Research campaign · Iteration 1 · unassessed
Milton Berg reconstructed buy/sell signals
Which publicly disclosed Milton Berg buy, sell, and warning templates can be reconstructed as independent Other-indicators with verbatim provenance, without claiming the proprietary MB Edge model?
- Expressions
- 4
- Logged trials
- Not recorded
- Independent events
- Not assessed
- Evidence
- unassessed
What the research found
Imported 53 reconstructed signals in 8 classes from the miltonberg workspace into Other → Indicators. Each card is an independent overlay, not an AllocateSmartly replica and not a claim that we replicate proprietary MB Edge internals.
Official class: mb_edge_official (110 BUY/SELL legs) and mb_edge_first_buy (55 BUY legs) are the published binary history used as alignment ground truth.
Volatility (6): VIX/VXN +10%×2d, VXN +35%/3d, VXN short/long ratio SELL (2025-12-11 cited), 5d-vs-30d optimism warnings.
ROC (7): NASDAQ −7%/3d, 6d/18d ROC ≥105%, weakest-5d-in-180 and 5y, S&P −10% 3d/5d washouts, NASDAQ +10%/10d thrust.
Volume (10): 10-to-1×3d, 12:1/7d, 15:1 day, net up-vol −45%, 375d volume regime, NDX 200d volume extreme, 70:1 day, high-vol-after-advance.
Breadth (7): A/D divergence (1987 historical), DJU record-volume peak, 60d new low, 95%>10d thrust (historical_only without constituents), NASDAQ A/D 8-of-10 down.
Price pattern (10): −12%+3% rebound, drawdown exits 4/7/8%, RUT island reversal, V-bottoms, consecutive-gain streak, deep-decline 18/25/40% into 60d lows.
Sentiment (6): AAII and put/call trailing-2y tails (reconstructed 95th percentile); CFTC and NAAIM placeholders remain missing_data.
Regime (5): Montgomery calendar windows, M2 vs IP/PPI drain, TRIN reflecting boundaries, 1987 structural/portfolio-insurance and Fed-hike complacency (mostly historical).
Alignment vs official legs (±5 calendar days) is stored per signal in the fixture. Highest published alignments in the source catalog include vbottom_no_retest and drawdown_exit_8pct; many volume/sentiment tails are frequent and therefore low-alignment by construction.
Prior berg-turning-points campaign used public-proxy families and was a negative/data-limited result. This import is the later reconstruction with verbatim quotes, stated dates, and computed events — complementary, not a silent replacement.
Mechanism and falsifiers
Not recorded in this iteration.
Not recorded in this iteration.
Confidence and limitations
These are reconstructions from public articles, transcripts, and stated dates — not the live MB Edge product.
Thresholds Berg published are used verbatim; unmarked tails are reconstructed and labeled as such in signal descriptions.
NYSE ADVN/DECN, CFTC, and NAAIM remain unavailable; those signals are missing_data or historical_only.
The 20-day active window on Other Indicators is a visualization convention except for the disclosed 8% drawdown exit.
Event dates are pinned from the 2026-08-24 miltonberg export; live re-fetch of TradingView/FRED series is out of scope for this import.
Not recorded in this iteration.
Compare expressions
Download evidenceExploratory results. Check each period, proxy and cost assumption before comparing. — means not recorded.
| Expression / family | CAGR | Sharpe | Max drawdown | Test period | Assessment |
|---|---|---|---|---|---|
| miltonberg.mb-edge-officialofficial | — | — | — | Not recorded |
exploratory |
| miltonberg.drawdown-exit-8pctprice pattern | — | — | — | Not recorded |
exploratory |
| miltonberg.ew-roc6-18-ratio105roc | — | — | — | Not recorded |
exploratory |
| miltonberg.vbottom-no-retestprice pattern | — | — | — | Not recorded |
exploratory |
Open an expression to inspect its rules and request confirmation. The request must be submitted by a trusted repository collaborator.
miltonberg.mb-edge-official
- variant id
miltonberg.mb-edge-official
- family id
official
- parameters
- round trips
55
- created by
source_baseline
- reason tested
published MB Edge legs as alignment ground truth
- status
exploratory
- tags
useful_as_filter
fidelity_mismatch
miltonberg.drawdown-exit-8pct
- variant id
miltonberg.drawdown-exit-8pct
- family id
price_pattern
- parameters
- drawdown
0.08
- created by
source_baseline
- reason tested
disclosed MB Edge sell rule
- status
exploratory
- tags
useful_as_exit
miltonberg.ew-roc6-18-ratio105
- variant id
miltonberg.ew-roc6-18-ratio105
- family id
roc
- parameters
- ratio
1.05
- fast
6
- slow
18
- created by
source_baseline
- reason tested
disclosed 17-occurrence ROC ratio buy
- status
exploratory
- tags
useful_as_entry
promising
miltonberg.vbottom-no-retest
- variant id
miltonberg.vbottom-no-retest
- family id
price_pattern
- parameters
- low lookback
60
- rebound
0.04
- window
5
- created by
source_baseline
- reason tested
four cited V-bottoms
- status
exploratory
- tags
useful_as_entry
structurally_interesting
Interactive lab
Interactive history unavailableLoading available evidence…
Research record
Source claims, inferred rules, experiments and the evidence behind the assessment.
Source
- kind
public reconstruction
- title
Milton Berg Signals — Reconstruction, Spec & Provenance
- links
research/miltonberg/
research/findings/milton-berg-signals.md
- prior experiments
berg-turning-points
What the source claims
MB Edge is a binary 100% S&P 500 or 100% T-bills model: act on the first buy signal and exit on an 8% closing drawdown. The published history has 55 round trips since 1957 (110 official legs).
Equal-weight 6-day/18-day ROC ratio >= 105% is a disclosed buy: 17 stated occurrences, low never broken, median +32.5% max 12-month gain.
V-bottoms: new 60-day low then >=+4% within 5 days without retest — cited 2009-03-09, 2018-12-24, 2020-03-23, 2025-04-08.
April 4 2025 cluster: VIX +10% for two days, NASDAQ 3-day ROC <= -7%, and related panic marks arriving together.
Named sentiment inputs (AAII, put/call, NAAIM, CFTC) have no published numeric tails; 95th-percentile reconstructions are marked as reconstructed thresholds.
Signals are event-dated, no lookahead, cluster-deduped. Viewer holding windows of 20 sessions are a site overlay, not a Berg exit except the 8% drawdown rule.
NYSE issue A/D counts, CFTC noncommercial extremes, and NAAIM exposure are not freely ingestible here; those rows stay missing_data or historical_only.
Rules actually disclosed
MB Edge is a binary 100% S&P 500 or 100% T-bills model: act on the first buy signal and exit on an 8% closing drawdown. The published history has 55 round trips since 1957 (110 official legs).
Equal-weight 6-day/18-day ROC ratio >= 105% is a disclosed buy: 17 stated occurrences, low never broken, median +32.5% max 12-month gain.
V-bottoms: new 60-day low then >=+4% within 5 days without retest — cited 2009-03-09, 2018-12-24, 2020-03-23, 2025-04-08.
April 4 2025 cluster: VIX +10% for two days, NASDAQ 3-day ROC <= -7%, and related panic marks arriving together.
Named sentiment inputs (AAII, put/call, NAAIM, CFTC) have no published numeric tails; 95th-percentile reconstructions are marked as reconstructed thresholds.
What had to be inferred
Signals are event-dated, no lookahead, cluster-deduped. Viewer holding windows of 20 sessions are a site overlay, not a Berg exit except the 8% drawdown rule.
Research questions
Which publicly disclosed Milton Berg buy, sell, and warning templates can be reconstructed as independent Other-indicators with verbatim provenance, without claiming the proprietary MB Edge model?
Data
- inputs
TVC:SPX
TVC:IXIC
TVC:VIX
USI:UVOL
USI:DVOL
CBOE put/call
AAII
- native berg feeds
False
- event fixture
research/miltonberg/events_fixture.json
Baseline implementation
- variant id
miltonberg.mb-edge-official
- family id
official
- parameters
- round trips
55
- created by
source_baseline
- reason tested
published MB Edge legs as alignment ground truth
- status
exploratory
- metrics
- label
ground_truth
- tags
useful_as_filter
fidelity_mismatch
What to try interactively
- name
holding_window
- type
integer
- default
20
- options
5
10
20
40
Suggested next research
Re-evaluate the computable subset against quantmodel market.sqlite when overlapping series exist.
Keep missing_data rows fail-closed until free A/D, CFTC, or NAAIM panels exist.
Trial ledger
- iteration number
1
- objective
Import reconstructed signals into Other indicators and document provenance on a Research Library page.
Next questions
Re-evaluate the computable subset against quantmodel market.sqlite when overlapping series exist.
Keep missing_data rows fail-closed until free A/D, CFTC, or NAAIM panels exist.
Review status
published. Research publication does not imply official admission.
Return to pending research