FLG/FNB regional banks: the spread just collapsed 3.7zFLG-0.6330*FNBBATS:FLG-0.6330*BATS:FNBRORO_Labs The trade. We are long the spread on Flagstar Financial / F.N.B. Corporation. In plain terms: **buy FLG, sell FNB** — $10,000 of FLG (727 shares) against $8,668 of FNB (460 shares) on the hedge ratio of 1 : 0.63. This is not a bet on either bank going up or down on its own. It is a bet that the *gap* between them — which has just been stretched violently wide — closes back toward its normal level. Because one leg is bought and one is sold, the position is roughly market-neutral: the broad market, and the regional bank complex with it, can rise or fall together and the trade mostly cares about the relative move between these two names. Both are Banks – Regional, same sector, same rate cycle, same deposit and credit forces. The position is **1 day old**. The frozen trade rules. These were locked by our tracker the moment the signal fired and do not change mid-trade: - **Entry:** −3.66σ — the stretch at which the pair fired. FLG had fallen far enough against FNB to be a genuine outlier. - **Target:** exit at **0.22σ** — essentially all the way back to the average, about **3.9σ of travel** from here. - **Stop:** **±5.29σ** — the line where the idea is declared wrong. - **Timeout / max hold:** a maximum-hold rule closes the trade if it overstays its welcome, regardless of where the spread sits. Where it stands. The tracker reads **−3.6801σ** against an entry of −3.66σ — the spread has drifted a fraction further against the position in its first session, which is noise, not a story. Target sits at 0.2238σ, so the full journey is roughly 3.90σ of mean-reversion. The stop band is ±5.29σ, leaving about **1.6σ of room** before the idea is wrong. That is a wide leash by design: this pair's spread is volatile, and the frozen rules were solved to match its behaviour rather than a generic template. Our stack's latest general read on the pair is −3.6562σ — a whisker away from the frozen tracker number, and pointing the same direction. Unrealised P&L is flat at day one; the day's mark is +$26. 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 trades under these rules, not as a promise: - **Average return:** +2.53% per trade - **Average days in position:** 15.8 - **Max drawdown:** 5.2% - **Record:** 4 wins from 4 trades — a small sample, and we say so — Sharpe 1.98 The four optimisation trades: Nov 5 → Dec 1 (+3.34%, 17 days), Feb 13 → Feb 19 (+3.08%, 3 days), Mar 25 → May 15 (+1.82%, 36 days), Jun 15 → Jun 25 (+1.89%, 7 days). Note the spread of holding periods — three days at the fastest, thirty-six at the slowest. At **1 day in against a 15.8-day average**, this position has barely started. The returns are modest per trade; the appeal is the consistency and the shortish holds, not the size of any single win. How clean is the spread?** From our research stack's pair diagnostics: - **Hurst 0.24** — the rubber-band score. Below 0.5 means stretches tend to snap back rather than run; 0.24 is firmly in rubber-band territory. - **Half-life 24.9 days** — when this spread stretches, it typically takes about five trading weeks to fade halfway home. Patience is part of the trade, and that squares with the 15.8-day in-sample average hold only if reversion runs faster than typical. - **Bond 94.7%** — how tightly the two names are economically tethered. High, as you would expect from two regional banks. - **Hedge ratio 1 : 0.63** — the dollar balance that makes the pair market-neutral, and the reason leg two is $8,668 rather than $10,000. - **Spread volatility 0.300** — the pair's normal daily wobble, and the reason a ±5.29σ stop is not as loose as it first looks. - **Score 0.677, eligible** — a relative ranking of signal quality in the latest run. It ranks; it does not predict. The chart. The last bar is the whole story: the spread fell off a cliff, a single session dropping from around 110 to the low 80s on the ratio, with the on-chart z-line plunging from roughly +0.5 to −2.79. That is not a slow drift — it is a step change, and step changes are exactly what a mean-reversion trade is designed to lean against. The caveat is honest: a gap that large can be a repricing rather than a stretch, and the frozen stop exists precisely because we cannot tell the difference in advance. The second opinion: POSI v3.** The pane on this chart is a deliberately lightweight, one-size-fits-all solver, learning its own moving-average window (31 bars, auto) on the fixed ratio. It classifies the series as mean-reverting — "the gap snaps back" — and calls it tradeable. Its stretch reads −2.8σ at the **97th percentile**, meaning today's gap is more unusual than 97% of the history it has seen. Room to line 35.89% with an expected ~7 bars back to the average, reward-to-risk 2.6:1, and an 83% probability of hitting target before stop — the script's own odds, on its own model. Stability: stable. Reason: optimal reversion trade. Two gaps. First, the pane z of −2.79 is not the tracker's −3.68σ; the pane is a generic on-chart solver, while the tracker runs the pair-specific terms frozen when the signal fired. Different models, same direction. Second, the pane's suggested terms (`L&L |E/S/T|` 1.7 / 3.0 / 1.8) sit well inside our frozen −3.66 entry and ±5.29σ stop. That is expected: the pane would have entered earlier and stopped out sooner, because it applies one template to every series; our optimiser ran per-pair discovery on this spread's own history and froze wider, slower terms to match its volatility. The difference is **scope, not quality**. Verdict: the second opinion agrees — stretched, mean-reverting, and worth taking. Chart it yourself. The spread ratio is `FLG-0.6330*FNB` on a daily. One caveat: hedge ratios drift. A ratio solved months ago goes stale as the two businesses diverge — static spread symbols need periodic re-solving. How this strategy works. Our research stack scans thousands of pairs of economically related stocks — 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 (entry, target, stop, timeout) on the pair's own history, scores the result to rank the best opportunities, and tracks the strongest signals live — freezing the rules the moment the signal fires and monitoring the position through to its exit. The core idea: 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.