Methodology & provenance
Data dictionary
| Field | Definition |
|---|---|
| main_street | 0–100 consumer demand: measured purchases, app downloads, web traffic, and brand engagement, aggregated per company |
| wall_street | 0–100 analyst sentiment and institutional positioning |
| likefolio_score | 0–100 composite of demand and sentiment |
| divergence_gap / percentile | main_street − wall_street, and its cross-sectional percentile; extremes (≥ 90th) are logged as dated events with forward outcomes |
| estimated | per-row provenance flag: true = model-backfilled rather than observed-at-the-time |
Brands map to parent tickers, so a product trend reads as
its listed company. Coverage: 3,976 names (US-listed plus ADR/OTC), rescored every trading day.
Machine-readable: GET /v1/meta.
History provenance — three tiers, plainly labeled
| Tier | What it is | Sold as |
|---|---|---|
| As-published vintages | Nightly append-only freeze of the published scores (started July 2026; currently 91 vintage days). SHA-256 batch manifests held in external object storage; recomputable on demand. | Point-in-time |
| Reconstructed history | 2016–2026 series generated under
current-model methodology, with per-row estimated flags and
admin-edit exclusions documented. | Reconstructed history — never "point-in-time" |
| Recovered vintages | As-delivered artifacts from historical licensee files and partner archives, matched against the reconstructed series where available. | Verified against archival snapshots |
No-restatement policy: once a (date, ticker) row is frozen in the vintage store, it is never updated or deleted. There is no restatement path in the writer, and the external checksum manifests make any tampering detectable. Verification output is available in diligence.
Signal fact sheets
Main Street
Main Street Score — Signal Fact Sheet
Generated 2026-07-05 from 1,170,301 ticker-day observations, 595 tickers, 2016-01-25 → 2026-07-03. Estimated-provenance rows in panel: 0.0%.
Predictive power (rank IC vs forward returns)
| Horizon | Mean IC | t-stat* | % days positive | N days |
|---|---|---|---|---|
| 1d | +0.0183 | +7.33 | 59.3% | 1,735 |
| 5d | +0.0357 | +6.30 | 64.9% | 1,747 |
| 21d | +0.0717 | +5.72 | 76.8% | 1,731 |
| 63d | +0.1295 | +4.64 | 90.8% | 1,689 |
| 90d | +0.1614 | +4.46 | 92.8% | 1,662 |
*t-stats on non-overlapping day subsamples (overlap-robust).
Consistency by year (21d IC)
| Year | 21d IC |
|---|---|
| 2019 | +0.0235 |
| 2020 | -0.0485 |
| 2021 | +0.0981 |
| 2022 | +0.0777 |
| 2023 | +0.1094 |
| 2024 | +0.1244 |
| 2025 | +0.1082 |
| 2026 | +0.0324 |
Decile portfolios (21-day rebalance, equal weight)
- Average 21d return by decile: D1:-0.24% D2:+0.31% D3:+0.87% D4:+0.89% D5:+0.28% D6:+1.78% D7:+1.85% D8:+2.12% D9:+2.48% D10:+2.38%
- Long-short (D10−D1), annualized: +31.4% gross (Sharpe 1.78), +21.8% net of 20bps/side (Sharpe 1.23)
- Max drawdown (gross): -18.2% · 68.7% of rebalance periods positive · 83 rebalances 2019-09-03 → 2026-06-01
Orthogonality to price factors
- Rank correlation with momentum (12-1): 0.181; short-term reversal (21d): 0.072; realized vol (63d): -0.112
- Residual 21d IC after removing all three factors: 0.053 (t = 5.47, 78.0% days positive) — the signal is not repackaged price action.
Regime robustness (21d IC)
- up market: +0.0740 (74.9% days positive, n=974)
- down market: +0.0688 (79.3% days positive, n=757)
- 2020 covid: -0.0499 (33.6% days positive, n=232)
- 2022 bear: +0.0777 (84.9% days positive, n=251)
- 2023 2025 bull: +0.1140 (89.4% days positive, n=752)
- 2026 ytd: +0.0324 (74.8% days positive, n=139)
Methodology: cross-sectional studies on LikeFolio's covered universe (equal weight, gross of borrow; 20bps/side cost model where noted). History 2016-2026 is reconstructed history under current-model methodology with per-row provenance flags; as-published point-in-time vintages accrue in the append-only score_vintage store from July 2026 forward under a written no-restatement policy. Hypothetical results; past performance does not guarantee future results.
LikeFolio Score
LikeFolio Score — Signal Fact Sheet
Generated 2026-07-05 from 1,170,301 ticker-day observations, 595 tickers, 2016-01-25 → 2026-07-03. Estimated-provenance rows in panel: 0.0%.
Predictive power (rank IC vs forward returns)
| Horizon | Mean IC | t-stat* | % days positive | N days |
|---|---|---|---|---|
| 1d | +0.0149 | +7.57 | 56.9% | 2,643 |
| 5d | +0.0310 | +6.82 | 63.8% | 2,655 |
| 21d | +0.0619 | +6.87 | 77.9% | 2,639 |
| 63d | +0.1086 | +5.82 | 89.1% | 2,597 |
| 90d | +0.1328 | +7.92 | 90.9% | 2,570 |
*t-stats on non-overlapping day subsamples (overlap-robust).
Consistency by year (21d IC)
| Year | 21d IC |
|---|---|
| 2016 | +0.0677 |
| 2017 | +0.0613 |
| 2018 | +0.0057 |
| 2019 | +0.0668 |
| 2020 | -0.0400 |
| 2021 | +0.0971 |
| 2022 | +0.0757 |
| 2023 | +0.0997 |
| 2024 | +0.1072 |
| 2025 | +0.0939 |
| 2026 | +0.0348 |
Decile portfolios (21-day rebalance, equal weight)
- Average 21d return by decile: D1:+0.32% D2:+0.80% D3:+0.94% D4:+0.99% D5:+0.94% D6:+1.17% D7:+1.71% D8:+1.83% D9:+1.99% D10:+2.33%
- Long-short (D10−D1), annualized: +24.1% gross (Sharpe 1.53), +14.5% net of 20bps/side (Sharpe 0.92)
- Max drawdown (gross): -14.6% · 65.9% of rebalance periods positive · 126 rebalances 2016-01-25 → 2026-05-27
Orthogonality to price factors
- Rank correlation with momentum (12-1): 0.137; short-term reversal (21d): 0.012; realized vol (63d): -0.101
- Residual 21d IC after removing all three factors: 0.0477 (t = 6.13, 75.3% days positive) — the signal is not repackaged price action.
Regime robustness (21d IC)
- up market: +0.0583 (75.3% days positive, n=1683)
- down market: +0.0683 (82.3% days positive, n=956)
- 2020 covid: -0.0353 (40.9% days positive, n=232)
- 2022 bear: +0.0757 (84.9% days positive, n=251)
- 2023 2025 bull: +0.1003 (91.0% days positive, n=752)
- 2026 ytd: +0.0348 (74.1% days positive, n=139)
Methodology: cross-sectional studies on LikeFolio's covered universe (equal weight, gross of borrow; 20bps/side cost model where noted). History 2016-2026 is reconstructed history under current-model methodology with per-row provenance flags; as-published point-in-time vintages accrue in the append-only score_vintage store from July 2026 forward under a written no-restatement policy. Hypothetical results; past performance does not guarantee future results.
Divergence Events
Divergence Extremes — Event Study Fact Sheet
Onset events: gap percentile crossing ≥ 0.90 (prior day below). Abnormal return = ticker forward return minus same-date universe median. 4,013 events, 2024-06 → present.
| Direction | Horizon | Mean abnormal | Median | Hit rate | t | N |
|---|---|---|---|---|---|---|
| bull | 21d | +0.64% | +0.08% | 50.3% | +2.39 | 2,872 |
| bull | 63d | +4.70% | +1.08% | 51.6% | +8.37 | 2,761 |
| bull | 90d | +6.14% | +1.95% | 53.3% | +8.91 | 2,670 |
| bear | 21d | +1.32% | -0.82% | 47.5% | +2.52 | 1,022 |
| bear | 63d | +1.33% | -4.01% | 41.3% | +1.31 | 905 |
| bear | 90d | -1.19% | -6.49% | 37.7% | -1.04 | 830 |
Read: bullish divergence onsets (consumer demand far above Street sentiment) show statistically strong positive abnormal returns. Bearish onsets are directionally negative but NOT statistically significant at current sample size — bear-side claims are not marketed.
Methodology: cross-sectional studies on LikeFolio's covered universe (equal weight, gross of borrow; 20bps/side cost model where noted). History 2016-2026 is reconstructed history under current-model methodology with per-row provenance flags; as-published point-in-time vintages accrue in the append-only score_vintage store from July 2026 forward under a written no-restatement policy. Hypothetical results; past performance does not guarantee future results.
Whitepaper: "Consumer Demand as an Alpha Source" — available in diligence along with study code and the red-flags review policy (weak results are documented internally and excluded from marketing, never massaged).