Estimated trading-cost parameters for crypto pairs, one row per pair and year, from 2013 on: 3,527 rows over 989 pairs (measured 2026-11-09). Coverage is thin before 2017 (two to four pairs a year) and widest in 2023 (738 pairs).

Columns

symbol: the pair, base then quote (BTCUSD, ETHEUR, ETHBTC). Cryptocurrencies-Daily-Price keys a coin by its bare symbol quoted in USD, so BTC there is BTCUSD here. 660 of the pairs are quoted in USD.
year: calendar year the parameters were estimated over.
price: the day's volume-weighted price, averaged over the sampled days.
spread_bps: the Roll (1984) effective-spread estimate from the day's trades, in basis points, averaged over the sampled days. A round trip crosses half of it on each side.
lambda_bps: Kyle's lambda, the price impact in basis points per million of the quote currency (dollars for BTCUSD, ether for XYZETH), from the 5-minute signed trade flow pooled over the sampled days. Null on 338 rows, where no slope was estimated for that pair and year.

How it was estimated

The source publishes crypto trades but no crypto quotes, so nothing here is a quoted spread: the spread is inferred from the serial covariance of trade prices. The whole UTC day is read -- crypto has no quiet session -- on a sample of days in each month, not every day.

BTCUSD reads 4.65 bp in 2024 and 3.20 bp in 2025, ETHUSD 3.65 and 2.66; earlier years read wider (BTCUSD 49.8 bp in 2019, 10.1 bp in 2020).

Caveats

An estimate, not a measurement: Roll is noisy on a pair with few trades. 903 rows read above 100 bp, 81 above 1,000 bp and 7 above 10,000 bp; treat a thin pair's figure as "expensive", not as a price.
Sampled days: this gives the level of a pair's cost by year, not its day-to-day variation.
Exchange fees are not included.
Rows with no estimable spread, or a zero spread, are not published: a lookup that misses is uncovered, not free.

Downloads last month: 337

Total file size:

108 kB