xStocks Spread Dispersion.

1. Executive summary

Tokenized equities (xStocks) on Solana do not have one liquidity problem — they have a dispersion problem. The deep majors already trade tight to NBBO fair value (NVDAx 12 bps, TSLAx 13 bps in market hours). The long tail and single-name ETFs do not (MSFTx 46 bps, COINx 32 bps, HOODx 31 bps). And across every name, the after-hours window is taxed: the TSLAx that costs 13 bps in market hours costs 71 bps after the close; HOODx jumps from 31 to 155 bps.

That dispersion is the wedge. Where a market is already tight (NVDAx, TSLAx in market hours), a sponsored market adds little. Where it is wide — long-tail names, single-stock ETFs, and all names after hours — a reward market that pays makers for measured execution can close a gap that demonstrably exists today.

Findings span 13 pairs, 243,280 swaps, $103,903,507 of on-chain volume over the trailing 30-day window (2026-04-14 → 2026-05-13). This is a complete on-chain capture — no sampling, no extrapolation. Every dollar figure is the literal sum of observed fills.

2. Headline findings

The picture is not "xStocks are expensive." It is that execution quality is spread across a wide range — by name and by session. The table reports every pair; spreads are the median (p50) and 90th-percentile (p90) deviation from NBBO fair value, in basis points.

All spreads in bps vs NBBO. Swap counts and volume are the full on-chain totals; spread percentiles are computed on fills with a fresh NBBO reference (see Methodology A.5). GLDx (Backed) wraps the GLD ETF, so its reference prices only in US market hours like the equities — its session split is measured on the same basis, which makes it a clean gold comparison alongside the single-name stocks.

3. Per-pair analysis

Each pair's section shows the full distribution against NBBO, broken out by session (market hours vs. after-hours), with three companion charts: the spread histogram, staleness vs. spread, and the intraday heatmap.

3.1 TSLAx/USDC

79,702 swaps · $26,571,331 volume · 2026-04-14 → 2026-05-13

Metric	Value
Market-hours p50	13.1 bps
Market-hours p90	51.4 bps
After-hours p50	71.1 bps
After-hours p90	284.5 bps
Fills measured (mkt / a-h)	29,834 / 15,637

3.2 NVDAx/USDC

40,557 swaps · $17,118,465 volume · 2026-04-14 → 2026-05-13

Metric	Value
Market-hours p50	11.6 bps
Market-hours p90	45.4 bps
After-hours p50	42.3 bps
After-hours p90	119.0 bps
Fills measured (mkt / a-h)	11,118 / 5,767

3.3 AAPLx/USDC

2,784 swaps · $941,097 volume · 2026-04-14 → 2026-05-13

Metric	Value
Market-hours p50	24.0 bps
Market-hours p90	69.8 bps
After-hours p50	62.6 bps
After-hours p90	294.1 bps
Fills measured (mkt / a-h)	879 / 872

4. SOL/USDC — crypto-native baseline & methodology validation

This section replicates the Jump Crypto (March 2026) propAMM spread analysis using the same on-chain data methodology, establishing that Qualiq's analytical approach is sound and that sub-2-bps execution is demonstrably achievable on Solana for crypto-native pairs.

Interpretation

Raydium CLMM SOL/USDC shows somewhat wider spreads than Jump Crypto's propAMM finding because CLMM is a passive AMM — it prices via inventory risk, not competitive quoting. PropAMMs (programmable AMMs with intent-based quoting) achieve tighter spreads by attracting competitive market-maker flow. The point for tokenized equities is the same: where execution is wide, competition among makers — rather than a single passive curve — is what compresses it. That is the gap a sponsored reward market is built to close.

A. Appendix — data sources & methodology

A.1 Data sources

A.2 Swap discovery methodology

1. Pool-based, not mint-based

   Token mint accounts receive 5,000–15,000 signatures/day (mints, burns, transfers, all pools). Pool state accounts receive only that pool's swap/LP transactions — 10–100× fewer, targeted to the specific liquidity we care about.

2. Complete capture, no sampling

   getSignaturesForAddress(pool_id, limit=1000) pages backward through the full history in the date range. Every transaction is decoded — there is no 1-in-N sampling and no extrapolation. The 243,280 swaps and $103.9M of volume are literal on-chain sums over the 30-day window.

3. Batch transaction fetch

   JSON-RPC batches of getTransaction calls per HTTP request, with a single-call fallback on any malformed batch response so no fill is silently dropped.

4. Swap detection

   Fee-payer's preTokenBalances and postTokenBalances must show equal and opposite changes in base- and quote-mint balances.

A.3 Fair-value methodology

• Equity pairs (NBBO proxy) — ASOF backward join to the most recent 5-min bar. market_open is computed from the swap's own UTC timestamp using NYSE trading hours (09:30–16:00 ET), not from the yfinance bar (which only contains market-hours data).
• Crypto pairs (CEX mid) — ASOF backward join to the most recent 1-min Binance bar. Always-open (no market-hours concept).

A.4 Spread formula

A.5 Reference freshness & limitations

• Window. All figures cover the trailing 30 days, 2026-04-14 → 2026-05-13, ending at the latest observed on-chain block across the pair set.
• Reference-staleness gate. A swap's spread is only meaningful against a fresh fair-value reference. yfinance publishes 5-min bars during the regular session only, so a swap that lands in a coverage gap or halt would otherwise be scored against an ancient mid and produce a spurious spread. We therefore compute spread percentiles only on fills whose matched reference is fresh enough: < 6 minutes (one bar) in market hours, and ≤ 1 day after hours (where the mid is the last regular-session close by design). Swap counts and volume use the full on-chain set; only the percentiles use this measured subset. Fig 3.x.b (staleness vs spread) shows the effect directly — spread is a clean function of reference age, which is exactly what the gate removes.
• After-hours fair value. With no 5-min bar after 16:00 ET, the mid for after-hours swaps is the last NYSE close. The after-hours numbers should be read as "vs last regular-session close" — a real and consistent off-hours premium, but a noisier reference than the intraday one.
• NBBO proxy. yfinance provides the bar close as mid, not a true NBBO bid/ask. A true NBBO requires a paid market-data tier. Close-as-mid slightly underestimates spreads in thin markets.
• Long-tail robustness. All 13 pairs carry thousands of fills each over the window, so the long-tail names (MSFTx, METAx, AMZNx, GOOGLx) are more robust than in earlier sampled work — though still thinner than the deep majors, and their after-hours p90s in particular should be read as directional.