Verdict: Volume Climax Base Rates (M-018) is rated “NOT VALIDATED” in its validation review dated 3 Sep 2026.
Key result: Forward base-rate estimator: empirical-Bayes, pooled over trailing 252-session sample (cap 200,000 events).
Key result: Region effect: material — e.g. Korean climaxes resolve far more bearishly than US/Taiwanese ones.
How it was tested: the record does not state separate training and test periods; out-of-sample design: none.
Horizon: 20 sessions forward from the event. Coverage: ~1,700-name panel (S&P 500 + Nifty + curated watchlists); events from the trailing 252 sessions.
Stage: Production-context. Version v1; the hub record is as of 6 Nov 2026.
The verdict
NOT VALIDATED as a predictive base rate — the event definition is sound and lookahead-free, but no held-out test of the pooled forward base rates themselves was found; treat Base Up20 etc. as a descriptive statistic pending that test.
Development summary: A clean, no-lookahead record-volume definition (current bar always excluded from its own comparison window) with EB-pooled, region-aware forward base rates — a well-built descriptive layer, correctly kept as context rather than a screen.
What the review allows
The tier (HVQ/HVY/HVE) and bar-type/MA context as a purely descriptive flag of what happened.
The pooled base rate as a rough, region-aware reference point for discussion, not as a validated forecast.
What the review rules out
Screen, rank or size on the pooled base rate — no held-out evaluation of it exists in reviewed evidence.
Assume the region effect (e.g. Korea vs US/Taiwan) is stable without a dated, quantified study behind it.
What it predicts
Item
Detail
Question it answers
When a daily bar prints a fresh quarter/year/all-time record of volume (HVQ/HVY/HVE), what did similar events historically do over the next 20 sessions, and is that base rate different by region?
Method
Record-volume tier + EB pooling
How it works
Deterministic record-volume tiering (rolling 63d/252d/expanding max, current bar excluded) + hierarchical empirical-Bayes pooling of forward outcomes by tier x bar-type x trend, region-aware
Inputs
Volume vs 63-day max (HVQ), Volume vs 252-day max (HVY), All-time volume record (HVE), Close location (accum/distrib), Nearest MA within 1 ATR
Output
Per-event tier (HVQ/HVY/HVE), bar type, MA context, and EB-pooled forward base rates (e.g. Base Up20) by region
Horizon
20 sessions forward from the event
Coverage
~1,700-name panel (S&P 500 + Nifty + curated watchlists); events from the trailing 252 sessions
Markets
US, India
Model type
Model · ML/Stat
Stage
Production-context
Version
v1, first recorded 3 Sep 2026
Validation evidence
Training data: not stated. Test data: not stated. Out-of-sample design: none.
Headline results from the model's development record, as the hub publishes them.
Measure
Result
Compared with
Note
Event definition
volume >= prior-quarter max (qmax), tiered HVQ/HVY/HVE
current bar excluded from every prior window (.shift(1))
no lookahead by construction
Region effect on Base Up20
material
a single global pooled rate
Korean climaxes resolve far more bearishly than US or Taiwanese ones (Volume Climax tab copy)
Pooling sample cap
200,000 events (POOL_SAMPLE)
full history when smaller
same capping pattern as
What the validation review found
Is the design sound? Sound event definition (no lookahead) and a sensible pooling hierarchy mirroring 's approach; the gap is evidentiary, not conceptual.
Data quality. Same ~1,700-name panel as the other technical engines; HVE's 400-bar floor is a reasonable, if coarse, guard against spurious all-time calls on short-history names.
Backtest. None found for this component specifically; is a schema/build contract, not a performance study.
Limitations
No held-out/train-test evaluation of the pooled base rates themselves is in the reviewed evidence; documents the build contract, not a backtest.
Region-conditioned rates fall back to global below 30 events (`REGION_EXP_MIN`), with no explicit thin-region flag noted on the row schema reviewed.
HVE is gated at >=400 bars specifically to avoid a spurious 'all-time record' call on a short-history name — a real but coarse fix (a 401-bar name still has a short 'all-time').
Questions
Is the Volume Climax Base Rates validated?
Its validation review dated 3 Sep 2026 rated it NOT VALIDATED as a predictive base rate — the event definition is sound and lookahead-free, but no held-out test of the pooled forward base rates themselves was found; treat Base Up20 etc. as a descriptive statistic pending that test.
What does the Volume Climax Base Rates predict?
It answers: When a daily bar prints a fresh quarter/year/all-time record of volume (HVQ/HVY/HVE), what did similar events historically do over the next 20 sessions, and is that base rate different by region? Output: Per-event tier (HVQ/HVY/HVE), bar type, MA context, and EB-pooled forward base rates (e.g. Base Up20) by region. Horizon: 20 sessions forward from the event. Coverage: ~1,700-name panel (S&P 500 + Nifty + curated watchlists); events from the trailing 252 sessions.
How was the Volume Climax Base Rates tested?
Training data not stated, test data not stated, out-of-sample design: none. Forward base-rate estimator: empirical-Bayes, pooled over trailing 252-session sample (cap 200,000 events); Region effect: material — e.g. Korean climaxes resolve far more bearishly than US/Taiwanese ones; Status: production-context — computed daily, not a screen input.