Network Fee Volatility, Measured

Fees are not a price; they are an auction outcome. Their variability is a property worth quantifying rather than describing.

The timescales

Within a block. The base fee adjusts by up to a fixed percentage per block, so short-term moves are bounded and rapid.

Within a day. The largest predictable variation. Peak to trough within a single day commonly spans a factor of two or three on mainnet.

Within a week. The weekly cycle adds to the daily one. Weekend lows against weekday peaks commonly span a factor of three to five.

Within a month. Dominated by events rather than by cycles. A single large mint or a liquidation cascade produces spikes far above anything the cycles generate.

The distribution shape

Fee data is not normally distributed. It has a long right tail: most observations cluster at a modest level, with occasional excursions many times higher.

That shape has a practical consequence. The average is a poor summary, because it is pulled upward by spikes you can avoid simply by not transacting during them.

The median describes what you will typically pay. The ninetieth percentile describes what you will pay if you transact without checking.

Measuring it yourself

Sample the base fee at fixed times daily for a month. Three samples a day is sufficient to see the daily pattern; more is better for the tail.

Convert each to currency at the rate of that moment, not at a single later rate, or the chart shows the asset price rather than the fee.

Report the median and the ninetieth percentile rather than the mean.

What the numbers imply

For scheduled activity, always transact at the weekly low. Consolidations, revocations and transfers between your own wallets are never urgent, and the saving is a multiple rather than a percentage.

For unscheduled activity, check before committing. The difference between transacting during a spike and waiting an hour is frequently larger than any other decision available to you.

For anything small, use a rollup. Fee volatility on mainnet makes small transactions uneconomic at the wrong moment, and the variation is largely absent on rollups.

The forecasting question

Short-horizon fee prediction works reasonably, because the base fee for the next block is determined by the current one.

Beyond a few blocks it is a demand forecast, which nobody does well. Gas trackers that project hours ahead are extrapolating recent conditions rather than predicting.

Use them for the next few minutes and ignore them beyond that.

The venue side

Withdrawal charges are set by venues and change less frequently than network fees. A platform passing through the network cost will show variation; one charging a fixed amount will not.

Comparing a venue’s charge against the current network fee, published by explorers and by venues such as venues that quote the network fee separately, is how you determine which model a platform uses and what margin it carries.