USDC / fiat liquidity

Liquidity is measured against the fiat FX rate (like the liquidity history), so a USDC premium or discount shows as less depth near FX; spread isolates the book itself. Spread is the hourly mean in ticks of 0.0001 EUR over the minutes with an uncrossed book: most venues trade at one such tick, so it shows how often and how far a venue's spread widens (OKX, Bitstamp and Cex quote 0.00001, so theirs can be a fraction of a tick). Until 2026-10-10 prices were stored rounded to 4 decimals, which merged OKX, Bitstamp and Cex levels and made their books look locked; their earlier depth and cost figures are approximate. Depth is EUR notional within the band; cost is the average fill for the size vs FX (n/a for an hour when the stored top 20 levels could not fill it in most samples). Hourly values are medians of once-a-minute snapshots; hours with fewer than 40 samples (recording gaps) and weekends (no FX rate) are left out. Volume covers only venues with recorded trades, so Niza is not part of this analysis; Binance and Bullish trades are recorded from 10 Oct 2026 on. Correlation is not causation; the adjusted ρ removes the typical level for each hour of day. With BTC/USDC volume, every venue's liquidity is compared with the same market-wide BTC/USDC volume (hourly candles of a fixed set of venues with history over the whole range, in USDC at the typical price); hours where one of these venues has no candle are left out.

For a trade with taker sign s (+1 buy, −1 sell) at price p: effective half-spread e = s·(p − m₀)/m₀, adverse selection a(h) = s·(m(t+h) − m₀)/m₀, realized half-spread r(h) = e − a(h), the part of the spread the maker of the trade keeps after h. The market mid m is the per-second median of the venue mids (uncrossed books, at least 3 venues) from the stored 1 s top-20 snapshots; m₀ is the mid of the second before the trade, because trade times are whole seconds and the trade may already be in that second's book. So the first horizons (1–5 s) are coarse. Mids are carried back up to 5 s over recording gaps; trades without a mid before them are left out (coverage), and trades more than 50 bp from the mid are treated as bad prints. e is measured against the market mid, not the venue's own mid: a venue that trades persistently away from the other venues (e.g. a negative e) shows its price offset here, not a negative spread. Coinbase reports the maker side and is flipped; Niza has no trades; bit2me (mirror of Kraken) is excluded.