The Blockchain Folk Theorem - Department of Economics...
Transcript of The Blockchain Folk Theorem - Department of Economics...
Introduction Preview of results Model Analysis & results Conclusion
The Blockchain Folk TheoremAn economic analysis of the blockchain
Bruno Biais (TSE), Christophe Bisiere (TSE),Matthieu Bouvard (McGill) and Catherine Casamatta (TSE)
Preliminary
April 2017
Introduction Preview of results Model Analysis & results Conclusion
Blockchain
Distributed ledger, records transactions and ownership, operatedwithin a peer to peer network
Bitcoin blockchain: ownership of bitcoins
Blockchain can be used for other assets & contracts (Ethereum)
If reliable and stable, new cost effective way to record transactionsand ownership
Is it?
Introduction Preview of results Model Analysis & results Conclusion
Nakamoto (2008): Steps in Blockchain
“The steps to run the network are as follows:1. New transactions are broadcast to all nodes.2. Each node collects new transactions into a block.3. Each node works on finding a difficult
proof-of-work for its block.4. When a node finds a proof-of-work, it broadcasts
the block to all nodes.5. Nodes accept the block only if all transactions in it
are valid and not already spent.6. Nodes express their acceptance of the block by
working on creating the next block in the chain, using thehash of the accepted block as the previous hash.”
Introduction Preview of results Model Analysis & results Conclusion
Blockchain
!me
ini!alstockoftransac!onsinB1
minersworkoncryptographicproblem(proofofwork)
minersolvesB1
broadcaststoall
nodescheckproofofwork,transac!onsvalid,notspent,thenexpressacceptancebychainingnextblocktoB1
Block1
Block1
Block1B1 B2 B3
tB1 tB2 tB3
transac!onsbetween0andtB1
0
minersolvesB2
broadcaststoall
nodescheckthenchainnextblocktoB2
…
Introduction Preview of results Model Analysis & results Conclusion
Nakamoto (2008): Longest chain rule (LCR)
To which previously solved block will miners chain their block?
“Nodes always consider the longest chain to be thecorrect one and will keep working on extending it.”
If miners follow LCR → no fork, single chain: consensus on thestate of the registry
In practice forks sometimes occur, and consensus disappears (atleast temporarily)
Introduction Preview of results Model Analysis & results Conclusion
March 11, 2013 fork on Bitcoin
Vitalik Buterin (March 13, 2013, bitcoinmagazine.com)
“Starting from block 225430, the blockchain literally split
into two, with one half of the network adding blocks to one
version of the chain, and the other half adding to the other.
[...] The split lasted for 24 blocks or 6 hours [...]”
Introduction Preview of results Model Analysis & results Conclusion
July 20, 2016 fork on EthereumSource: blog.ethereum.org
Introduction Preview of results Model Analysis & results Conclusion
Can we expect miners to follow LCR?
We take a game-theoretic approach to study the strategies ofminers: to which block do they chain their own block?
We consider a frictionless environment: all miners observe blocksand transactions simultaneously, no double spending
In this perfect environment, can we expect stable and reliabledistributed consensus?
Is LCR an equilibrium of the game between miners? Yes
Is it the only equilibrium? No
Introduction Preview of results Model Analysis & results Conclusion
What are miners mining for?
When miner solves block Bn in Bitcoin blockchain, he broadcaststhis, including his reward, in bitcoins, in block Bn
This reward can’t be spent immediately, only after sufficientlymany blocks chained to Bn (k-blocks rule)
Miners want to
• mine on chain which they think will become consensus, sotheir rewards are valuable → coordination
• protect value of reward they obtained and still hold → vestedinterest
Introduction Preview of results Model Analysis & results Conclusion
Mining is a coordination game
Miners want to attach their blocks to the chain to which theyexpect the others will attach their blocks
=⇒ Choice of which block to mine = coordination game
=⇒ Multiple equilibria, some without forks and others with forks
Problem with forks → loss of rewards for orphaned blocks,blockchain instability, uncertainty about consensus, underminescredibility/value of cryptocurrency
Introduction Preview of results Model Analysis & results Conclusion
Coordination issues during March 2013 Bitcoin fork
Two competing chains (0.7 versus 0.8): miners didn’t know whichchain to coordinate on
23:22 Gavin Andresen: the 0.8 fork is longer, yes? somajority hashpower is 0.8 ... first rule of bitcoin: majorityhashpower wins
23:23 Luke Dashjr: if we go with 0.8 we are hard forking23:43 BTC Guild: I can single handedly put 0.7 back to
the majority hashpower. I just need confirmation that that’swhat should be done
23:44 Pieter Wuille: that is what you should do, but we
should have consensus first
Eventually, BTC Guild chose 0.7 and 0.8 was orphaned
Reflecting uncertainty, price of Bitcoin dropped 25% during fork
Introduction Preview of results Model Analysis & results Conclusion
Mining generates vested interest
Miner keeps rewards for solving block (k-block rule, preferredsavings vehicle)
As long as keeps rewards earned on a chain, vested interest in thatchain being consensus
→ keeps mining on that chain to make sure it is consensus
→ earns block on that chain
→ strengthens vested interest
→ persistent competing chains?
Introduction Preview of results Model Analysis & results Conclusion
Vested interest during March 2013 Bitcoin fork
23:24 Luke Dashjr: it’s either lose 6 blocks [mined on 0.8]or hardfork [to 0.8]
23:25 Pieter Wuille: all old miners will stick to their oldchain regardless of the mining power behind the other
23:57 BTC Guild: I’ve lost so much money in the last 24hours from 0.8
By switching BTC Guild loses the work they’ve done on
0.8 since the fork started. On the other hand they are more or
less assured that the 0.7 branch will win
Source: freedom-to-tinker.com/2015/07/28/analyzing-the-2013-bitcoin-fork
Introduction Preview of results Model Analysis & results Conclusion
Miners and pools
Miners, m ∈ M = {1, ...M}, risk neutral, rational, infinitelypatient
In practice miners work in pools, which coordinate the efforts oftheir miners, in particular as regards which blocks they mine. Mcan be interpreted as number of pools
April 20th 2017: 14 mining pools = 93% of total hash capacity →14 = reasonable order of magnitude for M
Relatively limited number of pools motivates game theoreticapproach (but coordination effects also there with competitivebehaviour)
Introduction Preview of results Model Analysis & results Conclusion
Hash rate distribution in the Bitcoin network
Source: https://blockchain.info/pools?timespan=4days
Introduction Preview of results Model Analysis & results Conclusion
Transactions and blocks
We assume:
• Exogenous continuous flow of transactions sent forconfirmation by end-users
• Instantaneously observed by miners (removes one of thepotential causes for forks noted by Nakamoto (2008) and yetforks arise in equilibrium)
Stock of transactions available at time 0 = block B1
All participants start mining B1 at time 0
Introduction Preview of results Model Analysis & results Conclusion
Time to solve blocks
Nakamoto (2008): time it takes m to solve block = exponentialwith intensity θm (proof of work)
We assume accordingly that probability that m solves his blockbetween t and t + dt is θmdt independently of
• how long m has been mining the block
• which block m is mining
• which block the others are mining
In practice, θm also depends on difficulty of cryptographic problem,revised every 2,016 blocks (on Bitcoin). We consider a stationaryenvironment (θm constant): appropriate for short term horizon (≤2 weeks)
Introduction Preview of results Model Analysis & results Conclusion
Times at which miner m solves blocks
Introduction Preview of results Model Analysis & results Conclusion
Times at which miner m’ solves blocks
Introduction Preview of results Model Analysis & results Conclusion
Times at which miner m” solves blocks
Introduction Preview of results Model Analysis & results Conclusion
Stochastic maturity
At time zm (exponential with intensity λm) miner m hit by liquidityshock → must consume real goods → sells cryptocurrency earnedas reward to a new miner, who also inherits his beliefs
Exit compensated by entry → stationary environment
Before zm miner m keeps cryptocurrency obtained as reward(k-blocks rule)
Introduction Preview of results Model Analysis & results Conclusion
Rewards for solving blocks
In block he mines, m includes his reward: G units of cryptocurrency(eg Bitcoins or ETH) (+ transaction fee, much smaller)
Value of reward (obtained at zm) for mining block in chain Bdepends on number of miners active in B
For large part of the analysis, we only require G increasing andG (0) = G (1) = 0 : when there is no miner or only one miner on achain, the rewards on that (quasi) orphaned chain are valueless
Introduction Preview of results Model Analysis & results Conclusion
LCR is a Nash equilibrium
Proposition 1:There exists a Nash equilibrium in which, at each date, allparticipants choose to mine the longest chain.
If m follows LCR, like the others=⇒ at zm only one chain (with M active miners)=⇒ each block mined by m earns G (M)=⇒ optimal (at least weakly) for m to follow LCR
Introduction Preview of results Model Analysis & results Conclusion
Coordination rather than competition
Miners are not really competing to solve their block before theothers
That someone else solves his block before me, does not, in itself,reduce my gains
The only thing that matters is that we all coordinate well, and allwork on the same chain, so that we all obtain maximum rewardsfor the blocks we solve
Introduction Preview of results Model Analysis & results Conclusion
Fork with orphaned branch
So far, focus on LCR. But LCR not always equilibrium:
Proposition 3: There exists a Nash equilibrium in which all minerscreate a fork from the main chain f blocks backward, and thenfollow the LCR on that fork.
Intuition: expect all miners fork f blocks backward → expectrewards on that chain to be more valuable → also fork:coordination game - strategic complementarity (would be therealso if miners competitive)
Consequence: Fork becomes the only active chain; rewards in theabandoned chain are lost; others don’t earn more =⇒ forkequilibrium Pareto dominated by LCR equilibrium
Introduction Preview of results Model Analysis & results Conclusion
Coordination & forks
Arvind Narayanan (2015) “Analyzing the 2013 Bitcoin fork:centralized decision-making saved the day”:
“One way to look at this is that BTC Guild sacrificedrevenues for the good of the network. But these actionscan also be justified from a revenue-maximizingperspective. If the BTC Guild operator believed that the0.7 branch would win anyway (perhaps the developerswould be able to convince another large pool operator),then moving first is relatively best, since delaying wouldonly take BTC Guild further down the doomed branch.”
http://freedom-to-tinker.com/2015/07/28/
analyzing-the-2013-bitcoin-fork
Introduction Preview of results Model Analysis & results Conclusion
Persistent forks
Can it happen that some miners fork to new chain whose parent isnot the previously solved block while others do not?
Old chain: {B1, ...Bn−f ,Bn−f+1, ...,Bn−f+i}
New chain: {B1, ...Bn−, Bn−f+1, ..., Bn−f+j}.
t
Bn−f Bn−f+1 Bn−f+i
tBn−f
Old chain with M −K miners
Bn−f+1 Bn−f+j
tBn
New chain with K miners
Introduction Preview of results Model Analysis & results Conclusion
Persistent fork
Vested interest: number of blocks solved by a miner on old chainat time of fork
Rank the miners by vested interest in old chain
Proposition 4: Under certain conditions, there exists a Nashequilibrium in which miners with low vested interest on the oldchain decide to fork f blocks backward and keep working on thisfork, while miners with large vested interest continue to mine onthe old chain.
Introduction Preview of results Model Analysis & results Conclusion
Intuition for Proposition 4
For some to fork, we need K ≥ M2 : Persistent forks can occur only
if majority of miners fork
Benefit from forking = blocks mined on new chain more valuable;large if expects to solve many blocks in the future, i.e. if largecomputing power and unfrequent liquidity shock
Cost of forking = reduces value of blocks already mined on oldchain; large if vested interest large and frequent liquidity shock
Persistent fork equilibrium is Pareto dominated
Introduction Preview of results Model Analysis & results Conclusion
ConclusionLCR equilibrium and single persistent chain with no fork cannot beruled out... but cannot be taken for granted
Coordination game: Number of miners/computing power on achain → Credibility of chain → Value of rewards for blocks minedon that chain → Attractiveness of that chain
Multiple equilibria
• Instability?
• Pareto dominated (waste of resources)
Next steps
• double spending & non instantaneous transmission: sameeffects?
• endogenous computing power: overinvestment?
• protocol design to reduce inefficiencies: communicationamong miners, reward design