This paper measures how much on‑chain activity falls when gas prices rise on Ethereum mainnet (layer 1, or L1) and on the Arbitrum rollup (layer 2, or L2). The authors find that demand is nearly inelastic on Ethereum in 2025: a 10% rise in the base fee reduces total gas use by about 0.06% (elasticity ≈ −0.006). On Arbitrum, for October 2025 through April 2026, the pooled estimate is slightly larger but still small: a 10% fee rise cuts gas use by about 0.36% (elasticity ≈ −0.036). These are causal estimates intended to help calibrate fee‑mechanism simulations and policy analysis.

To get causal estimates the researchers use transaction data and a panel regression with wallet and time fixed effects. They remove a common bias from congestion by using each wallet’s own lagged base fee as an instrument. In plain terms, fees on these chains are adjusted automatically by a rule called EIP‑1559 (Ethereum Improvement Proposal 1559), so current fees partly reflect past fees. The lagged fee gives a source of fee variation that is predetermined relative to a wallet’s current choices. Simple regressions that ignore this problem can even show the wrong sign: a raw pooled ordinary least squares regression on Arbitrum gave a positive coefficient (+0.094), which is driven by congestion rather than true user response.

The paper uses large datasets. For L1 it covers all Ethereum transactions in 2025 and studies 44,449 wallets that submitted at least 500 transactions that year, producing 35.3 million wallet‑day observations. For Arbitrum it covers October 2025–April 2026 with 23,119 wallets meeting the same 500‑transaction threshold, giving 16.9 million wallet‑hour observations. Arbitrum also records gas consumption by resource type, which the authors use to break demand down by operation: computation, storage reads and writes, storage growth, history growth, calldata, and refunds.