In a scenario where data is bought and sold in a double-auction market, will the trading price and volatility of a given piece of data become inversely proportional to its accuracy over time? In other words, will "true" data tend to propagate until there is no scarcity, while falsehoods tend to trade at higher and more volatile prices?
The reason why this would be useful to know is that, if "truth is cheap", then volatility and long-term price trend can be used as an indicator of accuracy.
Definition of terms:
- "True" data is more accurate, less noisy, more useful to its buyers. A simple example of true data would be a current and correct list of phone numbers for a set of people.
- "False" data is less accurate, noisy, obsolete, or outright wrong, less useful and more risky for its buyers. A simple example of false data would be an old and partially incorrect list of phone numbers for a set of people.
Note that by these definitions, "true" or "false" is not a boolean -- a set of data can be partially false.
Note also that, until the buyer tests the data, they have no way of knowing whether it is true or false -- they may need to buy two conflicting sets of the same data and test both -- this generates some intermittent demand for "false" data.
I think the answer to this question may be "yes".
Sellers would tend to raise their ask prices in order to compensate for their increased risk when selling data they themselves believe to be suspect, leading to a wider bid-ask spread and lower liquidity. See the "possible implementation" section below for what the sellers' risk might be.
Likewise, buyers would tend to lower their bids for data that scores lower in their purchasing heuristics -- their scoring might include known prior data quality from the same seller, number of sellers offering the same data hash, age of data, etc. Again see the "implementation" section below.
If a seller believes a piece of data is more correct, they will believe there to be less risk, and they will move ask prices lower in order to remain competitive. This lowers the bid-ask spread as well as volatility of matched prices.
If "truth is cheap" turns out to be correct, then systems based on this principle could be useful both in low-level distributed algorithms and in high-level social systems and governance. I've been pondering this question myself for a while for an open-source project I want to launch, and I figured I'd check in here before starting significant work on it.
One specific low-level problem I want to be able to solve, for example, is in choosing the most valid block of data for a particular file or record in a distributed system. Existing distributed systems have difficulty hosting more than one economic actor on a peer-to-peer basis without explosive growth of the data itself, and a decreasing signal-to-noise ratio. If anyone can add data to a commons regardless of validity or usefulness, then the commons tends to become more noisy over time. The traditional cure for this is active curation by founders or their delegates. That doesn't work as well in truly distributed systems where founders don't own most of the nodes.
Traditionally, conflicts in distributed data systems are often prevented by locking before writing, and resolving conflicts when they do arise is traditionally handled by voting or merging, often with humans in the loop. Replication of data for safety in distributed systems is traditionally either manually specified or on a best-effort basis.
I think market structure likely matters a great deal in whether "truth is cheap" works or not. Here are some example conditions which I think would enable this:
- A subset of traders have ways of testing data for accuracy. They discard inaccurate data, and offer unlimited copies of the remainder for sale.
- A subset of traders need accurate data for their own internal use, and have the option of applying complex algorithms to choose between competing sellers.
- Sellers provide a cryptographic hash of the data in their offer, but not the data itself. This allows the buyer to count how many sellers are offering the same data, and enables the buyer to choose the lowest price.
- Sellers digitally sign their offers.
- Sellers incur risk -- testers may sell information describing bad data, including the original offer and seller's signature. Another possible source of risk could be contracts with some sort of forced settlement similar to options assignment.
I'm obviously handwaving a lot of details in the interest of brevity, and am completely skipping over currency, accounting, and transparency of peer account balances. Feel free to ask if something seems missing.