Skip to content
StockMarketAgent
§ Tool · Macro analogs

Historical analog explorer — building 5-year history

The historical analog explorer is building its first five-year history. Actual collected coverage is 2024-05 to 2026-05; first analog match available 2027-05.

§ What this is warning you about first

The analog result is intentionally withheld until the dataset reaches the minimum history threshold. A premature nearest-neighbour claim would look precise while resting on too little regime history. The table below is blank by design — a placeholder forward-return is worse than no forward-return.

Building
Status · no analogs published
2024-05 → 2026-05
Coverage · collected history
24 / 36
Snapshots · minimum before matching
2027-05
First match · expected threshold month
§ Regime-snapshot ledger

24 of 36 monthly snapshots collected

67%
Collected2024-05 → 2026-05
Remaining12 months
First analog match2027-05
§ The feature vector

Five standardised macro dimensions

Each monthly snapshot is a 5-D vector. Every dimension is z-scored or percentile-normalized so it is comparable across decades — a 200bps yield curve in 1995 and the same curve in 2025 do not mean the same thing relatively, and standardisation fixes that.

01
Curve slope
10Y − 2Y Treasury spread, normalized
02
Breadth percentile
% of S&P 500 above 200d MA, 10y percentile
03
Valuation percentile
Forward P/E vs 10y history
04
Credit spreads
IG & HY OAS, normalized
05
Regime score
Composite leading/lagging indicator

Matches are ranked by cosine similarity, not Euclidean distance — similarity measures the shape of the vector (the direction it points in 5-D space), not its magnitude. Two periods can sit at very different absolute levels yet share a similar pattern of relative tilts; that pattern is the signal.

§ Forward outcomes (available once history is collected)

Three nearest analogs — and what happened next

Each published row will surface the analog window, a one-line label, the cosine-similarity score (0–100), the S&P 500 total return over the following 12 months, the max drawdown along the way, and a note on what was distinctive.

⌗
Forward-outcome rows appear once 36 regime snapshots are collected

Currently 24 of 36. First analog match available 2027-05. The table is intentionally blank rather than back-filled with synthetic rows — a placeholder forward-return is worse than no forward-return.

PeriodLabelSimilarityFwd 12mMax DD
§ Illustrative shape · not a published match

What an honest analog surface looks like

Sample rows for the design only — gated out of the live surface above. The point: the closest matches do not all point the same way. One bear precursor, one mid-cycle scare-then-recover, one melt-up. The dispersion is the information a single-best-match framing would erase.

illustrative — not published
PeriodLabelSimilarityFwd 12mMax DDNote
2007 Q3 → 2008 Q3Late-cycle credit deterioration84−18.2%−34.1%Bear-led curve steepening; breadth collapsed before the headline index did.
2015 Q3 → 2016 Q3China growth-scare correction and recovery76+2.4%−12.8%Late-cycle scare without recession; sharp rotation, no permanent damage.
1998 Q4 → 1999 Q4Pre-mania melt-up69+16.4%−7.6%Narrow leadership broadened into a final-stage melt-up before the 2000 unwind.
§ Notes

No historical analog rows are published yet. The dataset temporal coverage is 2024-05/2026-05, and the first analog match becomes available 2027-05. The conditional distribution from a similarity-weighted sample is the honest input; a single point estimate would erase the dispersion. This is index-level (S&P 500), monthly cadence — once live it joins the sector heatmap and breadth tracker as the third lens of the platform's monthly macro snapshot.

All tools →Read methodologycreativeWorkStatus = InProgress · refresh = monthly
§ FAQ

Five things worth knowing

Q01Why is there no analog match yet?+
The tool is deliberately in a building-history state. It collects one regime snapshot per month and needs a minimum of 36 (three years) before it will publish anything. With 24 collected, it is 12 months short — the first analog match becomes available 2027-05. The result is intentionally withheld until the dataset reaches the threshold, because a premature nearest-neighbour claim would look precise while resting on too little regime history.
Q02Why 36 snapshots, not ship now with what you have?+
A nearest-neighbour analysis from a short ledger is too easy to over-cite. With only ~20 observations, similarity scores look high simply because the search space is small, and the "closest analog" might be eight months old — almost the same point in time, not a meaningful historical comparison. A user shown an "85% similar" match would naturally weight it heavily even though it came from a tiny haystack. 36 monthly snapshots is the calibration floor where finding a genuinely similar setup is informative rather than coincidental.
Q03Why not back-fill 30 years of history to launch faster?+
The regime-snapshot ledger is collected forward in time at a specific monthly cadence using specific data sources. Back-filling would require reconstructing decades of macro feature vectors from imperfect historical data, which introduces measurement noise that is hard to bound. The forward-collection-only approach trades launch speed for credibility — the numbers you eventually see are all measured the same way.
Q04When it launches, why three matches instead of the single closest one?+
Showing the single nearest neighbour creates false precision. Three matches form a conditional distribution — the range of outcomes that historically followed similar setups. The illustrative sample is deliberately diverse (−18% / +2% / +16%): when highly-similar setups produced very different outcomes, the prior on the current setup is genuinely uncertain, which is more informative than a single best-match point estimate. The dispersion is the information.
Q05Will this predict the market?+
No. Even after launch, "the closest analog returned −18%" is a calibration of the prior, not a forecast. Past performance does not constrain the future, but it does calibrate the prior. It is also index-level (S&P 500), not stock-specific, and refreshes monthly — a context tool for monthly allocation reviews alongside the sector heatmap and breadth tracker, not an intraday signal.