Posts / #POST-236547

If you were building a pair-trading universe for crypto from scratch, which venues, instruments, and quote currency would you anchor it to?

May 24, 2026 · 01:10

R/CRYPTOMARKETS

↗ https://reddit.com/r/CryptoMarkets/comments/1tlxrtz/if_you_were_building_a_pairtrading_universe_for/

\*\*TL;DR\*\*

I’ve built a statistical-arbitrage scanner that runs against roughly 1250 large-and-mid--cap US equities — the full rig: Engle-Granger cointegration, Ornstein-Uhlenbeck mean-reversion fits, half-life and Hurst filters, plus those frozen exit plans we lock in at entry. It works on equities because shorting is cheap, the universe is clean, and the relationships behave like dogs on a leash — they wander but they come back.

Extending the same engine into crypto has delivered the same quiet revelation every honest quant eventually meets: the universe the model prices and the one a real account can actually go both long \*and\* short in are two different animals entirely.

Before I publish any “Top 100” crypto pairs list, I thought I’d ask the people who actually trade this stuff for a living: what’s the right venue + instrument + denominator stack to build a repeatable edge around?

\*\*What the numbers are showing\*\*

We’re sitting on roughly 1,600 cointegrated candidate pairs pulled from spot data. About 900 of them are clearing the eligibility gates right now — Bond Strength, Hurst, half-life, p-value — all the usual filters.

If anything, the mean-reversion statistics look cleaner than they do on US equities: bigger residuals, faster cycles, half-lives often landing in that comfortable 2-3 week window instead of the 4-8 we see in equities. Signal density is high. The execution path, however, is where the probability surface starts to bend in ways the back-test never quite warned you about.

\*\*Where the model and reality quietly diverge\*\*

A proper pair trade needs a clean, reliable short on the relative outperformer. For most altcoins, that published “USD price” you see on the chart is not really a USD price — it’s the USDT book multiplied by whatever the prevailing USDT/USD rate happens to be. Below the top twenty names, actual USD spot volume is somewhere between one and five percent of the USDT volume. Below the top two hundred, the USD book is essentially theoretical.

That leaves the executable universe forking into three practical tiers:

Some of the highest-ranked statistical pairs I’m seeing sit squarely in tier three. Which is the honest way of saying the strategy works beautifully — on paper.

\*\*The question I keep coming back to\*\*

If you were designing a published “top N” crypto pair-trading universe — the way a US equity quant would calmly publish a top-250 list — how would you actually scope it?

A few sub-questions I’d value real-operator views on:

1. \*\*Denominator.\*\* USDT is clearly the unit of account for something like ninety percent of global crypto volume, yet it remains a private-company IOU with a modest history of partial depegs. Do you build the entire universe USDT-quoted and treat USDT/USD as its own separate risk factor, or do you split into a tight USD tier and a wider USDT tier?
2. \*\*Instrument.\*\* Spot pairs or perpetual futures? Perps solve the shorting problem cleanly — no inventory, no locate, funding is simply the cost — but that funding rate is live, dynamic, and perfectly capable of flipping sign mid-trade. Does it make sense to publish a pair signal whose true “borrow cost” remains unknown at the moment of entry?
3. \*\*Venue cut-off.\*\* Do you insist both legs have a liquid perpetual listing on at least one major venue (Binance, Bybit, OKX, Hyperliquid, dYdX), or do you accept spot-margin borrow as a fallback for names that only clear one side? My instinct leans toward the stricter rule — anything that cannot be reliably shorted gets a quiet “not retail-shortable” badge and drops out — but I’m genuinely interested in the counter-argument.
4. \*\*Jurisdiction.\*\* US-accessible venues (Coinbase, Kraken, Hyperliquid, dYdX, GMX) versus the rest of the world (add Binance, Bybit, OKX, Bitget). Two separate products, or one product with a venue tag per pair?
5. \*\*Top 10 / Top 100 framing.\*\* On equities we publish a top-250 because that is roughly the cohort where cointegration holds and execution costs are uniformly cheap. Crypto feels chunkier: the top twenty majors behave like one big BTC-beta asset class, the fifty-to-one-hundred-fifty alt-L1s, L2s and DeFi names carve out their own sector cohorts, and the long tail starts to look a lot like gambling. Does a single “Top 100” still make sense, or are we actually looking at two or three category-specific lists?

\*\*Where I’m leaning at the moment\*\*

Two coverage tiers, labelled with complete honesty:

\- A \*\*USD-quoted tier\*\* of roughly twenty-five to forty tokens, built around what a US retail account can actually execute cleanly on Coinbase or Kraken, with optional long-only or inverse-substitution framing.

\- A \*\*USDT/perp-quoted tier\*\* built around tokens that carry a liquid perpetual listing on at least one of the major venues, with both clean spread P&L \*and\* funding-adjusted P&L shown side by side.

I keep circling back on whether to publish anything at all for the long-tail, spot-borrow-only tier. The statistical relationships are genuinely interesting; the execution realities are genuinely brutal.

\*\*Deeper plumbing available\*\*

If anyone wants the longer version — Tether redemption mechanics, depeg history, perpetual funding arithmetic, US versus non-US friction stack laid out side by side — I wrote a more detailed piece on it. Happy to drop the link in the comments rather than clutter the body.

\---

\*\*Question to the people actually running systematic strategies in crypto right now:\*\* what venue + instrument + denominator combination did you ultimately settle on, and what do you wish you’d known about the funding-rate cost before you went live?