String Depot

A postmodern market
May 16, 2023

1. Postmodern Stocks

When the stock market was first invented, people would buy a stock in hopes that the company would make money and then give them, the shareholders, some of the money in the form of dividends. Eventually they realized the getting-money-back part wasn’t actually necessary: today something like half of publicly traded companies don’t even pay dividends, they just reinvest their profits back into growth. On some level, this is weird: the value of a stock more or less tracks a company’s expected future profits, even though there’s not necessarily any expectation that shareholders will get those profits back. But it works because the shareholders still retain fractional control over the company in the form of voting rights, and so they basically own an asset, even if they never personally touch its output.

What if you took this further: what if stocks didn’t actually come with any share in the company? No dividends or voting rights? Then you’re just betting on the company - sort of. But unlike a wager on, say, a horse race, you’re not betting on any particular proposition about the company. You’re just betting something like “this stock ticker will sell for more later than it sells for now”.

Would you expect the price of a headless stock ticker, unaffiliated with the actual company, to correlate with that company’s profitability? Kind of. In one sense, this is a Keynesian beauty contest, and “present value of expected future cash flows” is a Schelling-point criterion, if only because of the resemblance to a normal stock market. But it’s arguably not the only Schelling point: “longest name wins” or “best-known company” or “memeiest company” might contribute to the pricing too.

It turns out that this happens from time to time on the actual stock market. In 2021, Ethan Allen’s ticker symbol was ETH, and so people memed the price up, because it sounded like Ethereum (they’ve since changed to ETD).

2. Postmodern Prediction Markets

Prediction markets are, in one sense, a generalization of stock markets. Instead of buying “I think the present value of’s expected future cash flows is $100 million”, you can buy “I think’s revenue next year will be 3% higher than this year” or “I think the CEO of will tweet the rocket emoji this month” or “I think widget prices will drop precipitously”.

The difference between regular markets and prediction markets is that with the latter, you’re betting on the evaluation of some proposition. This is more expressive, but it’s also inelegant in an obvious way and also a subtler way. The obvious way is that somebody has to evaluate the statement. Right now that’s a central party (the central market maker for real-money markets and Metaculus; the market creator for Manifold). You could imagine a modification where you have to get a consensus vote, with a board of directors or a jury or something, about which way the proposition resolved, but regardless, it’s messier than the stock market, where a company just earns what it earns and nobody has to pass the official judgment of how much value it created. And even then, you’re never going to completely eliminate ambiguity in the wording.

The subtler inelegance is that people (and large language models) don’t really conceive of the world in terms of propositions. All our mental models are just like, big complicated representations of concept-space. They don’t map neatly to Boolean true/false outputs. Proposition-evaluation is an abstraction we can use, but language is nebulous and fuzzy because the concepts people use to understand the world are nebulous and fuzzy, because the world is such a high-dimensional firehose of stuff.

Prediction markets themselves have landed on something like this: people on Manifold Markets have started making markets called “_____ Stock (Permanent)”. The permanent means they’ll never resolve, the stock means whoever made the market suggests you trade up and down based on sentiment about Pepsi or Emmanuel Macron or crypto or Norway.

As of this writing, Manifold provides this as the default description for a stock-type market: BUY: good SHORT: bad Market trades based on sentiment & never resolves.

I think you can go farther, though. In my ontology, BUY: buy, SHORT: short. Market trades based on market.

3. Postmodern Altcoins

When crypto was first invented, people made tokens for particular projects; then they realized they could also make shitcoins for fun; then they discovered the absolute minimum viable shitcoin, which is just a name.

In the winter of 2021, the Omicron variant of Covid started surging. So somebody made a coin with no special properties – it was an OlympusDAO clone – other than being called Omicron, and it went to the moon, because people wanted to bet on omicron. Not “omicron will cause at least 20,000 deaths” or “the WHO will declare omicron a variant of concern” but just “omicron”.

Nobody resolves this, because you can’t. Omicron-coin is real estate in concept-space.

4. String Depot

I made a thing called String Depot, where you can buy and sell words. It’s like omicron, but for every word – every <32-character-string (technically bytes32), in fact – so you don’t have to spin up a clone of Litecoin or whatever every time you need to bet on a subject instead of a predicate. You can buy stuff like “toadstool” and “kjarjge;ijregaio” and “alphabet inc” and “desantis 2024” and “superconductor” and “frank ocean will release new music this year” and “I still love you Kyle, take me back”.

How do you price something like this? You could use a Uniswap-like automated market maker (AMM). But since every single possible bytes32 can trade, we get a long tail of low-liquidity token pools. So we want a pricing mechanism that doesn’t require much liquidity for each individual token.

The solution String Depot uses is that strings are priced using a bonding curve. That means there isn’t a fixed supply; under the hood, each buy is actually a mint and each sell is a burn. This particular bonding curve is quadratic**, so the price of the marginal token is proportional to the square of the supply.

String Depot is currently live on Polygon.

*I know, I know, they’re technically bytes32 not strings – did you get your home at Home Depot?

** Q: Why’d you do a quadratic bonding curve, instead of linear or something else? A: I’m cargo-culting Bitclout. Please tell me if you know a good argument for a different curve!