpf. "Manipulating political stock markets: A field experiment and a century of observational data." University of North Carolina (2008).

Rhode/Strumpf, 2014 Rhode, P. W., and K. S. Strumpf. "The long history of political betting." The Oxford Handbook of the Economics of Gambling (2014): 560-586.

Rice, 2014 Rice, Andrew. "The Fall Of Intrade And The Business Of Betting On Real Life." Buzzfeed.com, February 21, 2014: https://www.buzzfeed.com/andrewrice/the-fall-of-intrade- and-the-business-of-betting-on-real-life?utm_term=.nqmaQw1Na#.dkKLBrjAL (Accessed June 27, 2018).

Rogers, 2003 Rogers, Everett M. "Diffusion of innovations. Free Press." New York 551 (2003).

Roll, 1984 Roll, Richard. "Orange juice and weather." The American Economic Review 74.5 (1984): 861- 880.

Rothschild, 2009 Rothschild, David. "Forecasting elections: Comparing prediction markets, polls, and their biases." Public Opinion Quarterly 73.5 (2009): 895-916.

Schumpeter, 1939 Schumpeter, Joseph Alois. Business cycles: a theoretical, historical, and statistical analysis of the capitalist process. New York: McGraw-Hill, 1939.

Servan-Schreiber et al., 2004 Servan-Schreiber, Emile, et al. "Prediction markets: Does money matter?". Electronic markets 14.3 (2004): 243-251.

Seward, 2013 Seward, Zachary M. “This is how much a Bloomberg terminal costs.” Qz.com, May 15, 2013: https://qz.com/84961/this-is-how-much-a-bloomberg-terminal-costs/ (Accessed November 5, 2018).

Shrier et al., 2016 Shrier, David, et al. Prediction markets. Technical report. Massachusetts Institute of Technology, Cambridge, 2016.

Shvarts/Green, 2007 Shvarts, R., & Green, S. Cost-Benefit Analysis and Prediction Markets and the Proposal for University of Oregon Prediction Markets. Diss. Department of Economics, University of Oregon, 2007.

103 Snowberg/Wolfers, 2010 Snowberg, Erik, and Justin Wolfers. "Explaining the favorite–long shot bias: Is it risk-love or misperceptions?" Journal of Political Economy 118.4 (2010): 723-746.

Snowberg et al., 2005 Snowberg, Erik, Justin Wolfers, and Eric Zitzewitz. "18 Information (in) efficiency in prediction markets." Information efficiency in financial and betting markets (2005): 366.

Snowberg et al., 2008 Snowberg, Erik, Justin Wolfers, and Eric Zitzewitz. "Prediction Markets: From Politics to Business (and Back)." Handbook of Sports and Lottery Markets. Elsevier, 2008. 385-402.

Snowberg et al., 2012 Snowberg, Erik, Justin Wolfers, and Eric Zitzewitz. "Prediction markets for economic forecasting." Handbook of Economic Forecasting. Vol. 2. Elsevier, 2013. 657-687.

Spann/Skiera, 2009 Spann, Martin, and Bernd Skiera. "Sports forecasting: a comparison of the forecast accuracy of prediction markets, betting odds and tipsters." Journal of Forecasting 28.1 (2009): 55-72.

Spears et al., 2009 Spears, Brian, and Christina LaComb. "Examining trader behavior in idea markets: an implementation of GE's imagination markets." The Journal of Prediction Markets 3.1 (2009): 17-39.

Statista, 2019 Statista. “Size of the Bitcoin blockchain from 2010 to 2019, by quarter (in megabytes).” Statista.com, 2019: https://www.statista.com/statistics/647523/worldwide-bitcoin-blockchain- size/ (Accessed April 22, 2019).

Stox Whitepaper, 2018 Stox. “Stox Platform for Prediction Markets.” Whitepaper V03, 2018: https://resources.stox.com/stox-whitepaper.pdf (Accessed August 23, 2018).

Strumpf, 2012 Strumpf, Koleman. "Introduction to special issue on corporate applications of prediction markets." The Journal of Prediction Markets 3.1 (2012).

Sunstein, 2004 Surowiecki, James. The wisdom of crowds. Anchor, 2005. 1-7. (A Review by Cass R. Sunstein).

Surowiecki, 2004 Surowiecki, James. "The wisdom of crowds: Why the many are smarter than the few and how collective wisdom shapes business." Economies, Societies and Nations 296 (2004).

104 Swan, 2013 Swan, Melanie. “Prediction Markets.” A presentation at the Singularity University, San Jose, 2013: https://www.slideshare.net/lablogga/prediction-markets-singularity-university-2013 (Accessed May 17, 2018).

Tetlock, 2004 Tetlock, Paul. "How efficient are information markets? Evidence from an online exchange." Social Science Research Network (2004).

The Economist, 2002 Economist. “Doomsday derivatives.” Economist.com, October 17, 2002: https://www.economist.com/finance-and-economics/2002/10/17/doomsday-derivatives (Accessed April 18, 2019).

Thompson, 2013 Thompson, Clive. Smarter than you think: How technology is changing our minds for the better. Penguin, 2013.

Tompkins et al., 2008 Tompkins, Robert G., William T. Ziemba, and Stewart D. Hodges. "The Favorite-Longshot Bias in S&P 500 and FTSE 100 Index Futures Options: The Return to Bets and the Cost of Insurance." Handbook of Sports and Lottery Markets. Elsevier, 2008. 161-180. Twesige, 2015 Twesige, Richard Lee. "A simple explanation of Bitcoin and Blockchain technology." (2015).

Tziralis/Tatsiopoulos, 2012 Tziralis, George, and Ilias Tatsiopoulos. "Prediction markets: An extended literature review." The journal of prediction markets 1.1 (2012): 75-91.

Vetr Blog, 2017 Vetr. ”Vetr presents at Demo Day 4.0, hosted by FinTech Sandbox.” Vetr Blog, April 21, 2017: http://blog.vetr.com/2017/04/21/vetr-presents-demo-day-4-0-hosted-fintech-sandbox/ (Accessed July 26, 2018).

Vetr Company Presentation, 2017 see. Vien, 2017: https://www.youtube.com/watch?v=jU3YGEZsIXA#t=119 (Accessed July 26, 2018).

Vetr Story, 2018 Vetr. “About Vetr.” Vetr.com, 2018: https://www.vetr.com/about/story (Accessed July 26, 2018).

Vien, 2017 Vien, Mike. “Vetr presentation.” Boston, April 20, 2017: https://www.youtube.com/watch?v=jU3YGEZsIXA#t=119 (Accessed July 26, 2018).

105 Williams, 2005 Williams, Leighton Vaughan, ed. Information efficiency in financial and betting markets. Cambridge University Press, 2005.

Wolfers/Zitzewitz, 2004 Wolfers, Justin, and Eric Zitzewitz. "Prediction markets." Journal of economic perspectives 18.2 (2004): 107-126.

Wolfers/Zitzewitz, 2006 Wolfers, Justin, and Eric Zitzewitz. Prediction markets in theory and practice. No. w12083. national bureau of economic research, 2006.

World Blockchain Foundation, 2018 World Blockchain Foundation. “All About Blockchain.” Worldblockchainfoundation.org, 2018: https://worldblockchainfoundation.org/blockchain-101/ (Accessed December 12, 2018).

Zheng et al., 2017 Zheng, Zibin, et al. "An overview of blockchain technology: Architecture, consensus, and future trends." 2017 IEEE International Congress on Big Data (BigData Congress). IEEE, 2017.

Zitzewitz, 2007 Zitzewitz, Eric. “Manipulation can Affect Prices.” Midasoracle.org, May 30, 2007: http://www.midasoracle.org/2007/05/30/manipulation-can-affect-prices/ (Accessed April 17, 2019).

106