Flash Boys 2.0: priority gas auctions and front-running (Daian et al. 2019)

When an arbitrage opportunity appears on a decentralized exchange, only the first transaction to execute can take it, and transactions in a block run in fee order. Daian et al. watched bots compete by repeatedly replacing their pending transaction with one paying a higher fee: a priority gas auction. Each replacement must raise the fee by a minimum step (10% in go-ethereum), so with bots that value the opportunity alike, bidding stops only when the next step would exceed its value \(V\):

\[\text{price} > \frac{V}{1.1}, \qquad \text{profit} < V - \frac{V}{1.1} \approx 0.09\,V.\]

Competition hands more than nine tenths of the value to whoever orders the block. Daian et al. called it miner extractable value, and showed that it can even pay a miner to reorder or rewrite recent blocks.

import matplotlib.pyplot as plt

import blockchainkit as bk
from blockchainkit.economics.visualizers import plot_gas_auction

Two bots and one opportunity

OPPORTUNITY = 50_000
auction = bk.economics.priority_gas_auction(OPPORTUNITY, ["bot A", "bot B"], start=10, seed=2)
print(f"{len(auction.bids)} bids; {auction.winner} pays {auction.price:,} for {OPPORTUNITY:,}")
assert auction.price > OPPORTUNITY / 1.1 and auction.profit < 0.1 * OPPORTUNITY

fig, (left, right) = plt.subplots(1, 2, figsize=(12, 4.5))
plot_gas_auction(auction, opportunity=OPPORTUNITY, ax=left)

86 bids; bot B pays 48,164 for 50,000

The bump decides how much the bots keep

bumps = [1, 2, 5, 10, 12, 15, 20, 25, 30]
kept = []
for bump in bumps:
    shares = [
        bk.economics.priority_gas_auction(OPPORTUNITY, ["a", "b", "c"], bump=bump, seed=s).profit
        / OPPORTUNITY
        for s in range(20)
    ]
    kept.append(sum(shares) / len(shares))
    assert max(shares) < bump / (100 + bump) + 1 / OPPORTUNITY
right.plot(bumps, kept, "o-", color="#2563eb", label="winner's profit")
right.plot(bumps, [b / (100 + b) for b in bumps], ":", color="black", label="bound b / (100 + b)")
right.set(xlabel="minimum raise (%)", ylabel="fraction of the opportunity kept")
right.set_title("Everything else goes to the block producer")
right.legend()
fig.tight_layout()

plt.show()

Exercise

If the block producer runs its own bot, it pays its fees to itself. How much of the opportunity does it keep, and what does that suggest about who ends up ordering transactions?