on an earnings print that was mostly non-cash...

**The trade.** We are short the spread on AMZN/NVDA. In plain terms: short Amazon, long NVIDIA — $10,000 of AMZN against $4,000 of NVDA on the solved hedge ratio of 1 : 0.54. This is not a call on either company going up or down on its own. It is a bet that the *gap* between the two, which has just stretched to an unusual extreme, closes back toward its normal level. One leg long and one leg short makes the position roughly market-neutral: the broad market can rally or sell off, and the trade mainly cares about the relative move between the two names. The signal is fresh — it passed every entry filter in the run stamped 2026-07-31.

**Why the gap opened.** Amazon closed the week up 15.32% at $271.58 after reporting Q2 net income of $62.6B, or $5.75 a share against roughly $1.81 expected. Inside that print sits a $53.4B non-operating, non-cash gain from revaluing its Anthropic stake. Strip it out and normalised earnings land nearer $1.60 — below the consensus the headline appeared to crush — while trailing twelve-month free cash flow reads –$7.6B after capex rose $66.1B year on year. NVIDIA, meanwhile, closed at $200.75 (+2.93%) and does not report until 26 August. One name repriced violently on a single number; the other did not move much at all. That is exactly the kind of one-sided repricing a spread trade is built to fade — with the honest caveat that a market willing to pay for a headline once can pay for it again.

**The frozen trade rules.** These are locked by our tracker at the moment the signal fires and do not change mid-trade:

- **Entry** — 3.11σ. That is how far the spread was stretched above its own average when the pair fired.
- **Target** — back to the average, roughly 3σ of travel from here. The exit level is the tracker's, not a level eyeballed off the chart.
- **Stop** — a frozen σ band beyond entry, set per pair. It is the line where the idea is declared wrong, and it is fixed before the trade starts.
- **Max hold** — a timeout closes the position if it overstays, regardless of where the spread sits.

Stop and timeout figures are not quoted here because they are not on the material in front of me, and I would rather leave a number out than make one up.

**Optimisation history (in-sample).** Before this signal fired, our optimiser tested these exact rules against this pair's own history. That is in-sample — the rules were fitted on this data — so treat it as indicative of how the pair behaves under these rules, not as a promise of anything:

- **Trades:** four, four wins. A small sample, and we say so.
- **Returns:** +5.46%, +6.35%, +3.66%, +5.90% — an average near +5.3% per trade.
- **Holds:** 23 days, 4 days, 35 days, 8 days — an average around 17.5 days.
- **Max drawdown / Sharpe:** not shown on the record supplied, so not quoted.

The individual runs: Aug 1 → Sep 4 (+5.46%, 23d), Oct 29 → Nov 4 (+6.35%, 4d), Feb 6 → Mar 30 (+3.66%, 35d), May 1 → May 13 (+5.90%, 8d). The spread of holding periods matters more than the win column — four days and thirty-five days are both normal outcomes here.

**How clean is the spread?** The diagnostics from our research stack are unusually consistent for once:

- **Hurst 0.17** — the rubber-band score. Below 0.5 means stretches tend to snap back, and 0.17 is emphatic.
- **ADF p 0.046** (stat –2.90) — the formal snap-back test clears the 5% bar. Hurst and ADF agreeing is not the norm; here they do.
- **Half-life 32 days** — when this spread stretches, it typically takes about a month to fade halfway home. Patience is part of the trade, and it lines up with the 35-day outlier in the optimisation history.
- **Skew 0.03, excess kurtosis 0.03, JB p 0.986** — the spread's daily moves are close to a textbook bell curve, with no fat-tail bulge. That matters, because it means a 3σ reading is genuinely a 3σ event rather than a statistic flattered by a lopsided distribution.
- **Pearson r 0.31** — day-to-day the two names only loosely track each other. The relationship shows up over weeks, not sessions.
- **Bond 100%, gravity 0.022, score 0.899** — the pair's structural link and ranking quality in the latest run.

**The chart.** The lower pane is the spread itself, AMZN−0.5435×NVDA, with a 60-day mean and its bands. It closed at 162.47, up 25.47% on the session, against a 60-day average of 136.83 and an upper 2σ band at 157.74 — outside both band sets, and effectively all of that move arrived in one candle. Through July the spread had been drifting around and below its mean, including a dip toward 120 late in the month. So this is not a slow grind wider that has been fighting a position for weeks; it is a single-session dislocation. Our stack reads the stretch at 3.11σ using the window it solved for this pair, while the chart's bands use a fixed 60-day setting on a fixed ratio — different windows, so different numbers. Both say the same thing in plain words: this is far from normal.

**Second opinion.** There is no POSI v3 pane on this chart — the on-chart read here is the plain 60-day band structure, and it agrees with the tracker: the spread is trading outside its 2σ envelope after a one-day move, on a series whose own statistics say stretches fade. Agreement is not confirmation. It just means the on-chart view and the solved view are not arguing.

**Chart it yourself.** The spread ratio is AMZN-0.5435*NVDA — paste it as a symbol and you get the yellow line above. One caveat worth knowing: hedge ratios drift. A ratio solved in winter is stale by summer, so static spread symbols need periodic re-solving, and the bands you see are only as good as the ratio underneath them.

**How this strategy works.** Our research stack scans thousands of pairs of economically related companies — same sector, same customers, same economic forces. For each candidate it tests whether the spread between them genuinely mean-reverts, optimises per-pair trade rules on the pair's own history, scores the result to rank the best opportunities, and tracks the strongest signals live — freezing entry, target, stop and timeout at the moment the signal fires and monitoring the position to its exit. The core idea is narrow on purpose: the two companies are economically related, and their co-movement is statistically proven at the point in time of the signal, not assumed forever. Relationships drift, which is why everything is re-tested and re-solved run after run.

Not financial advice. Do your own research.