tl;dr Here's an unusual theory: Bitcoin is decentralized only when mining is unprofitable on average. This is fine - people will participate anyway, for the same reason they play state lotteries and satoshi dice.
Introduction The Bitcoin whitepaper sketches a consensus proof assuming 51% of the mining power is honest. But this assumption isn't meant to be swallowed whole - it's the role of Bitcoin's incentive scheme to *keep* the majority honest, by encouraging lots of users to participate independently. To date there's been no formal justification of the particular incentive scheme, nor any economic model to clarify what rationality assumptions are necessary, nor a usable definition for what "decentralized" means. Here's a description of a simple economic game that characterizes all this.
Decentralized Mining Game First assume there is a population of potential participants (later I'll make assumptions about their budgets, preferences, etc). The game has two phases, a negotiation/bidding phase and a mining phase.
- In the negotiation phase, every participant decides how much money to spend on Bitcoin mining. Since each participant's payoff depends on the total amount of hashpower (everyone else's participation), this requires many rounds of negotiation or a magic auctioneer who knows everyone's curve. Imagine that it works like this: each participant is told the total hashpower of everyone else's bids, and is given the choice of changing their own bid. This is repeated until no one changes their current bid so an equilibrium is reached.
- In the mining phase, everyone mines at the rate of their final bid. The game ends after the first (or some finite number) of blocks is found, and everyone learns their payoff.
This is a bit pedantic and oversimplified, especially because the mining and bidding are separate phases, and the game ends rather than running forever, but it's abstract and amenable to analysis this way. Now suppose the population consists of two types of participants, Clients (which represent solo miners and pool participants), and Servers (which represent
large professional mining operations. I can now define "decentralized" as a desirable outcome for the bidding phase: an incentive scheme is "decentralized" if some minimum fraction of the total hashpower comes from the Clients, and not just from the Servers.
Expected Value rationality model To make things interesting, suppose that the Servers enjoy a discount on hashpower due to bulk economies of scale, and that every participant only tries to maximize their expected profit. This leads to an undesirable result for Bitcoin. If the Servers are in equilibrium, then the marginal profit from adding more mining power is zero. When this is the case, then it is *unprofitable* for Clients to mine at all.
Cumulative Prospect Theory rationality model Even though "maximize expected value" is the common wisdom for deciding whether to bet on a risk, it's known this doesn't predict very well how people *actually* make risky decisions. (See
Allais Paradox) There'd be no casinos (except for fraud), no state lotteries, and no satoshidice. There's a popular alternate model of rationality called
Cumulative Prospect Theory that's almost as simple and describes empirical evidence much better. It essentially says that people overweight small probabilities and underweight near certainties. This predicts, in particular, that people play lotteries with sufficiently large jackpots, even when the chance of winning is small. Not everyone's curves are the same - a related phenomenon is that poor people tend to play lotteries with much lower variance and smaller jackpots (
Are All Lotteries Regressive? Evidence from the Powerball).
So this suggests an alternate rational model, where the Servers are wealthy (above median), and are essentially the same as expected-profit maximizing (except around really large multi-million dollar jackpots), while Clients are poorer (below median?) and overweight small probabilities, even for smaller jackpots (like thousands of dollars or less).
In this model, there's a solution where the jackpot size is chosen appropriately, such that it's unprofitable on average for everyone, including the Servers, yet Clients continue to play anyway.
DiscussionThe reward right now for winning a single block is several thousand dollars. This is probably not enough to excite a billionaire or large firm to invest in mining Bitcoin directly. (Selling mining equipment or hosted mining is another story though.) In fact, the prevalence of pooled mining suggests there's an appetite for even *lower* variance and smaller jackpots, which could mean even lower block bonuses may be sustainable. Modern lotteries are designed to appeal to as wide a market as possible, and typically involve a
large number of games with different jackpots and probabilities. In this view, Bitcoin's payout scheme, with only one game, is very primitive and suboptimal. Pure gambling sites like SatoshiDice compete directly against Bitcoin mining for the same market. There is obviously a lot more subtlety to include in the economic model, such as the improvement in miner efficiency over time, fixed costs, and the fact that mining is continuous rather than a one shot game. This also doesn't account for other reasons to mine, like entertainment value or altruistically contributing to the network. I'm not sure whether any of these would change the main outcome.
I really like the lottery model of Bitcoin participation. Bitcoin doesn't just compete with state monopoly on currency, it also competes with the state monopoly on lotteries! State lotteries are a form of voluntary tax - they are typically justified by directing lottery profits towards a public good, such as education.
Historically, lotteries were used to fund defense, like armies and even the Great Wall of China. Unprofitable Bitcoin mining can be seen as a mechanism for raising funds towards the public good of securing digital transactions.
I anticipate that one reaction to this theory is that I'm calling Bitcoin miners foolish for participating in unprofitable bets. To the contrary, I am beginning to think that this CPT is not just a better descriptive rationality model than EU, but maybe it's also a reasonable prescriptive one too. Perhaps the overweighting of small probabilities is an evolved behavioral trait that leads self-interested individuals to contribute to public goods without coercion. Bitcoin might simply be harnessing this more effectively than state lotteries.