Cross posting from here for visibility: Price Elasticity of Gas Demand on Ethereum and Arbitrum - Arbitrum Research
Problem
We study the question of how the blockspace demand is affected by the gas fee change on Ethereum base layer and Arbitrum One rollup. This is a non-trivial question as the gas fees are themselves determined by the aggregate blockspace demand, through an EIP-1559 style mechanism, creating endogeneity. Naive approaches such as regression methods lead to incorrect conclusions. For example, a simple pooled OLS regression on 16.9 million Arbitrum observations yield a positive elasticity of +0.094, which contradicts the expectation of negative demand response. This is a structural bias that cannot be overcome with more data. Getting a true causal estimate of the price elasticity requires more care.
Methodology
To address the above problem, we combine three steps. First, we cluster wallet addresses into six behavioral clusters based on their on-chain activity, transaction frequency, and resource consumption. This is important because a pooled estimate averages across different types of wallets such as oracle updaters, MEV searchers, and retail users. Second, we use two-way fixed effects (wallet and time) analysis, where each data point is a wallet, fixed time interval pair and the transaction count acts as a control variable. Third, we use wallet’s own lagged base fee as an instrument variable to separate the causal fee variation from congestion driven endogeneity.
In addition to the aggregate gas, we also extend the analysis to seven resource dimensions on Arbitrum: computation, storage reads/writes, storage growth, calldata, history growth, and refunds.
Findings
Both chains are inelastic in the aggregate. The pooled elasticities are −0.006 on Ethereum and −0.036 on Arbitrum, meaning a 10% fee increase reduces total gas demand by roughly 0.06% and 0.36% respectively. L2 exhibits a higher level of elasticity partly because the users are more price sensitive and also because of the presence of probabilistic backrunners.
The pooled elasticity estimates mask the heterogeneity. We find that cluster level elasticities are nearly 6x the pooled estimate. Moreover, the resource level decomposition reveals an even wider spread in the elasticities. Insights obtained in the analysis can be used to answer questions of pricing mechanism, opcode repricing, explaining usage patterns, etc.
For details, check out the paper: https://arxiv.org/pdf/2606.13555. Any feedback is welcome.