Distributed Prediction Contracts
A Contrarian Market Prediction Method
TABLE OF CONTENTS
1. OPENING
2. HOW DOES THIS WORK?
3. THE FIVE CONDITIONS
4. REAL LIFE EXAMPLES
5. REWARD VALUE DETERMINATION
6. EQUILIBRIUM AND INFORMATION
7. VOTER BEHAVIOR
8. TRUSTED VOTE MANAGER
9. ABOUT GO BLOCK USA
10. PICKFLIX
11. OPEN AND FREE USE
OPENING
A Distributed Prediction Contract is a contrarian market prediction model developed by GoBlock USA LLC. This model uses the concepts of crowd solving and behavioral economics to closely approximate the results of future events whose outcomes can be measured along a scale.
A Distributed Prediction Contract is a scalar prediction model. Events that are binary (yes/no, on/off, win/lose) are not scalar, and therefore may not be accurately predicted by a Distributed Prediction Contract. Scalar events are those that have results which can occur along a continuum of possible outcomes (or are “distributed” along a continuum of outcomes). The purpose of the Distributed Prediction Contract is to predict where on that continuum an event will land.
Why a Distributed Prediction Contract is also considered a contrarian market prediction model will be explained as well - and needs a little more background first.
HOW DOES IT WORK?
A Distributed Prediction Contract utilizes crowd solving logic. Therefore, a “crowd” of people are required to predict the outcome. The people in the crowd who are performing the act of predicting (called “voters”) purchase votes (or “contracts”) for a particular outcome of the scalar event. The voters hold these contracts for future redemption once the event has concluded, and the results are measured. The Reward Value (or redemption value) of each contract is determined based on those results.
Traditional prediction models reward voters by how CLOSE the actual result is to the predicted result. However, a contrarian prediction model rewards voters by how DIFFERENT the actual result is from the predicted result. The greater the degree of difference of actual from predicted, the greater the change in the Reward Value of the contact. Therefore, to maximize reward in a contrarian prediction model, voters are incented to purchase contracts that differ in their prediction from what the crowd is predicting.
A real-life example will be provided next, and we will first talk about the five conditions that must exist for a Distributed Prediction Contract to work properly.
THE FIVE CONDITIONS
1. Number of Entities. There must be at least two competing entities (companies, products, entrants, protagonists, etc.) involved in the market being predicted.
2. Non-Random Outcome. The outcome cannot be determined randomly, such as by dice, coin flip, wheel, acts of god or nature, or other games of chance. The outcome must instead be influenced by the behavior of the market.
3. Cost to Vote. The cost to vote (e.g., purchase a contract) must be equal for all competing entities, and equal for all outcomes. In certain market prediction scenarios, it may be important to increase (or decrease) contract price as the voting window begins to close. But, for purposes of this introductory paper, we will force the price to stay constant during the entire voting window.
4. Information. All voters must have access to real time and accurate information on the number of contracts purchased by all voters. Only with this information will a voter be able to know the crowd-predicted outcome at the time of purchase, and make a decision on whether to vote differently than the crowd.
5. Trusted Vote Manager. An impartial and trusted party is required to manage the sale of voting contracts, distribution of information, certification of results, determination of rewards, and buyback of the voting contracts.
REAL LIFE EXAMPLES
A Distributed Prediction Contract can be used to predict a wide range of market-based outcomes, such as:
1. Which new smart phone will dominate the market?
2. Will a new TV show steal viewership from an existing hit TV show?
3. What percentage of drivers will use pay express lanes vs free regular lanes?
Let’s take the example of the smart phone market…
Assume there are three smart phone providers (A,B,C) all releasing a new smart phone in the near term. Each phone will have its own contract (A,B,C), and voters will buy contracts for the smartphone for which they want to vote. As more contracts are sold for a particular phone, that is an indicator that more voters feel that phone will sell more units.
Since this is contrarian prediction model, voters want to maximize their reward by purchasing the contract for the phone which balances two things:
· Casting a vote for a phone that they believe will do well in the market. That is, they believe there is an intrinsic value in the item for which they are voting.
· Casting a vote for a phone that has the fewest contracts (votes) sold. That is, the fewest people think the item has intrinsic value.
These two factors contradict and compete with each other in the decision-making process of the voter. If everyone’s opinion is that Phone A will sell more units, then everyone will be buying contracts for Phone A. But since contract Reward Values are determined by how different the actual result is from the predicted result, then it may make sense to buy a contract for Phone B instead. If the large crowd buying Phone A has overestimated the sales of Phone A, and Phone B results are therefore greater than predicted, then the reward for the Phone B contract will also be greater than predicted. Going against the crowd can be rewarding!
To step through the timing of the example:
Voting Opens: All contracts for all phones can be purchased for the same value. Let’s say $1 in this example. If a voter believes that Phone A will sell more units, they buy a contract for Phone A. If they believe instead that Phone C will be the leader, they buy a contract for Phone C. If a voter strongly believes that a certain phone will be the leader, they can vote multiple times for that phone, and buy multiple contracts for the same phone. Or, buy contracts for Phone C and Phone B. There is not a limit or restriction.
Open Voting Window: As contracts are purchased in the course of the open voting window, there becomes a distribution of contracts across Phones A, B, and C. Very simply, from this distribution, the expected market share of each Phone can be predicted based on the percentage of contracts sold. If 14% of the contracts are sold for Phone C, then the market should expect that approximately 14% of the actual phones sold will be Phone C.
Let use these hypothetical purchased contracts counts for our example and keep explaining.
Phone A: 200 contracts Percent of Total: 57.14%
Phone B: 100 contracts Percent of Total: 28.57%
Phone C: 50 contracts Percent of Total: 14.29%
TOTAL 350 contracts Percent of TOTAL: 100%
Expected Reward Value: If this turns out to be true (and 14.29% of the actual phones sold are Phone C), then the Reward Value for a Phone C contract will be equal to $1. We will do the math on this further in a bit, and this is a major concept that needs to be understood. If the crowd precisely predicts the actual result, then there is no gain or loss. The contract Reward Value is equal to the contract purchase price in this instance.
However, if the actual percentage of Phone C sold is GREATER than 14.29%, then the Reward Value of the contract will be HIGHER.
If the actual percentage of Phone C sold is LESS than 14.29%, then the Reward Value of the contract will be LOWER.
As stated above, it is the DIFFERENCE of the predicted results from the actual results which drives the reward. If the voters guess exactly right, then they break even.
Reacting to Votes during Open Voting Window: As voting continues, if voters feel that Phone C is under-predicted (that is, they believe Phone C will sell more than the 14.29% which the crowd predicts) then it is in their best interest to purchase a contract for Phone C. That purchase increases the number of contracts sold for Phone C and also the expected market share for Phone C. At the same time, that purchase also lowers the expected market share for Phones A and B. This fluctuation of contracts sold and predicted market share continues during the course of the open voting window.
Voting Close and Actual Results: When voting closes, the number of contracts sold becomes locked, and the actual results are determined from the actual market data. The Reward Value for each contract is then calculated. In this example, lets use the number of contracts sold below as the locked and final count.
Phone A: 250 contracts Percent of Total: 47.17%
Phone B: 130 contracts Percent of Total: 24.53%
Phone C: 150 contracts Percent of Total: 28.30%
TOTAL: 530 contracts Percent of TOTAL: 100%
At close of voting, Phone C is now predicted to sell 28.3% of the phones among the three companies.
If the ACTUAL results look like this:
Phone A: 45% of phones sold = LESS than Predicted
Phone B: 25% of phones sold = MORE than Predicted
Phone C: 30% of phones sold = MORE than Predicted
We can now calculate the Reward Value for each contract using $1 as the starting point, and then increase (or decrease) that reward by the relative difference of predicted results from actual results.
ACTUAL RESULT ÷ PREDICTED RESULT = RATIO X $1 = REWARD
Phone A: 45% 47.17% .954 $.954 / contract
Phone B: 25% 24.53% 1.019 $1.019 / contract
Phone C: 30% 28.30% 1.060 $1.060 / contract
In this example, Phone A under-performed the prediction, and that contract is now worth less. Phone B and C both over-performed the prediction, and those contracts are now worth more.
It is worth noting that even though Phone A sold the most units in the market, Phone A underperformed its prediction, and therefore the contract decreased in value.
This highlights that the Distributed Prediction Contract is not a method to predict a “winner”, but instead a method to predict an overall market, and then to reward voters based on the relative over (or under) performance of the market to that prediction.
Redeeming Reward: Now that results are in, and the Reward Value calculated, the voters can redeem their contracts for the Reward Value. The process is now complete.
REWARD VALUE DETERMINATION
To discuss the determination of Reward Value further, there are 3 factors which affect that value calculation:
1. Total Number of All Contracts Purchased
2. Number of Contracts Purchased for each Protagonist in the market
3. Actual Relative Result of each Protagonist to each other
Therefore, two relationships can be derived from these factors:
I. As the Actual Relative Result of a Protagonist increases, its Reward Value increases. *
II. As the Number of Contracts Purchased for a Protagonist increases, its Reward Value decreases.
Therefore, the mathematical formula to calculate the Reward Value for each Protagonist's contract is:
Reward Value = Actual Relative Result ÷ Contracts Purchased for Protagonist X Contract
of Protagonist Total Number of all Contracts Price
That is, the better a Protagonist does against its competition (Actual Relative Result) and the fewer the number of voters which predicted that result (Number of Contracts Purchased for Protagonist) then the higher the Reward Value of the contract. Hence, the value in being contrarian, and picking against the crowd!
Take an example of Starbucks vs Dunkin Donuts, and a prediction of how many cups of coffee of each will sell in a promotional period. Here are the final contract counts for voting, and the market results (once voting closed and results were in), and the resulting Reward Values.
Purchase Price per Contract = $1
Starbucks Dunkin Donuts TOTAL
Cups of Coffee Sold 1,100,000 650,000 1,750,000
Percent of Cups (Relative Result) 62.86% 37.14% 100%
Contracts Sold 100 50 150
Predicted Market Result 66.67% 33.33%
Reward Value per Contract $.943 $1.114
In this case, Dunkin Donuts performed better than predicted and the Reward Value is higher than the purchase price.
* There are some market prediction scenarios where the actual relative result is preferred to decrease when compared to the competition; and therefore, the Reward Value is calculated as an inverse measure. For example, fewer negative side effects of a drug is more desirable when compared to those of another drug. So, a market prediction model for medical trials may want to determine which drug will create fewer side effects. In this case, rewards would be calculated on the inverse. But, for the examples in this paper, we will keep it simple with a direct relationship.
EQUILIBRIUM AND INFORMATION
Under a contrarian market prediction method, a voter is incented to buy contracts in a Protagonist that she feels is not correctly predicted by the crowd. For example, if she has an opinion that Dunkin Donuts will perform better than the current prediction, then she will begin buying Dunkin contracts. This will increase the predicted market result (relative to Starbucks), but also decrease the Reward Value of the Dunkin contract. Eventually, the predicted result for Dunkin will be too high, and the Reward Value for Dunkin too low, and the voters will begin to buy Starbucks contracts.
In this way, the voters will search for an equilibrium (or balance point) that closely approximates the predicted market result while providing an economic reward for the voter.
(see illustration at top of page)
Informed and Relevant Voters: In order to allow this “tug of war” to occur on the journey to equilibrium, voters require accurate and up to date information on the contract counts for each Protagonist. While not always possible to provide perfect information, the farther the Distributed Prediction Contract system moves away from perfect information, the greater the margin of error in the equilibrium prediction.
The voters also have to be relevant to the market being predicted. For example, Chinese farmers are not (usually) relevant to the American baby diaper market. Instead, American parents of young children should be voting on the American baby diaper market. And, the Vote Manager should provide relevant data about the baby diaper market to help the relevant voters make an educated choice. The farther the Distributed Prediction Contract system moves away from educated and relevant voters, the greater the margin of error in the equilibrium prediction.
VOTER BEHAVIOR
The Distributed Prediction Contract system is premised on the following economic behaviors of the voter.
Stake: there must be stake (or risk) to the voter. Without stake, a voter may allow other factors to influence their voting decision (such as emotion, prejudice, favoritism, or whim). When value is at risk, decisions are taken more seriously, and decisions are based on market conditions. Stake also helps encourage a higher percentage of relevant voters. Without stake, voters could vote on any market (relevant or non-relevant) without consequence.
Anonymity: all voters and votes must be anonymous. If a voter is subject to public scrutiny, the voter will tend to use emotion or agenda, rather than economic or financial influencers, and may corrupt their voting decisions.
Convergence: given everything above (stake, anonymity, informed and relevant voters) the crowd’s votes in a Distributed Prediction Contract system will begin to converge on the predicted outcome. The size of the crowd required to properly predict the result differs by instance, but generally as the number of voters increases, the margin of error in the prediction decreases.
TRUSTED VOTE MANAGER
Finally, the voters must trust that the Reward Value of their contract will be redeemable and calculated correctly based on the actual market results. To the end, the Vote Manager must provide accurate and timely information during the open voting period, report reliable and verifiable actual market results, show transparent Reward Value calculations, and provide timely redemption of the contracts to the voters.
Without a trusted Vote Manager to issue and redeem voting contracts, there is no Distributed Prediction Contract system.
ABOUT GO BLOCK
GoBlock USA LLC is a group of motivated blockchain enthusiasts who are excited to be involved in this emerging technology. Our background is in traditional enterprise and internet systems connecting supply chains and the real world. Based in Atlanta, Georgia, and with our fantastic development team in San Francisco, we have a mantra: "Be Excellent to Each Other and Block On!"
Have questions about GoBlock or the Distributed Prediction Contract, contact us at contact@goblock.co
PICKFLIX
GoBlock USA LLC is deploying a Distributed Prediction Contract system in the market of new release theatrical movies. We will be the trusted Vote Manager in this market. Enabled by Ethereum blockchain smart contracts, voters can predict the box office market share of competing movies in a new release period. Using this Distributed Prediction Contract of movies, we hope to gather valuable information in the further research of the Distributed Prediction Contract method.
The Ethereum blockchain Smart Contract provides for a strong and reliable vehicle to meet the requirements of a Distributed Prediction Contract system. A Smart Contract provides access to a large voting crowd for just about any market, and meets the requirements of stake, transparency, anonymity, trusted and immutable information, auditable calculations, and payout certainty.
You can vote in this movie market at www.pickflix.co
OPEN AND FREE USE
We are blockchain believers. As the blockchain is open and transparent, so is GoBlock USA LLC. The information in this paper is free for you to use, reproduce, and improve upon in any way you see fit – non-commercial or commercial. Only by sharing information and research will the blockchain grow, and we hope this research is one more step in that journey.
Our request is only that you cite the source of your information as “developed by GoBlock USA LLC and used by general permission”.
Be Excellent to Each Other and Block On!
Obtaining Equilibrium Consensus using a Distributed Prediction Contract