Collective Intelligence: How do we produce consensus value?
At Lithium we believe in the superiority of our valuation model and its fundamental principle: Collective Intelligence. Across the ages, society has trusted the wisdom of the crowd where they believe in the collective knowledge of a group as expressed through their aggregated opinions. Consensus from multiple predictions are generally preferred to a single, isolated prediction.
Of course, if building a consistent and unbiased valuation model was as easy as gathering public opinion, there would be no need for Lithium, and the Web3 world would be much simpler. However the challenge is not collecting data and opinion, it is first ensuring the data is precise and meaningful, and second, how to process this data and summarize it into a single consensus.
To gather pertinent pricing insight from our community, we take guidance from the principles of the “Keynesian Beauty Contest”, the fruit of one of the greatest minds of modern Economics. Per Keynes’ theory, traders are essentially attempting to estimate others’ interpretation of the world’s anticipation. We consider this theory as the underlying thought process among our Experts which enables an accurate and well-thought price estimate. Click here to learn more about how our community prices NFT as a beauty contest.
Would simple averaging work? The Coin Toss taught us NOT!
Collecting meaningful data is only the first step to creating a superior pricing model. Once the data is collected, the question of how to aggregate, process and use it to build a single consensus becomes the next logical step.
Consider the most common aggregation method, averaging. While it may be tempting to process estimates from the community by simply averaging them, research and academia struggle to find common ground when it comes to the best way to aggregate. Using Ville Satopaa’s coin toss example we can demonstrate how averaging data is a bad idea when applied to prediction.
Prediction Coin Toss Example
Consider this scene: 2 individuals and 1 coin. Each individual must predict on which side the coin will land every time it is flipped into the air.
Person 1 magically knows the result every time. He says definitely heads or definitely tails each time.
Person 2 has no idea. He is smart and reasonably says the probability of heads is 0.5 for every coin toss.
In this 2 person game, the resulting probability prediction would be 0.75. If this becomes a game with lots of players, the probability prediction would become very close to 0.5
The average of these predictions would be a bad estimate
Simple averaging leads to poor forecasting accuracy as each person has access to varying amounts of information and varying prediction performance. Capturing these variations among the community is crucial to appropriately summarize community predictions into a single consensus.
As demonstrated from the Coin Toss example, when it comes to probabilities it is recommended to move predictions closer to upper or lower bounds (1 & 0), which is called extremization. However, when these forecasts involve real values, such as NFT pricing, extremization isn’t applicable anymore as it becomes difficult to ascertain which values may represent a confident forecast.
Taking this into account, Lithium has devised a novel extremization process called Reputation Clearing Mechanism. The Reputation Clearing Mechanism takes into consideration the confidence level and historical performance signaled by Reputation Point input when Price Experts submit their estimates. This may sound confusing at first, but throughout this article we will be going over the finer details and elements that constitute the Reputation Clearing Mechanism, so that by the end of this read, all should have a good grasp on how Lithium approaches the data processing to form a sensible and pertinent consensus.
Reputation Clearing Process
In previous article, we talked about what Reputation Point is and how it is the key influencing factor on a contribution’s weight in the final output. When a Price Expert is submitting the two estimates — Price to Buy and Price to Sell, each pair of estimates is tied to the Reputation Points the Expert has locked up for the Price Quest and is pooled with other estimates denominated in Reputation Points. At the close of a Pricing Quest, we match all these buy and sell estimates, starting from matching the lowest selling estimate to the lowest qualifying buying estimate. For example, a Price to Buy estimate of ETH 100 with 10 units of Reputation will be able to take up 10 units of Price to Sell estimate less than or equal to ETH 100. What’s more, each Expert’s estimates are translated into a Premium-to-Floor Price figure in real time in order to ensure all inputs are standardized for aggregation. For instance if an Expert considers an NFT to be worth 120ETH while the current Floor Price is 100, his valuation is considered to be 1.2x the Floor Price.
Through this Reputation Clearing, the protocol summarizes all prediction inputs into a single output, the Consensus, along with a valuation range with a Low Value and High Value. This allows us to overcome the complexity of extremizing predictions for real values, as we can bring out a reliable upper bound which is the price level above which the community collectively expects there would be no further interest in selling and buying.
It is the price of the last matched buy and sell estimates from the Reputation Clearing system . It indicates the price level above which the community believes that there would be no buyers and sellers willing to transact further.
It indicates the price below which the community believes there would be plenty of interest in the market ready to buy. It is the last matched buying and selling estimates, starting from matching the highest buying interest with the highest qualifying selling estimates .
It is the price above the Consensus Value that eager buyers could be willing to pay. It is the last matched buying and selling estimate from the Reputation Clearing excluding all estimates below or equal to the Consensus Value. It aims to reflect the situation where the perfect buyer could be willing to pay a premium in order to acquire a non-fungible token.
Readers should note that qualifying selling estimates with respect to a buying estimate are the highest selling estimate that is equal to or lower than such buying estimate. On the flip side, qualifying buying estimates with respect to a selling estimate are the highest buying estimate that is equal to or higher than the selling estimate.
Lithium Finance is the first decentralized NFT valuation protocol powered by collective intelligence and machine learning. Redefining NFT valuation approach through incentivizing honest assessment from community to reveal market sentiments.