BNB Chain's gas fee looks useful for watching next-day token liquidity. Across its non-native tokens, the gas coefficient is 0.2938, which Wang and Krištoufek translate into a 52.3% increase in Amihud illiquidity for a one-SD rise in log gas. The same-chain monitor is credible. The paper calls gas fees a "predictive" signal, though its regressions leave the warning horizon unmeasured.
Why fees might reach the order book
Non-native tokens on a smart-contract chain compete for block space, paid for in ETH, BNB or SOL. Higher gas raises the fixed cost of every swap. Arbitrageurs and market makers may widen quotes or abandon small pools, leaving trades with greater price impact. The authors expect BNB Chain's validator-priority auction to matter especially in its retail, small-cap universe, where fixed fees eat more of a trade. Ethereum has the base-fee burn introduced by the EIP-1559 fee reform, batching and Layer-2 migration as buffers. Solana uses local fee markets; a chain-wide fee can therefore miss what an individual program pays.
The CoinMarketCap panel contains 4,224 single-platform tokens and 2,039,270 token-days. Its main regression uses 2,003,075 observations across 2,831 dates, from 2017-08-23 to 2025-05-23; Artemis supplies gas fees in USD. Illiquidity is log(1 + |r| / (USD volume + 1)). The regression interacts one-day-lagged log gas with a platform dummy and includes BTC and native-token illiquidity factors. Each factor is a fixed-weight distributed lag over the prior seven days. Token fixed effects, lagged volatility and lagged log volume enter as controls, with errors clustered on token and date.
The coefficients are 0.0834 for Ethereum (SE 0.0112), 0.2938 for BNB Chain (SE 0.0287) and 0.0365 for Solana (SE 0.0171). Under two-way clustering on token and date, Ethereum and BNB Chain are significant at 1%; Solana is significant at 5%. A one-SD move in log gas corresponds to within-token illiquidity changes of 0.075, 0.181 and 0.040, respectively. The paper expresses those as Amihud increases of 18.0%, 52.3% and 6.8%. All three signs agree with its fee mechanism, as does their ranking.
That much I grant.
How early is the signal?
One day of lag inside a daily regression establishes timing. The controls include lagged volatility and lagged log volume, whose coefficient is -0.2426, yet omit lagged illiquidity. A persistent common state could therefore lift gas first and illiquidity next. The authors describe the test as conditional association after token fixed effects, lagged token-level controls and a BTC illiquidity benchmark. Their abstract and conclusion go further, calling gas fees a predictive signal and a usable leading indicator. We did not find an out-of-sample test or a forecast metric.
The authors acknowledge the shared driver: "Both gas fees and token-level Amihud illiquidity respond to common factors including block-space demand", alongside activity composition and broader crypto stress. They say causal identification would require fee-design shocks such as EIP-1559, Layer-2 launches and Solana outages, which they leave for later. A monitor can be useful without a causal estimate. Its incremental forecasting power out of sample remains untested.
The dollar fee and the missing date effects
The paper gives this reason for omitting date fixed effects: "We do not include unrestricted date fixed effects, because the BTC and gas-fee variables vary" at the date or platform-date level and would be absorbed. Date effects would absorb BTC. The three platform-interacted gas terms, however, have different fees on a given date, leaving cross-chain variation from which to estimate their slopes. In their absence, the BTC factor is the sole market-wide control; its coefficient is -0.0559, the wrong sign for a market-stress control. Gas measured in USD can also rise when ETH, BNB or SOL rallies, even if congestion does not change.
Scale raises another question. Raw USD volume sits in the denominator, putting the transform near zero for liquid tokens. We did not find a discussion of that scaling.
The stability tests miss gas
The authors call the gas coefficient "large and uniformly positive". Gas appears in one full-sample column. There, its inclusion moves the BNB native coefficient from -0.0729 to an insignificant -0.0175. The authors write: "The operational measure substitutes for the unstable price-based proxy." Their subsample tests then follow the BNB native coefficient, which changes from 0.2051 pre-2022 to -0.2210 pre-Terra. They take that instability to suggest gas tracks the friction more reliably than native-token Amihud as sample composition changes.
The gas slope itself gets no such subsample test. The regime split presented as a check on the main illiquidity result estimates the aggregate native factor: 0.098 in low-BTC-illiquidity states and 0.060 in high ones. The 30, 60 and 90-day rolling windows also estimate only the native factor. Elsewhere, the cross-chain ratios are 2.4x and 4.5x in one passage, then 3.5x and 8x in another. Both sets are correct: the former uses standardized effects, the latter raw coefficients. The reader has to supply that distinction.
On Ethereum, adding gas raises the native coefficient from 0.1736 to 0.2002; on Solana, it rises from 0.1937 to 0.2603. Once the controls are partialled out, those increases imply a negative correlation between gas and native illiquidity. Section 2 had proposed native illiquidity as a proxy for the same congestion, with added speculative noise, and predicted that gas would absorb some native transmission. The authors instead call the two measures "related but distinct dimensions of stress". That describes the estimates while leaving the movement against the proposed noisier proxy unexplained.
No measured return edge
Native-token returns show no co-movement with token returns. The native return factor is -0.0042 at baseline and -0.0135 with BTC; no platform term is significant. Native volatility, at 0.0858, loses significance after BTC enters (0.1211). The authors view these nulls as consistent with a frictions channel. For a trader, the paper supplies no return edge. The liquidity relationship might help size quotes or reduce inventory, though it comes with no measured cost saving.
We could not test the gas result ourselves. We have no historical gas-fee series for Ethereum, BNB Chain or Solana, and our crypto universe lacks the token-to-chain mappings needed to reconstruct a 4,224-token panel.
A rolling out-of-sample fit of next-day illiquidity on lagged gas, measured in native units and including date fixed effects, would change my view. If BNB Chain's slope remains near 0.29 there, gas belongs on the liquidity dashboard.