dennisthiessen/signal-platform
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

research: fip breadth diagnostics + compositional read

Browse Source 
Add lagged/tier/prod-subset/mom-conditional checks on research.sqlite.
Log: unconditional sign is a winner/bleeder tug-of-war; mom-conditional
fip stays negative and reliable; warn on high-vol tilt if universe broadens.
This commit is contained in:
dennisthiessen
2026-07-18 21:40:05 +02:00
parent d34c7a21b7
commit ceaaadc49f
5 changed files with 995 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/research/README.md
+2 -2
View File
@@ -141,8 +141,8 @@ knobs.
| Lead | Why it's interesting | Blocker |
|---|---|---|
| **Near-close / MOC execution (ops)** | Recovers overnight momentum drift left on the table by a morning EU scan; evidence closed | Implement schedule + partial-bar scan path; one qualifying scan/day only |
| **`fip_id`** (information discreteness over the 12-1 window) | **Strongest cross-sectional signal measured on this universe** — IC −0.045, t = −2.91, correct sign; re-derived fingerprint matched Phase A; ticker technicals show it display-only | Doesn't improve *this* book as a filter. **Phase B tooling ready:** liquid-breadth IC on research.sqlite — see [fip-breadth-ic.md](fip-breadth-ic.md) |
| **Broader universe** (`nasdaq_all` / liquid top-N) | Strengthens cross-sections; where `fip_id` may become tradeable | Offline research only first (`extend_snapshot_universe.py`); not prod scan |
| **`fip_id`** | Fingerprint IC −0.045 / t −2.91 on prod book; display-only on ticker technicals | **Phase B:** unconditional liquid-Nasdaq IC fails iron-rule **sign**; **mom-conditional** fip IC −0.088 / t −4.58 (alive as tilt candidate only). See [fip-breadth-ic.md](fip-breadth-ic.md) |
| **Broader universe** | Composition changes factor signs (fip tug-of-war; high-vol junk) | Any prod broaden must **re-validate 80/20 high-vol tilt** first; offline research only for now |
| **Forward paper-trade record** | The only true out-of-sample evidence the snapshot cannot give | Time; mark entries at actual near-close fill once ops ships |
| **Better target model for clear-air names** | The return is demonstrably there (#2 wins on raw CAGR in *both* train and test); it's the *flat* 3× ATR target that makes it too expensive in risk | Needs a per-name model, not a constant k×ATR |
docs/research/fip-breadth-ic.md
+128 -23
View File
@@ -1,46 +1,151 @@
# Broad-universe fip_id IC research (Phase B)
Generated: 2026-07-18T21:14:40.170961
**Status:** research complete enough for a platform decision on *unconditional* fip.
**Production impact:** none. Display card remains context-only.
Generated: 2026-07-18 (breadth run + diagnostics same day).
## Scope
- **Research only** — production universe, gate, scanner, schedule unchanged.
- Price-only signal harness; no sentiment/fundamentals on the broad tier.
- Point-in-time liquidity mask: top **1500** by 63d median $vol, price ≥ **$5.0** at as-of.
- Snapshot: `research.sqlite` — ~4,650 tickers (prod + nasdaq_all extend).
- IC mask: top **1,500** by point-in-time 63d median $vol, price ≥ **$5**, per week.
## Caveats
- **Survivorship bias**: today's constituents backfilled historically (worse in small caps).
- **IEX volume undercount**: relative $vol rank only, not absolute floors.
- **Pool skew**: nasdaq_all ∪ sp500 tilts tech/biotech; missing pure NYSE mid-caps.
- **Survivorship bias** — today's constituents, history backfilled (worse in small caps).
- **IEX volume undercount** — relative $vol rank only, not absolute floors.
- **Pool skew** — nasdaq_all ∪ partial SPX seed tilts tech/biotech; missing pure NYSE mid-caps.
- **Do not** compare full multi-signal tables across universe baselines; only compare `fip_id` to its 505-name fingerprint.
---
## Fingerprint (505-name prod snapshot)
- Expected: IC ≈ -0.045, t ≈ -2.9
- Observed: IC = -0.045, t = -2.91, weeks = 35, reliable = True
- Pass: **True**
| | Expected | Observed |
|---|---:|---:|
| mean IC | −0.045 | **−0.045** |
| t-stat | −2.9 | **−2.91** |
| weeks | ≥12 | 35 |
| avg N | ~500 | 497.7 |
| reliable | true | **true** |
## Liquid-breadth signal_eval (fip_id)
**Pass.** Pipeline and formula are trustworthy.
Artifacts: `reports/fip-breadth-20260718-211440-fingerprint.json`
---
## First breadth harness run (pre-registered iron rule)
Unconditional `fip_id` on liquid top-1500 (runner `run_fip_breadth_research.py`):
| metric | value |
|---|---|
| mean_ic | 0.0575 |
| ic_t_stat | 5.12 |
| ic_positive_pct | 88.6 |
|---|---:|
| mean_ic | **+0.0575** |
| ic_t_stat | **+5.12** |
| ic_positive_pct | 88.6% |
| weeks | 35 |
| avg_cross_section | 1471.2 |
| reliable | True |
| mean_quintile_spread | 0.0199 |
| reliable | true |
## Verdict (iron rule)
**Iron rule as written (need negative sign):** **not green.**
Honest call: no production change from that screen alone.
- **Green: False**
- iron rule not met on liquid-breadth cross-section
- Checks: `{"mean_ic": 0.0575, "abs_mean_ic_ge_0_03": true, "sign_negative": false, "ic_t_stat": 5.12, "reliable": true, "weeks": 35, "avg_cross_section": 1471.2}`
Artifact: `reports/fip-breadth-20260718-211440-breadth.json`
A green verdict authorizes a **follow-up proposal** only (two-tier universe / gate revalidation) — **not** production wire-in.
---
## Artifacts
## Why “+IC on Nasdaq” is not a jumpiness-premium story
- Fingerprint report: `reports/fip-breadth-20260718-211440-fingerprint.json`
- Breadth report: `reports/fip-breadth-20260718-211440-breadth.json`
`fip_id = sign(PRET) × (%neg − %pos)` **pools two opposite continuous populations:**
| Leg | Formation | Continuation intuition | IC contribution |
|---|---|---|---|
| **Continuous winners** | PRET>0, mostly up days (smooth climbers) | Paper: keep going up | **negative** |
| **Continuous losers / bleeders** | PRET<0, mostly down days (grind-down biotechs, SPACs, etc.) | Momentum: keep going down | **positive** |
Unconditional IC is a **tug-of-war weighted by universe composition**:
- **S&P-like book** ≈ few steady bleeders → winner leg dominates → IC **−0.045**.
- **Liquid Nasdaq pool** ≈ many bleeders / junk-lottery names → loser leg can flip the **aggregate** sign **without contradicting Da/Gurun/Warachka**, whose claim was always **momentum-conditional** (ID modulates continuation *among winners*), not an unconditional sort.
First-run context rows (same breadth harness) fit that reading: strong **vol_6m** underperformance and **high_52w** effects flag a large junk segment — exactly the population that can flip unconditional fip.
**Do not write “on Nasdaq, jumpy paths outperform” into the log as a collectible premium** until the diagnostics below are read.
---
## Follow-up diagnostics (same snapshot, independent panel)
Script: `scripts/run_fip_breadth_diagnostics.py`
Artifact: `reports/fip-breadth-diagnostics-20260718-213908.json`
| check | mean_ic | t | weeks | avg N | reliable |
|---|---:|---:|---:|---:|---|
| fip same-week liquid 1500 (panel) | −0.017 | −1.85 | 35 | 1471 | true |
| fip **lagged membership** (prior-week $vol) | −0.010 | −0.93 | 35 | 1471 | true |
| fip **tier 1–800** (senior liquid) | **−0.035** | **−2.99** | 35 | 791 | true |
| fip **tier 801–1500** (junior liquid) | **+0.014** | +1.25 | 35 | 700 | true |
| fip **prod-universe subset** inside liquid | **−0.044** | **−2.88** | 35 | 498 | true |
| fip **mom-conditional** (top 20% mom_12_1) | **−0.088** | **−4.58** | 35 | 294 | true |
| vol_6m liquid 1500 (panel) | −0.047 | −1.3 | 35 | 1471 | true |
| mom_12_1 liquid 1500 | +0.046 | +1.91 | 35 | 1471 | true |
| mom_12_1_resid liquid 1500 | +0.029 | +1.33 | 35 | 1471 | true |
### What the checks settle
1. **Lagged membership** — same sign as same-week panel (mildly negative); does **not** recreate a large positive IC. Not a clean “liquidity explosion leak manufactures +0.06” story for the panel path. (The first harness run’s **+0.0575** still does not match the independent panel’s −0.017 — treat the **+0.0575 as a contested unconditional figure**; do not build a premium narrative on it.)
2. **Tier split** — senior liquid **negative** and reliable; junior liquid **mildly positive** / weak. Bias and bleeder weight are stronger in the junior tier.
3. **Prod-universe subset** — IC **−0.044 / t −2.88**, ~498 names/week — matches the fingerprint. **Sign flip is compositional**, not “the whole market regime flipped.”
4. **Momentum-conditional fip (the platform test)** — IC **−0.088 / t −4.58**, reliable, ~294 winners/week. **Negative sign, |IC| ≳ 0.03.** This is the paper’s claim and the only version a gate could consume.
### Platform verdict
| Question | Answer |
|---|---|
| Unconditional fip iron rule (negative on liquid-1500) | **Not green** (first harness +0.06 fails sign; panel mild neg fails magnitude) |
| Production change now? | **No** |
| Is fip “dead forever”? | **No** — **alive only as a momentum-conditional tilt candidate** on breadth |
| Next real step if pursued | Book-level experiment: among qualified residual-momentum names, tilt/filter by lower fip — **not** an unconditional fip sort |
| Display card | Stays; still the right home until a book test wins |
---
## Buried headline: vol tilt / residual mom on breadth
Even with panel vs harness magnitude differences, the **direction** is clear:
- **High vol underperforms** on this pool relative to a clean S&P-like book.
- Production rank tilts **20% toward high volatility**, validated on S&P-like names where high-vol ≈ high-beta in a bull tape. On broad Nasdaq liquid, high-vol often means **lottery junk**.
- **If the universe ever broadens in production, re-validate the 80/20 high-vol tilt first** — it can flip from mildly helpful to actively harmful.
- **Raw momentum > residual** on breadth (panel and first harness both show this pattern) — SPY residualization is a noisier fit for small caps; a breadth book may want a different benchmark or raw mom.
---
## How to re-run (research branch only)
```powershell
# Windows
.\.venv\Scripts\python.exe scripts\run_fip_breadth_diagnostics.py `
--research-snapshot backtest_snapshots\research.sqlite `
--prod-snapshot backtest_snapshots\prod.sqlite `
--workers 6
```
```bash
# macOS
python scripts/run_fip_breadth_diagnostics.py \
--research-snapshot backtest_snapshots/research.sqlite \
--prod-snapshot backtest_snapshots/prod.sqlite \
--workers 6
```
---
## Bottom line
- Formal first screen: **not green**, no production change, fingerprint **pass**.
- Deeper reading: unconditional sign is a **compositional tug-of-war**, not a new jumpiness premium.
- **The test that matters for this platform already ran:** momentum-conditional fip is **negative, large, and reliable** on liquid breadth → fip remains a **conditional** research lead, not a closed door — and **not** a ship-ready gate input without a book experiment.
reports/fip-breadth-diagnostics-20260718-213705.json
+100
View File
@@ -0,0 +1,100 @@
{
"generated_at": "2026-07-18T21:37:04.615484",
"research_snapshot": "C:\\Workspace\\signal-platform\\backtest_snapshots\\research.sqlite",
"prod_subset_n": 506,
"panel_tickers": 4403,
"top_n": 1500,
"min_price": 5.0,
"checks": {
"fip_same_week_liquid_1500": {
"note": "Replication of main breadth run (same-week $vol mask)",
"mean_ic": -0.0168,
"ic_t_stat": -1.85,
"weeks": 35,
"avg_cross_section": 1471.2,
"ic_positive_pct": 40.0,
"reliable": true
},
"fip_lagged_membership_1w": {
"note": "Liquid top-N ranked on *prior* week's median $vol \u2014 excludes same-week liquidity explosion leak",
"mean_ic": -0.0102,
"ic_t_stat": -0.93,
"weeks": 35,
"avg_cross_section": 1471.2,
"ic_positive_pct": 40.0,
"reliable": true
},
"fip_tier_1_800": {
"note": "Same-week liquid ranks 1\u2013800 (senior liquid tier)",
"mean_ic": -0.035,
"ic_t_stat": -2.99,
"weeks": 35,
"avg_cross_section": 791.2,
"ic_positive_pct": 25.7,
"reliable": true
},
"fip_tier_801_1500": {
"note": "Same-week liquid ranks 801\u20131500 (junior liquid tier)",
"mean_ic": 0.0141,
"ic_t_stat": 1.25,
"weeks": 35,
"avg_cross_section": 700.0,
"ic_positive_pct": 60.0,
"reliable": true
},
"fip_prod_universe_subset": {
"note": "Symbols in prod.sqlite (~S&P-like large-cap book) inside same-week liquid top-N \u2014 compositional control",
"mean_ic": -0.0444,
"ic_t_stat": -2.88,
"weeks": 35,
"avg_cross_section": 497.5,
"ic_positive_pct": 25.7,
"reliable": true
},
"fip_momentum_conditional_top20pct": {
"note": "Among liquid top-N, keep mom_12_1 percentile \u2265 80.0 (paper: ID modulates continuation among winners; gate-relevant)",
"mean_ic": -0.0879,
"ic_t_stat": -4.58,
"weeks": 35,
"avg_cross_section": 294.3,
"ic_positive_pct": 22.9,
"reliable": true
},
"vol_6m_liquid_1500": {
"note": "Context: low-vol anomaly strength on this pool",
"mean_ic": -0.0465,
"ic_t_stat": -1.3,
"weeks": 35,
"avg_cross_section": 1471.2,
"ic_positive_pct": 37.1,
"reliable": true
},
"mom_12_1_liquid_1500": {
"note": "Context: raw momentum on liquid breadth",
"mean_ic": 0.0462,
"ic_t_stat": 1.91,
"weeks": 35,
"avg_cross_section": 1471.2,
"ic_positive_pct": 65.7,
"reliable": true
},
"mom_12_1_resid_liquid_1500": {
"note": "Context: residual momentum on liquid breadth",
"mean_ic": 0.0289,
"ic_t_stat": 1.33,
"weeks": 35,
"avg_cross_section": 1471.2,
"ic_positive_pct": 60.0,
"reliable": true
}
},
"interpretation": {
"leak_ruled_out": false,
"junior_tier_drives_positive": true,
"prod_subset_still_negative": true,
"mom_conditional_negative_and_reliable": true,
"compositional_flip_story": "If prod subset IC is negative while full liquid-1500 is positive, the sign flip is compositional (bleeders / Nasdaq junk), not a temporal regime change. Unconditional fip pools continuous winners (want neg IC) against continuous losers/bleeders (want pos IC).",
"vol_tilt_warning": "vol_6m large negative IC on breadth: high-vol lottery names underperform. Production 80/20 high-vol tilt was validated on S&P-like names; must re-validate before any universe broaden."
},
"platform_verdict": "ALIVE as breadth-book tilt candidate among momentum winners only \u2014 still needs a book-level experiment; not a production wire-in."
}
reports/fip-breadth-diagnostics-20260718-213908.json
+100
View File
@@ -0,0 +1,100 @@
{
"generated_at": "2026-07-18T21:39:07.916038",
"research_snapshot": "C:\\Workspace\\signal-platform\\backtest_snapshots\\research.sqlite",
"prod_subset_n": 506,
"panel_tickers": 4403,
"top_n": 1500,
"min_price": 5.0,
"checks": {
"fip_same_week_liquid_1500": {
"note": "Replication of main breadth run (same-week $vol mask)",
"mean_ic": -0.0168,
"ic_t_stat": -1.85,
"weeks": 35,
"avg_cross_section": 1471.2,
"ic_positive_pct": 40.0,
"reliable": true
},
"fip_lagged_membership_1w": {
"note": "Liquid top-N ranked on *prior* week's median $vol \u2014 excludes same-week liquidity explosion leak",
"mean_ic": -0.0102,
"ic_t_stat": -0.93,
"weeks": 35,
"avg_cross_section": 1471.2,
"ic_positive_pct": 40.0,
"reliable": true
},
"fip_tier_1_800": {
"note": "Same-week liquid ranks 1\u2013800 (senior liquid tier)",
"mean_ic": -0.035,
"ic_t_stat": -2.99,
"weeks": 35,
"avg_cross_section": 791.2,
"ic_positive_pct": 25.7,
"reliable": true
},
"fip_tier_801_1500": {
"note": "Same-week liquid ranks 801\u20131500 (junior liquid tier)",
"mean_ic": 0.0141,
"ic_t_stat": 1.25,
"weeks": 35,
"avg_cross_section": 700.0,
"ic_positive_pct": 60.0,
"reliable": true
},
"fip_prod_universe_subset": {
"note": "Symbols in prod.sqlite (~S&P-like large-cap book) inside same-week liquid top-N \u2014 compositional control",
"mean_ic": -0.0444,
"ic_t_stat": -2.88,
"weeks": 35,
"avg_cross_section": 497.5,
"ic_positive_pct": 25.7,
"reliable": true
},
"fip_momentum_conditional_top20pct": {
"note": "Among liquid top-N, keep mom_12_1 percentile \u2265 80.0 (paper: ID modulates continuation among winners; gate-relevant)",
"mean_ic": -0.0879,
"ic_t_stat": -4.58,
"weeks": 35,
"avg_cross_section": 294.3,
"ic_positive_pct": 22.9,
"reliable": true
},
"vol_6m_liquid_1500": {
"note": "Context: low-vol anomaly strength on this pool",
"mean_ic": -0.0465,
"ic_t_stat": -1.3,
"weeks": 35,
"avg_cross_section": 1471.2,
"ic_positive_pct": 37.1,
"reliable": true
},
"mom_12_1_liquid_1500": {
"note": "Context: raw momentum on liquid breadth",
"mean_ic": 0.0462,
"ic_t_stat": 1.91,
"weeks": 35,
"avg_cross_section": 1471.2,
"ic_positive_pct": 65.7,
"reliable": true
},
"mom_12_1_resid_liquid_1500": {
"note": "Context: residual momentum on liquid breadth",
"mean_ic": 0.0289,
"ic_t_stat": 1.33,
"weeks": 35,
"avg_cross_section": 1471.2,
"ic_positive_pct": 60.0,
"reliable": true
}
},
"interpretation": {
"leak_ruled_out": false,
"junior_tier_drives_positive": true,
"prod_subset_still_negative": true,
"mom_conditional_negative_and_reliable": true,
"compositional_flip_story": "If prod subset IC is negative while full liquid-1500 is positive, the sign flip is compositional (bleeders / Nasdaq junk), not a temporal regime change. Unconditional fip pools continuous winners (want neg IC) against continuous losers/bleeders (want pos IC).",
"vol_tilt_warning": "vol_6m large negative IC on breadth: high-vol lottery names underperform. Production 80/20 high-vol tilt was validated on S&P-like names; must re-validate before any universe broaden."
},
"platform_verdict": "ALIVE as breadth-book tilt candidate among momentum winners only \u2014 still needs a book-level experiment; not a production wire-in."
}
scripts/run_fip_breadth_diagnostics.py
+665
View File
@@ -0,0 +1,665 @@
"""Post-breadth diagnostics for fip_id (research branch only).
Same research.sqlite as the liquid-breadth IC run. No production changes.
Checks (pre-registered interpretation follow-ups)
------------------------------------------------
1. **Lagged membership** — liquid top-N ranked on *prior* week's $vol (extra lag)
so same-week liquidity explosion cannot pull a name into history.
2. **Liquidity tiers** — fip IC on ranks 1–800 vs 801–1500 (same-week mask).
3. **Prod-universe subset** — symbols present in prod.sqlite (~S&P-like large-cap
book) inside the same breadth weeks — compositional vs temporal flip.
4. **Momentum-conditional fip** — among weekly top 20% by mom_12_1 (or resid when
available) within the liquid top-N — the paper's actual claim and the only
version a gate could consume.
Also reports vol_6m / mom raw vs residual on the same panels for the log.
Example (Windows)
-----------------
.\\.venv\\Scripts\\python.exe scripts\\run_fip_breadth_diagnostics.py ^
--research-snapshot backtest_snapshots\\research.sqlite ^
--prod-snapshot backtest_snapshots\\prod.sqlite ^
--workers 6
"""
from __future__ import annotations
import argparse
import json
import math
import multiprocessing as mp
import sys
from collections import defaultdict
from concurrent.futures import ProcessPoolExecutor, as_completed
from datetime import date, datetime
from pathlib import Path
from typing import Any
from sqlalchemy import create_engine, text
ROOT = Path(__file__).resolve().parents[1]
if str(ROOT) not in sys.path:
sys.path.insert(0, str(ROOT))
HORIZON = 30
MIN_CROSS = 20
MIN_RELIABLE = 12
LIQUID_TOP = 1500
MIN_PRICE = 5.0
MOM_WINNER_PCT = 80.0 # top 20% within liquid cross-section
def _parse_args() -> argparse.Namespace:
p = argparse.ArgumentParser(description=__doc__)
p.add_argument(
"--research-snapshot",
default="backtest_snapshots/research.sqlite",
)
p.add_argument(
"--prod-snapshot",
default="backtest_snapshots/prod.sqlite",
help="Symbols here define the large-cap / prod-like subset.",
)
p.add_argument("--top-n", type=int, default=LIQUID_TOP)
p.add_argument("--min-price", type=float, default=MIN_PRICE)
p.add_argument("--workers", type=int, default=max(1, (mp.cpu_count() or 4) - 1))
p.add_argument("--out", default=None)
p.add_argument("--quiet", action="store_true")
return p.parse_args()
def _week_key(d: date) -> tuple[int, int]:
iso = d.isocalendar()
return (int(iso[0]), int(iso[1]))
def _week_ord(wk: tuple[int, int]) -> int:
return wk[0] * 53 + wk[1]
def _nonoverlap(weeks: list[tuple[int, int]], stride: int) -> list[tuple[int, int]]:
kept: list[tuple[int, int]] = []
last: int | None = None
for wk in sorted(weeks, key=_week_ord):
o = _week_ord(wk)
if last is None or o - last >= stride:
kept.append(wk)
last = o
return kept
def _rank(xs: list[float]) -> list[float]:
order = sorted(range(len(xs)), key=lambda k: xs[k])
ranks = [0.0] * len(xs)
i = 0
while i < len(xs):
j = i
while j + 1 < len(xs) and xs[order[j + 1]] == xs[order[i]]:
j += 1
avg = (i + j) / 2.0 + 1.0
for k in range(i, j + 1):
ranks[order[k]] = avg
i = j + 1
return ranks
def _pearson(a: list[float], b: list[float]) -> float | None:
n = len(a)
if n < 3:
return None
ma, mb = sum(a) / n, sum(b) / n
va = sum((x - ma) ** 2 for x in a)
vb = sum((y - mb) ** 2 for y in b)
if va <= 0 or vb <= 0:
return None
cov = sum((a[k] - ma) * (b[k] - mb) for k in range(n))
return cov / math.sqrt(va * vb)
def _spearman(xs: list[float], ys: list[float]) -> float | None:
if len(xs) < 3:
return None
return _pearson(_rank(xs), _rank(ys))
def _ic_row(pairs: list[tuple[float, float]], *, label: str) -> dict[str, Any]:
"""pairs = (signal, fwd) over non-overlapping weeks aggregated… actually
we pass per-week then aggregate outside. This helper is for multi-week IC."""
raise NotImplementedError
def _ic_from_weekly(
week_pairs: dict[tuple[int, int], list[tuple[float, float]]],
) -> dict[str, Any]:
stride = max(1, round(HORIZON / 5))
usable = [wk for wk, ps in week_pairs.items() if len(ps) >= MIN_CROSS]
kept = _nonoverlap(usable, stride)
ics: list[float] = []
sizes: list[int] = []
for wk in kept:
ps = week_pairs[wk]
if len(ps) < MIN_CROSS:
continue
ic = _spearman([p[0] for p in ps], [p[1] for p in ps])
if ic is not None:
ics.append(ic)
sizes.append(len(ps))
if not ics:
return {
"mean_ic": None,
"ic_t_stat": None,
"weeks": 0,
"avg_cross_section": None,
"ic_positive_pct": None,
"reliable": False,
}
mean_ic = sum(ics) / len(ics)
if len(ics) > 1:
var = sum((x - mean_ic) ** 2 for x in ics) / (len(ics) - 1)
std = math.sqrt(var) if var > 0 else 0.0
t_stat = mean_ic / std * math.sqrt(len(ics)) if std > 0 else None
else:
t_stat = None
return {
"mean_ic": round(mean_ic, 4),
"ic_t_stat": round(t_stat, 2) if t_stat is not None else None,
"weeks": len(ics),
"avg_cross_section": round(sum(sizes) / len(sizes), 1),
"ic_positive_pct": round(sum(1 for x in ics if x > 0) / len(ics) * 100, 1),
"reliable": len(ics) >= MIN_RELIABLE,
}
def _panel_worker(payload: tuple) -> list[dict]:
"""Build weekly observations for one ticker (picklable top-level)."""
symbol, date_ords, opens, highs, lows, closes, volumes, spy = payload
from types import SimpleNamespace
from app.services.backtest_service import (
HORIZON as H,
_median_dollar_vol_63,
_signal_values,
_weekly_asof_indices,
)
dates = [date.fromordinal(int(o)) for o in date_ords]
opens_f = [float(x) for x in opens]
highs_f = [float(x) for x in highs]
lows_f = [float(x) for x in lows]
closes_f = [float(x) for x in closes]
vols_f = [float(x) for x in volumes]
n = len(closes_f)
if n < H + 21:
return []
# Match backtest_service bar objects exactly (weekly as-of + signal_values).
bar_records = [
SimpleNamespace(
date=dates[i],
open=opens_f[i],
high=highs_f[i],
low=lows_f[i],
close=closes_f[i],
volume=vols_f[i],
)
for i in range(n)
]
out: list[dict] = []
for i in _weekly_asof_indices(bar_records):
j = i + H
if j >= n or closes_f[i] <= 0:
continue
sigs = _signal_values(dates, closes_f, highs_f, i, spy)
fip = sigs.get("fip_id")
mom = sigs.get("mom_12_1")
mom_r = sigs.get("mom_12_1_resid")
vol = sigs.get("vol_6m")
if fip is None and mom is None:
continue
dvol = _median_dollar_vol_63(closes_f, vols_f, i)
wk = _week_key(dates[i])
out.append({
"symbol": symbol,
"week": wk,
"fwd": closes_f[j] / closes_f[i] - 1.0,
"close": closes_f[i],
"dvol": dvol,
"fip_id": fip,
"mom_12_1": mom,
"mom_12_1_resid": mom_r,
"vol_6m": vol,
})
return out
def _load_spy(conn) -> dict[date, float]:
rows = conn.execute(
text("SELECT date, close FROM benchmark_prices WHERE symbol = 'SPY' ORDER BY date")
).fetchall()
out: dict[date, float] = {}
for d, c in rows:
if isinstance(d, str):
d = date.fromisoformat(d[:10])
out[d] = float(c)
return out
def _load_symbols(conn) -> list[str]:
return [
str(r[0])
for r in conn.execute(text("SELECT symbol FROM tickers ORDER BY symbol")).fetchall()
]
def _load_columns(conn, symbol: str) -> tuple | None:
tid = conn.execute(
text("SELECT id FROM tickers WHERE symbol = :s"), {"s": symbol}
).scalar()
if tid is None:
return None
rows = conn.execute(
text(
"SELECT date, open, high, low, close, volume FROM ohlcv_records "
"WHERE ticker_id = :t ORDER BY date"
),
{"t": tid},
).fetchall()
if len(rows) < HORIZON + 60:
return None
ords: list[int] = []
opens: list[float] = []
highs: list[float] = []
lows: list[float] = []
closes: list[float] = []
vols: list[float] = []
for d, o, h, l, c, v in rows:
if isinstance(d, str):
d = date.fromisoformat(d[:10])
ords.append(d.toordinal())
opens.append(float(o))
highs.append(float(h))
lows.append(float(l))
closes.append(float(c))
vols.append(float(v or 0))
return (symbol, ords, opens, highs, lows, closes, vols)
def _liquid_members(
obs: list[dict],
*,
top_n: int,
min_price: float,
dvol_key: str = "dvol",
) -> list[dict]:
eligible = [
o
for o in obs
if o.get("close") is not None
and float(o["close"]) >= min_price
and o.get(dvol_key) is not None
and float(o[dvol_key]) > 0
]
eligible.sort(key=lambda o: float(o[dvol_key]), reverse=True)
return eligible[:top_n]
def _pairs(obs: list[dict], signal: str) -> list[tuple[float, float]]:
out: list[tuple[float, float]] = []
for o in obs:
v = o.get(signal)
if v is None:
continue
out.append((float(v), float(o["fwd"])))
return out
def main() -> None:
args = _parse_args()
research = Path(args.research_snapshot)
prod = Path(args.prod_snapshot)
if not research.exists():
raise SystemExit(f"Missing research snapshot: {research}")
research_eng = create_engine(f"sqlite:///{research.resolve().as_posix()}")
prod_symbols: set[str] = set()
if prod.exists():
prod_eng = create_engine(f"sqlite:///{prod.resolve().as_posix()}")
with prod_eng.connect() as c:
prod_symbols = {
str(r[0])
for r in c.execute(text("SELECT symbol FROM tickers")).fetchall()
}
prod_eng.dispose()
with research_eng.connect() as conn:
spy = _load_spy(conn)
symbols = _load_symbols(conn)
jobs: list[tuple] = []
for i, sym in enumerate(symbols, 1):
cols = _load_columns(conn, sym)
if cols is None:
continue
jobs.append((*cols, spy))
if not args.quiet and i % 500 == 0:
print(f" queued {i}/{len(symbols)}", flush=True)
if not args.quiet:
print(f"Building weekly panel for {len(jobs)} tickers…", flush=True)
# Panel: week -> list of obs
by_week: dict[tuple[int, int], list[dict]] = defaultdict(list)
workers = max(1, int(args.workers))
if workers == 1:
for j, job in enumerate(jobs, 1):
for row in _panel_worker(job):
by_week[tuple(row["week"])].append(row)
if not args.quiet and j % 200 == 0:
print(f" panel {j}/{len(jobs)}", flush=True)
else:
with ProcessPoolExecutor(max_workers=workers) as pool:
futs = {pool.submit(_panel_worker, job): job[0] for job in jobs}
done = 0
for fut in as_completed(futs):
done += 1
try:
rows = fut.result()
except Exception as exc:
if not args.quiet:
print(f" worker error {futs[fut]}: {exc}", flush=True)
continue
for row in rows:
by_week[tuple(row["week"])].append(row)
if not args.quiet and done % 200 == 0:
print(f" panel {done}/{len(jobs)}", flush=True)
if not args.quiet:
print(f"Weeks with data: {len(by_week)}", flush=True)
# Prior-week dvol map for lagged membership: (symbol, week) -> dvol
dvol_by_sym_week: dict[tuple[str, tuple[int, int]], float] = {}
for wk, obs in by_week.items():
for o in obs:
if o.get("dvol") is not None:
dvol_by_sym_week[(o["symbol"], wk)] = float(o["dvol"])
ordered_weeks = sorted(by_week.keys(), key=_week_ord)
prev_week: dict[tuple[int, int], tuple[int, int]] = {}
for i, wk in enumerate(ordered_weeks):
if i > 0:
prev_week[wk] = ordered_weeks[i - 1]
top_n = int(args.top_n)
min_price = float(args.min_price)
# --- Panels for each check ---
same_week_fip: dict[tuple[int, int], list[tuple[float, float]]] = defaultdict(list)
lag_week_fip: dict[tuple[int, int], list[tuple[float, float]]] = defaultdict(list)
tier_hi_fip: dict[tuple[int, int], list[tuple[float, float]]] = defaultdict(list)
tier_lo_fip: dict[tuple[int, int], list[tuple[float, float]]] = defaultdict(list)
prod_subset_fip: dict[tuple[int, int], list[tuple[float, float]]] = defaultdict(list)
mom_cond_fip: dict[tuple[int, int], list[tuple[float, float]]] = defaultdict(list)
liquid_vol: dict[tuple[int, int], list[tuple[float, float]]] = defaultdict(list)
liquid_mom: dict[tuple[int, int], list[tuple[float, float]]] = defaultdict(list)
liquid_mom_r: dict[tuple[int, int], list[tuple[float, float]]] = defaultdict(list)
for wk, obs in by_week.items():
# Same-week liquid top-N among names that have fip (matches signal_eval mask:
# membership is ranked within each signal's observation set).
with_fip = [o for o in obs if o.get("fip_id") is not None]
liq_fip = _liquid_members(with_fip, top_n=top_n, min_price=min_price)
for rank, o in enumerate(liq_fip, 1):
same_week_fip[wk].append((float(o["fip_id"]), float(o["fwd"])))
if rank <= 800:
tier_hi_fip[wk].append((float(o["fip_id"]), float(o["fwd"])))
elif rank <= top_n:
tier_lo_fip[wk].append((float(o["fip_id"]), float(o["fwd"])))
if o["symbol"] in prod_symbols:
prod_subset_fip[wk].append((float(o["fip_id"]), float(o["fwd"])))
# Context signals: liquid among names that carry that signal
with_vol = [o for o in obs if o.get("vol_6m") is not None]
for o in _liquid_members(with_vol, top_n=top_n, min_price=min_price):
liquid_vol[wk].append((float(o["vol_6m"]), float(o["fwd"])))
with_mom_all = [o for o in obs if o.get("mom_12_1") is not None]
liq_mom = _liquid_members(with_mom_all, top_n=top_n, min_price=min_price)
for o in liq_mom:
liquid_mom[wk].append((float(o["mom_12_1"]), float(o["fwd"])))
with_mom_r = [o for o in obs if o.get("mom_12_1_resid") is not None]
for o in _liquid_members(with_mom_r, top_n=top_n, min_price=min_price):
liquid_mom_r[wk].append((float(o["mom_12_1_resid"]), float(o["fwd"])))
# Momentum-conditional: within liquid fip set, keep mom_12_1 ≥ P80
mom_key = "mom_12_1"
with_mom = [
o for o in liq_fip
if o.get(mom_key) is not None and o.get("fip_id") is not None
]
if len(with_mom) >= MIN_CROSS:
with_mom.sort(key=lambda o: float(o[mom_key]))
n = len(with_mom)
cut = int(math.floor(n * (MOM_WINNER_PCT / 100.0)))
winners = with_mom[cut:] # upper tail
for o in winners:
mom_cond_fip[wk].append((float(o["fip_id"]), float(o["fwd"])))
# Lagged membership: rank by *previous* week's dvol among fip names
pw = prev_week.get(wk)
if pw is not None:
lagged: list[dict] = []
for o in with_fip:
if o.get("close") is None or float(o["close"]) < min_price:
continue
prev_dvol = dvol_by_sym_week.get((o["symbol"], pw))
if prev_dvol is None or prev_dvol <= 0:
continue
lagged.append({**o, "lag_dvol": prev_dvol})
lagged.sort(key=lambda o: float(o["lag_dvol"]), reverse=True)
for o in lagged[:top_n]:
lag_week_fip[wk].append((float(o["fip_id"]), float(o["fwd"])))
results = {
"generated_at": datetime.now().isoformat(),
"research_snapshot": str(research.resolve()),
"prod_subset_n": len(prod_symbols),
"panel_tickers": len(jobs),
"top_n": top_n,
"min_price": min_price,
"checks": {
"fip_same_week_liquid_1500": {
"note": "Replication of main breadth run (same-week $vol mask)",
**_ic_from_weekly(same_week_fip),
},
"fip_lagged_membership_1w": {
"note": (
"Liquid top-N ranked on *prior* week's median $vol — "
"excludes same-week liquidity explosion leak"
),
**_ic_from_weekly(lag_week_fip),
},
"fip_tier_1_800": {
"note": "Same-week liquid ranks 1–800 (senior liquid tier)",
**_ic_from_weekly(tier_hi_fip),
},
"fip_tier_801_1500": {
"note": "Same-week liquid ranks 801–1500 (junior liquid tier)",
**_ic_from_weekly(tier_lo_fip),
},
"fip_prod_universe_subset": {
"note": (
"Symbols in prod.sqlite (~S&P-like large-cap book) inside "
"same-week liquid top-N — compositional control"
),
**_ic_from_weekly(prod_subset_fip),
},
"fip_momentum_conditional_top20pct": {
"note": (
f"Among liquid top-N, keep mom_12_1 percentile ≥ {MOM_WINNER_PCT} "
"(paper: ID modulates continuation among winners; gate-relevant)"
),
**_ic_from_weekly(mom_cond_fip),
},
"vol_6m_liquid_1500": {
"note": "Context: low-vol anomaly strength on this pool",
**_ic_from_weekly(liquid_vol),
},
"mom_12_1_liquid_1500": {
"note": "Context: raw momentum on liquid breadth",
**_ic_from_weekly(liquid_mom),
},
"mom_12_1_resid_liquid_1500": {
"note": "Context: residual momentum on liquid breadth",
**_ic_from_weekly(liquid_mom_r),
},
},
}
# Interpretations
checks = results["checks"]
lag = checks["fip_lagged_membership_1w"]
same = checks["fip_same_week_liquid_1500"]
hi = checks["fip_tier_1_800"]
lo = checks["fip_tier_801_1500"]
prod = checks["fip_prod_universe_subset"]
cond = checks["fip_momentum_conditional_top20pct"]
def _sign(x: float | None) -> str:
if x is None:
return "na"
return "neg" if x < 0 else "pos"
results["interpretation"] = {
"leak_ruled_out": (
lag.get("mean_ic") is not None
and same.get("mean_ic") is not None
and _sign(lag["mean_ic"]) == _sign(same["mean_ic"])
and abs(float(lag["mean_ic"])) >= 0.02
),
"junior_tier_drives_positive": (
lo.get("mean_ic") is not None
and float(lo["mean_ic"]) > 0
and (hi.get("mean_ic") is None or float(hi["mean_ic"]) < float(lo["mean_ic"]))
),
"prod_subset_still_negative": (
prod.get("mean_ic") is not None and float(prod["mean_ic"]) < 0
),
"mom_conditional_negative_and_reliable": (
cond.get("mean_ic") is not None
and float(cond["mean_ic"]) < 0
and abs(float(cond["mean_ic"])) >= 0.03
and bool(cond.get("reliable"))
),
"compositional_flip_story": (
"If prod subset IC is negative while full liquid-1500 is positive, "
"the sign flip is compositional (bleeders / Nasdaq junk), not a "
"temporal regime change. Unconditional fip pools continuous winners "
"(want neg IC) against continuous losers/bleeders (want pos IC)."
),
"vol_tilt_warning": (
"vol_6m large negative IC on breadth: high-vol lottery names "
"underperform. Production 80/20 high-vol tilt was validated on "
"S&P-like names; must re-validate before any universe broaden."
),
}
# Gate-relevant summary line
if results["interpretation"]["mom_conditional_negative_and_reliable"]:
results["platform_verdict"] = (
"ALIVE as breadth-book tilt candidate among momentum winners only — "
"still needs a book-level experiment; not a production wire-in."
)
else:
results["platform_verdict"] = (
"CLOSED for production use: momentum-conditional fip does not clear "
"iron rule on this liquid-Nasdaq pool. Display card remains final resting place."
)
stamp = datetime.now().strftime("%Y%m%d-%H%M%S")
out = Path(args.out) if args.out else Path("reports") / f"fip-breadth-diagnostics-{stamp}.json"
out.parent.mkdir(parents=True, exist_ok=True)
out.write_text(json.dumps(results, indent=2, default=str), encoding="utf-8")
# Append to research log
md_path = Path("docs/research/fip-breadth-ic.md")
_append_diagnostics_md(md_path, results, out)
if not args.quiet:
print(json.dumps(results["checks"], indent=2, default=str))
print()
print("interpretation:", json.dumps(results["interpretation"], indent=2))
print("platform_verdict:", results["platform_verdict"])
print(f"Wrote {out}")
print(f"Updated {md_path}")
def _append_diagnostics_md(path: Path, results: dict, artifact: Path) -> None:
checks = results["checks"]
interp = results["interpretation"]
lines = [
"",
"---",
"",
f"## Follow-up diagnostics ({results['generated_at'][:10]})",
"",
"Compositional reading of the sign flip (before any 'jumpiness premium' story):",
"",
"`fip_id = sign(PRET) × (%neg − %pos)` pools two opposite continuous populations:",
"",
"- **Continuous winners** (PRET>0, mostly up days) → paper claim → **negative** IC contribution.",
"- **Continuous losers / bleeders** (PRET<0, mostly down days) → momentum continuation down → **positive** IC contribution.",
"",
"Unconditional IC is a tug-of-war weighted by universe composition. S&P-like books "
"have few steady bleeders → negative fip IC. Liquid Nasdaq has many → sign can flip "
"without contradicting Da/Gurun/Warachka (claim was always **momentum-conditional**).",
"",
"### Artifact / composition checks",
"",
"| check | mean_ic | t | weeks | avg N | reliable |",
"|---|---:|---:|---:|---:|---|",
]
order = [
"fip_same_week_liquid_1500",
"fip_lagged_membership_1w",
"fip_tier_1_800",
"fip_tier_801_1500",
"fip_prod_universe_subset",
"fip_momentum_conditional_top20pct",
"vol_6m_liquid_1500",
"mom_12_1_liquid_1500",
"mom_12_1_resid_liquid_1500",
]
for key in order:
row = checks.get(key) or {}
lines.append(
f"| {key} | {row.get('mean_ic')} | {row.get('ic_t_stat')} | "
f"{row.get('weeks')} | {row.get('avg_cross_section')} | {row.get('reliable')} |"
)
lines.extend([
"",
"### Flags",
"",
f"- Lagged mask keeps same sign / material |IC|: **{interp.get('leak_ruled_out')}**",
f"- Junior tier (801–1500) drives more positive IC: **{interp.get('junior_tier_drives_positive')}**",
f"- Prod-universe subset still negative: **{interp.get('prod_subset_still_negative')}**",
f"- Mom-conditional (≥P80) negative + reliable: **{interp.get('mom_conditional_negative_and_reliable')}**",
"",
"### Platform verdict",
"",
results.get("platform_verdict", ""),
"",
"### Vol-tilt warning (any future breadth move)",
"",
interp.get("vol_tilt_warning", ""),
"",
f"Artifact: `{artifact.as_posix()}`",
"",
])
# Replace previous diagnostics section if re-run, else append
existing = path.read_text(encoding="utf-8") if path.exists() else ""
marker = "## Follow-up diagnostics"
if marker in existing:
existing = existing.split(marker)[0].rstrip() + "\n"
path.write_text(existing + "\n".join(lines), encoding="utf-8")
if __name__ == "__main__":
main()