Milton Berg — Indicator Catalog (source notes)

---

name: Milton Berg indicator catalog — extreme-tail histogram methodology

description: Full catalog of Milton Berg's market-turning-point indicators from 2026-05-18 Shapiro interview — 30k+ indicators, ~2k tradeable combos, designed to be replicated in finance/indicators

type: reference

originSessionId: d08af12a-ffd1-4f37-8c52-4a5ad683e6e9

---

Milton Berg Indicator Catalog

Source: Jason Shapiro / Crowded Market Report interview, YouTube 68bnJkSZCzk, recorded 2026-05-18, Berg dating 2025-04 low signals.

Guest: Milton Berg — Milton Berg Advisors. Former: Richie Commodities (1978), Oppenheimer (80% cash before '87 crash), Michael Steinhardt, Soros, Druckenmiller (2000 top trade). 10 years of solo research culminating 2026.

Sites: miltonbergedge.com (retail, $10/mo), miltonberg.com (institutional), Twitter @bergmilton.

---

1. CORE METHODOLOGY

The histogram-tail principle

• Daily market action is assumed random in the middle of the distribution. Berg ignores mean/median behavior.
• All signals come from extreme tails of a histogram of the indicator's historical distribution.
• Rationale: extreme readings are non-random events; market turning points are non-random; therefore extremes correlate with turns.
• "Crossing a moving average happens all the time. That doesn't get us into a trade."

Combination logic — combos, not single indicators

• ~30,000 raw indicators tracked (e.g., "S&P up 4 days, up 5 days, up 6 days … up 10 days" = separate indicators each).
• ~2,000 tradeable combination models. Each combo = 3–4 simple extreme conditions firing on the same day.
• Each combo has historically signaled ~4 times = ~8,000 individual model dates total.
• A combo is only flagged when all conditions hit simultaneously — computer pops up the signal; otherwise silent.

Signal projection / stop framework

• For each combo, look at past instances → record max 1-year forward return and max drawdown.
• Projection: median of historical 1-year forward gains across instances.
• Stop: based on max historical drawdown of prior signals + ~50bp slack. Example: if max prior drawdown was 0.5%, use 1% stop below signal day. Many leverage stops are within 0.5% of trade date.
• Aggregation: when multiple combos fire in a window, take the median of medians as the composite projection.

Why momentum into a low/high is the key insight

• Conventional wisdom: momentum is bullish into highs, bearish into lows.
• Berg's framing: at the tail end of a bear market OR a bull market, momentum becomes extreme. So extreme positive momentum = exhaustion (top) OR thrust (bottom).
• Filter the difference with the 250-day MA: same momentum extreme means "low" if price < 250DMA, "top" if price >> 250DMA. The MA is a regime filter, not a crossover trigger.

Start date

• Models begin 1957-03-04 — date S&P expanded from 90 → 500 stocks. Earlier data excluded as non-comparable.
• Some indicators only begin in 2000 (require ETF/index instruments that didn't exist earlier).

---

2. INDIVIDUAL INDICATORS (the building blocks)

These are the atomic extremes Berg combines. Reproduce these first.

Volume regime

ID	Indicator	Threshold	Notes
V1	NDX 5-day volume rank	Highest in 200 days	Tail extreme, used at lows
V2	NYSE 5-day average volume rank	Highest in 375 days (≈1.5y) within past 20 days	High-volume regime — doesn't have to be day-of-low
V3	NASDAQ 10-day upside/downside volume thrust	≥ 1.89:1	Volume-thrust extreme
V4	NYSE upside volume as % of total 5-day volume	< 45%	Extreme downside volume regime (panic)
V5	Day-of-signal volume change	Up ≥ +20% vs prior day	Confirmation volume on rally day
V6	Day-of-signal volume change	+38% (observed example)	Stronger version of V5

Price momentum / rate-of-change

ID	Indicator	Threshold	Notes
P1	Nasdaq 3-day % change	≤ −7%	Panic decline momentum
P2	S&P 1-day % gain rank	Highest in 250 days	Thrust extreme
P3	Nasdaq 10-day % change rank	Highest in 180 trading days	Thrust
P4	S&P 500 10-day rate-of-change rank	Highest in 180 trading days	Thrust
P5	SOX 10-day % change	≥ +20%	Semis parabolic
P6	S&P decline depth at signal-day low	≥ −5% to −13% (multiple combo variants)	Drawdown filter
P7	S&P recovery from prior 10% decline	≥ +6% into new recovery high	Bottoming thrust
P8	Single-day gain on April 9 type	≥ +9% S&P or Nasdaq	Capitulation reversal
P9	Upside/downside volume ratio	100:1 on S&P 500 day	Generational breadth thrust

Streak / consecutive-day counts

ID	Indicator	Threshold	Notes
S1	VIX consecutive up days	≥ 10 (0.01% historical)	Extreme
S2	VIX consecutive down days	≥ 10 (0.04% historical)	Extreme
S3	S&P 500 up days in a row	≥ 9–11 (since 1928)	Tail
S4	S&P 500 down days in a row	≥ 9–11	Tail
S5	Russell up days in last 9	≥ 8 of 9	Breadth momentum
S6	Nasdaq up days in last 19	≥ 16 of 19	After 10% decline + 19-day hold

Breadth / A-D / new highs-lows

ID	Indicator	Threshold	Notes
B1	NDR Multi-Cap A/D line 7-day ratio	≥ 2.10:1	Index ~1,400 stocks
B2	NASDAQ 10-day A/D line ratio	≥ 1.30:1
B3	Reverse breadth thrust	A/D line spike to downside	Companion to V4; capitulation flush
B4	NASDAQ 5-day new lows − new highs	≤ −16	Cumulative deterioration extreme
B5	NYSE new lows, market-cap-weighted	% of total market cap of NYSE, not raw count	Critical Berg innovation — see §3
B6	NYSE 60-day new low test	S&P at new ≥60-day low	Capitulation filter
B7	30-day new closing high	S&P recovery high on signal day	Confirmation

Volatility extremes

ID	Indicator	Threshold	Notes
VX1	VIX intraday spread	trades between 45 and 60 in a session	Panic vol regime
VX2	VIX standard deviation (5-day) over 20-day	ratio < 0.87	Extreme downward VIX compression
VX3	VIX absolute level above 45	rare — "above 45 is high"	Standalone capitulation

Sentiment / positioning

ID	Indicator	Threshold	Notes
SE1	IPO dollar value	4-year high	Conventionally bearish, history shows mixed
SE2	Market Vane bullish %	≥ 1997 levels	Conventionally bearish, history shows OK
SE3	Margin debt / GDP	record high	Conventionally bearish, history shows OK
SE4	NDR overbought/oversold oscillator	< 10 (e.g. 4.44) on 0–100 scale	Genuine extreme
SE5	COT positioning	not extremely long	Shapiro's domain; Berg cites as confirmation
SE6	US Investing Championship % positive	only 16% positive in 2026 = participation proxy	Sentiment via P&L distribution

Trend / regime filter

ID	Indicator	Threshold	Notes
T1	250-day moving average	price below = downtrend regime (use to validate "low" signals); price above = uptrend regime	Used as filter, NEVER as crossover trigger
T2	"Hold the low" day count	days since prior swing low without violating it; common filter values 1, 2, 3, 4, 7, 9, 19	The longer the hold + new recovery high = stronger bottom

---

3. THE MARKET-CAP-WEIGHTED NEW LOWS INSIGHT (B5)

Berg's most actionable original innovation, worth its own section.

Problem: "Many new lows at all-time highs in index" — cited as bearish (1929, 1987, 2000).

Berg's correction: raw count of new lows is meaningless when the index is dominated by trillion-dollar names and the new lows are half-billion-dollar companies.

Fix: Compute new_lows_market_cap_pct = sum(market_cap of NYSE 52w-low stocks) / sum(market_cap of NYSE universe).

Calibration table (Berg's data):

• 1979: 4.38% — no crash
• 1980: 4.15% — no crash
• 1983: 3.90% — no crash
• 2026-05 reading: ~3.12% — not at extreme

Max forward drawdown in those historical readings: 11%, 14%, 4%, 2%, 8%, 6%, 1%, 1%, 0.6%, 1%. A couple of bear markets but no crashes.

Implementation: join raw_prices (we already have NDX/MAG7 + index members) with shares outstanding → market cap → flag 52w lows → sum % of total. Run daily. Add to dashboard.

---

4. COMBO MODELS — EXACT RECIPES BERG SHOWED

Combo C1 — "1957-style turn" (used Oct 21 1957, June 26 1962, Sept 22 1976 bear-rally, Dec 24 2018, April 8 2025)

All on same day:

1. V2: NYSE 5-day average volume highest in 375 days anytime in past 20 days
2. B6: S&P at ≥60-day new low
3. P6: S&P down ≥10–12% from prior high
4. V4: NYSE 5-day upside volume < 45% of total
5. B3: Reverse breadth thrust (A/D spike down)

Combo C2 — April 4 2025 (precedents: 1998, 2011)

1. V1: NDX 5-day volume at 200-day high
2. P1: Nasdaq down 7%+ in 3 days
3. VX1: VIX between 45–60

Combo C3 — April 7 2025 (precedents: 1998, 2018, 2020)

1. V2: NYSE 5-day average volume highest in 375 days
2. SE4: NDR overbought/oversold < 10 (was 4.44)
3. B4: NASDAQ 5-day new lows − new highs ≤ −16

Combo C4 — April 9 2025 "thrust day" (precedents: 1979, 1982, 1984, 2003 ×2, April 9 2025)

1. T1: S&P below 250DMA (downtrend filter)
2. P7: S&P up ≥6% from prior 10% decline into new recovery high
3. P2: S&P highest 1-day gain in 250 days
4. V5: Volume up ≥20% on the day

Combo C5 — April 30 2025 "held-low-7-days" (precedents: 1974, 1998, 2002 Oct, 2009 Mar, 2011, 2025 Apr)

1. T2: S&P down ≥5% but held its low for 7 days
2. P7 variant: S&P up ≥9% into new recovery high
3. B1: NDR Multi-Cap 7-day A/D ratio ≥ 2.10:1

Combo C6 — April 13 2026 "held-low-9-days + thrust" (precedents: Aug 25 1982, Aug 6 1984, Oct 21 1998, Oct 14 2011)

1. T2: S&P down 9% held low for 9 days
2. B7: S&P closed at 30-day new recovery high
3. P3: Nasdaq greatest 10-day gain in 180 days
4. P4: S&P greatest 10-day rate-of-change in 180 days

Historical: 100% of time up 120 days later across all precedents.

Combo C7 — April 13 2026 variant w/ Russell streak (precedents: Aug 23 1982, Oct 21 1998)

Same as C6, plus:

1. S5: Russell up 8 of 9 days

Combo C8 — April 13 2026 variant w/ Nasdaq volume thrust (precedent: Oct 14 2011 only)

1. T2: S&P down 13% held low 9 days
2. V3: NASDAQ 10-day upside/downside volume thrust ≥ 1.89:1
3. B2: NASDAQ 10-day A/D ratio ≥ 1.30:1

Combo C9 — April 17 "SOX parabolic" (precedents: Oct 4 1987 → 28% before crash; May 5 1997)

1. P5: SOX 10-day rate-of-change ≥ +20%
2. B7 variant: S&P at 2-year closing high

Never been bearish. Used as continuation, not turning point.

Combo C10 — April 27 "late entry, 16 of 19 days" (precedents: Nov 4 1985 → +76%, Oct 30 1986 → +42% before '87 blow-off, Nov 22 2023 → +34%)

1. P6 + T2: Nasdaq down 10% and held low for 19 days
2. S6: Nasdaq up 16 of those 19 days

Continuation signal; max drawdowns post-signal were 0.22% and 0.16% — extremely tight stop possible.

Combo C11 — "4-year IPO + VIX compression" (precedents: Nov 21 2014 = top region, May 29 2020, June 5 2020, May 2026)

1. SE1: 4-year high in dollar value of IPOs
2. VX2: VIX 5d/20d stdev ratio < 0.87

Berg doesn't trade on this — but cites to refute Twitter "bearish" arguments.

---

5. TRADE MANAGEMENT (the wrapper around the indicators)

Retail model rules (miltonbergedge.com)

• Entry: buy 100% S&P at close of first combo signal, no second-guessing, no confirmation wait.
• Exit: trailing stop, exit at S&P down −8% from highest close since entry.
• Reentry: wait for next combo signal.
• Cash leg: T-bills while flat.
• Frequency: ~1 round trip per year on average.
• Historical performance 1957–2026: 18.5% per annum, 91% of buy signals profitable, 95.5% of all trades profitable, 80.9% time in S&P / 19.1% T-bills, max loss per trade < 4%.

Institutional rules

• Each combo signal carries its own stop derived from the worst historical post-signal drawdown of that combo (+ ~50bp).
• Leverage stops can be as tight as 0.5% below signal day.
• Pyramid up as additional combos fire confirming the trend.
• Use the most-momentum index for the long (SOX, OEX, QQQ), use S&P futures only when going leverage (lowest vol).
• Current position (May 2026): 175% long.

Current outlook from Berg (snapshot 2026-05-18)

• Median projection from combined April 2026 signals: S&P 8944.29 within 12 months (vs ~7500 at recording). ~+19%.
• Bullish: short covering still driving (USIC 16% positive = retail not really long).
• Indifferent: high IPO count, low VIX, high margin/GDP — all historically not crash signals.
• Inflation framing: bullish for stocks until the Fed actually fights it; don't preempt.

---

6. REPLICATION PLAN FOR JARVIS (proposed)

Phase 1 — atomic indicators (build first, store as time series)

Data already in finance/research.db and /var/lib/jarvis/data/raw_prices:

• V1, V2, V3, V5: need NYSE/Nasdaq volume — we have SPY/QQQ proxies in raw_prices; for true NYSE/Nasdaq composite volume need to add a data source.
• P1–P9: pure price math from raw_prices. Easy first wins.
• S1–S6: trivial streak counters.
• B5 (market-cap-weighted new lows): needs shares-outstanding × close for NYSE universe. High-value original. Use yfinance for shares outstanding daily.
• VX1–VX3: VIX data already in svix_compression table.
• SE4 (overbought/oversold oscillator): replicable as %B / Williams / Stoch — emulate NDR.
• T1, T2: trivial.

Phase 2 — combo engine

• Each combo C1–C11 = boolean AND of atomic conditions on the same day.
• For each combo, find historical instances 1957→present, log forward 5/10/15/20/120-day returns and max drawdown.
• Output: combo_signals table (date, combo_id, atomic_conditions_met, forward_returns, max_dd, stop_price).

Phase 3 — daily scanner

• Cron 4:30PM ET weekdays.
• Output: dashboard panel "Berg signals fired today" with combo ID, prior precedents, median projection, suggested stop.

Phase 4 — backtest the retail rule

• Long-or-flat S&P, buy on first signal of any combo, trail −8%, T-bills otherwise.
• Compare 1957–2026 CAGR to Berg's claimed 18.5%.

Open data gaps

• True NYSE composite volume (not SPY proxy) — Polygon should serve.
• NDR-equivalent multi-cap index → use Russell 3000 or Wilshire 5000 A/D.
• US Investing Championship % positive — manual scrape; quarterly.
• Market Vane bullish % — paid service; substitute with Investors Intelligence or AAII.

---

7. ATTRIBUTION TO PRIOR LESSONS

• Programmatic First (CLAUDE.md): Berg explicitly says "I couldn't use AI to do it because AI doesn't understand the nuances of markets" — pure rule-based extreme detection. Aligned with our Karpathy principle.
• Always Be Testing: Berg backtests every combo against 1957→present and only acts when the historical precedent count is ≥1 and direction is consistent. Aligned.
• Walk-forward / simpler models generalize (recent lesson 7h ago): Berg uses 3–4 simple conditions per combo, no fitted parameters beyond thresholds — consistent with our walk-forward result that DEMA/TEMA overfit.

---

8. NEXT ACTIONS (queued, not yet executed)

1. P1 — Build atomic indicator V1–V5, P1–P9, S1–S6 in finance/indicators/berg_atoms.py. Pure math from raw_prices. ~half-day.
2. P1 — Build B5 (market-cap-weighted new lows). Needs daily shares outstanding for NYSE universe. Backtest 1990→present where shares-out is reliable.
3. P2 — Build combo engine (finance/indicators/berg_combos.py) wiring atoms → C1–C11. Generate historical signal log.
4. P2 — Add dashboard panel /finance/berg-signals.
5. P3 — Backtest retail rule, publish CAGR vs SPY.
6. P3 — Scrape miltonbergedge.com periodically for any new combo recipes Berg publishes (he says he holds proprietary, but blog posts may reveal).

---

9. EXPANDED CATALOG FROM X / TWITTER & SECONDARY SOURCES (2026-05-25)

Following deep search across @bergmilton X posts + secondary appearances (Forward Guidance, prior Shapiro shows, Real Vision). Reorders by implementation confidence.

9.1 New HIGH-confidence atoms (pure math, threshold explicit)

ID	Indicator	Threshold	Source	Confidence
P_ROC74	S&P 5-day rate-of-change ≥ +7.4%	"Every bull market since 1928 began with a 5-day move ≥ 7.4%" — Forward Guidance Feb 2024	0.90 HIGH
T_RIN_HI	NYSE TRIN intraday extreme	TRIN ≥ 12.5 (capitulation) or ≥ 15.5 (rare panic) — 2020-03-12 was 15.50, Oct 2008 was 12.52	0.85 HIGH
B_DA_SP600_50_1	S&P 600 daily decliners-to-advancers	≥ 50:1 D/A intraday — Berg X 2026: "18 cases since 1995, 100% higher 120 days later"	0.80 HIGH
B_AD_5D_087	NYSE 5-day A/D ratio ≤ 0.87 at index ATH (within 5 days of all-time high)	"Breadth divergence at peak" — combo precondition Berg cites for tops	0.75 MEDIUM-HIGH
VX_VIX_5_30	VIX 5-day average / 30-day average	< 0.85 (compression at top) or > 1.50 (expansion at bottom)	0.70 MEDIUM

9.2 Macro overlays Berg cites (lower confidence — context, not triggers)

• Edson Gould "Two Tumbles and a Jump" — 2 consecutive Fed funds cuts (after a tightening cycle) historically bullish; Berg uses as regime overlay. Confidence: 0.60 MEDIUM. Implementation: trivial Fed funds delta.
• Montgomery cycles — Berg has referenced but doesn't publish formula; appears to be a fixed-period cycle overlay. Confidence: 0.25 SPECULATIVE — defer.
• Argentina M2 momentum — Berg X 2026: cites Argentina monetary regime change as analog for emerging-market bottoms. Not a quant atom; macro narrative.

9.3 Corrections to Section 5 retail stats

The 18.5% CAGR claim from the Shapiro video reconciles with Berg's X-published track record:

• 55 round-trips over 1957-2026 (≈ 1 every 15 months, NOT "1/year" — bursty)
• 90.9% profitable trades (not 95.5% as I'd transcribed)
• 21.7% average gain per winning trade
• Time in market ~78%, T-bills ~22%
• Max drawdown per trade < 4%, longest losing streak = 2 consecutive losses
• Compounded result: ~18.5% CAGR vs SPY 1957→2026 ~10.2% CAGR

9.4 Indicators Berg DISCUSSED but did NOT specify formula (defer)

• "Specialist short ratio" extremes — proprietary 1970s NYSE data, no longer published
• "Member firm short selling" — same data issue
• "Mutual fund cash positions" — ICI data exists but Berg's threshold not stated
• "Insider buying clusters" — Berg cites Vickers data but no replicable threshold

9.5 Implementation ordering by confidence (HIGH-conf first)

1. P_ROC74 (5d ROC ≥7.4%) — trivial pct_change; data: SPX
2. T_RIN_HI — needs TRIN feed (NYSE advances/declines × volume ratio) → Polygon
3. B_DA_SP600_50_1 — needs SP600 daily A/D → Polygon mid-cap breadth
4. B_AD_5D_087 — needs NYSE 5d A/D ratio + SPX ATH detector → mostly raw_prices, A/D from Polygon
5. VX_VIX_5_30 — VIX already in svix_compression table → immediate

9.6 Cross-references

• B_AD_5D_087 dovetails with our existing bubble-bounds indicator (Pomorski paper): both flag "ATH with deteriorating internals." Composite signal candidate.
• P_ROC74 competes with breakout-academy ATR filter (lesson 2026-05-24): ATR(5)>ATR(20) HURT daily Sharpe. Berg's 7.4% threshold is roughly equivalent to a 2-σ daily move — re-check whether the level vs ratio framing matters.
• T_RIN_HI is a tape-reading classic; check feedback_pipeline_first_for_resilience.md for prior-art pipeline pattern when wiring the TRIN feed.

---

10. PHASE 2 SHIP — 2026-05-25 (combo engine live)

Code shipped

• finance/indicators/berg_atoms.py — 25 atoms (24 prior + P6_HIST)
• finance/indicators/berg_combos.py — all 11 combos C1–C11 wired
• finance/scripts/berg_gate_g2.py — Gate G2 backtest harness
• finance/tests/test_berg_combos.py — 16 tests; 49/49 green with atoms suite

Gate G2 result — PASS 7/7 = 100% on testable precedents

combo	live?	matches	notes
C2	LIVE	2/2 (1998-08-31, 2011-08-08 proxies)	NDX panic + VIX 45–60
C4	LIVE	1/1 (2003 thrust day)	April thrust
C6	LIVE	2/2 (2011-10-14, 2026-04-13)	held-low-9 + 4-way thrust
C10	LIVE	2/2 (2023-11-22, 2026-04-27)	late entry 16-of-19 (after P6_HIST fix)
C1, C3, C5, C7, C8, C9, C11	BLOCKED	n/a	missing data feeds (NYSE breadth, SOX, NDR, IPO calendar)

P6_HIST atom — semantic correction

Original P6 (p6_drawdown) required current-bar drawdown ≤ threshold. But Berg's verbal phrasing "the market DECLINED 10%, held the low for 19 days" describes a historical regime: by the time the recovery thrust fires (S6: 16-of-19 up days), price has recovered past the −10% line. Without this fix, C10 missed all 3 testable precedents.

New atom: p6_drawdown_reached_within(prices, threshold=-0.10, lookback_days=60) — true if any close within the lookback window reached the drawdown threshold. Confidence 0.90 HIGH.

C10 now uses P6_HIST@-10%_60d_NDX + T2_held@19d_NDX + S6_NDX_16of19.

Coverage caveat

26 of 33 named precedents are pre-2003 (research.db starts ~2003). G2 PASS rate based on the 7 in-range cases. Phase 0 backfill (1957→2003) is the next high-leverage move to validate the long tail.

Date provenance

Catalog dates for C6/C7/C10 are 2026 (verbatim from Berg 2026-05-18 interview, dating live-as-of-recording signals). Earlier text in the catalog correctly shows 2026 throughout §4. An interim mistaken edit to 2025 was reverted after diagnostic verified 2026-04-13/2026-04-27 in our price data.

Verdict

Phase 2 SHIPPED. G2 authorizes Phase 3 (dashboard + cron) for the 4 live combos. Phase 0 (data backfill) recommended in parallel to unlock blocked combos and validate the 26 pre-2003 precedents.

---

11. EXTRAS — 2026-05-28 deep research (vetted)

Vetted from background agent (general-purpose + WebSearch). Raw at finance/research/berg_extras_raw.md. Each item confidence-rated.

11.1 CRITICAL CORRECTIONS to prior catalog

• Indicator universe (UPDATED): 30,000 indicators daily → 1,346 distinct individual indicators + 816 distinct buy-signal configurations + ~2,000 tradeable models. Prior "30k / 2k" framing collapsed indicator count with model count.
• Retail model exit (UPDATED): NOT a fixed −8% trailing stop. Per miltonbergedge.com/faq, model "lacks fixed percentage decline triggers — has ability to adjust the exit level based upon conditions." The "8%" in Soar Financially summary is journalist simplification. Section 5 "trailing stop −8%" entry is superseded.
• Performance figures (CONFIRMED): 18.5% CAGR (1957→2026), 90.9% trade win-rate, 21.7% avg gain per winner, 78%/22% time in market vs T-bills, max DD per trade <4%. 55 round-trips / 68 years = ~1 every 15 months (bursty, NOT 1/year).

11.2 MISATTRIBUTION WARNING (negative finding)

Per Trading Flip transcript synthesis + primary source check, the following are NOT in Berg's actual public framework despite frequent third-party tweet attribution. Treat any tweet attaching these to Berg as misattribution unless verified from primary source:

• Specialist short ratio extremes
• Member firm short selling
• Vickers insider buying
• ICI mutual fund cash positions
• Hindenburg Omen (as a Berg signal)
• Zweig breadth thrust by name (Berg uses thrust concepts but not the Zweig label)
• 90% up/down days by name (Berg uses volume thrust but not the Lowry 90% framing)

11.3 NEW HIGH-confidence atoms (verified from primary)

ID	Indicator	Threshold	Source	Confidence
V_NYSE_5D_375	NYSE 5-day total volume — greatest in 375+ days	≥375d lookback breach	miltonberg.com home	HIGH
V_5D_VS_90D_DEV	5-day avg volume / 90-day avg deviation	crosses +50% AND +100% at bull starts	Wealthion 2023	HIGH
V_NYSE_UD_9_1	NYSE up:down volume ratio, 2 consec sessions	≥9.0 for 2 consecutive days	Wealthion 2023	HIGH
B_NHNL_60D_30	60-day net new highs / (NH+NL) at bull start	≥30% net	Wealthion 2023	HIGH
B_SP500_PCT_BELOW_20DMA	% of S&P 500 stocks below 20DMA	≥90% (oversold bottom)	Wealthion 2023	HIGH
B_SP600_PCT_BELOW_10DMA	% of S&P 600 stocks below 10DMA	≥95%	miltonberg.com	HIGH
VX_VIX_3D_35	VIX 3-day rate of change	≥+35% in 3 days	miltonberg.com	HIGH
T_RIN_15_50	TRIN extreme — historic high (Jan 8 1988 = 15.50)	TRIN ≥ 15.50	Trading Flip transcript	HIGH (factual)
T_RIN_12_52	TRIN — Oct 2011 panic bottom (12.52)	TRIN ≥ 12.52	Trading Flip transcript	HIGH (factual)
P_5D_ROC_74_CONFIRM	5d ROC ≥+7.4% bull confirmation. Precedents: 1975, 1978, 1980, 1987, 1991, 2003, 2009, 2011, 2020	≥+7.4% on SPX	Wealthion 2023 (full precedent set)	HIGH
P_5D_ROC_LACKING	Counter-cases: 6.68% (Jun '22), 6.36% (Oct '22) failed bull-confirm	<7.4%	Wealthion 2023	HIGH

11.4 NEW MEDIUM-confidence atoms

ID	Indicator	Threshold	Source	Confidence
V_NASDAQ_RECORD	NASDAQ single-day vol record (May 17 2024)	New ATH	miltonberg.com (penny-stock distorted post-2020 — data quality flag)	MEDIUM
V_5D_NETUD / V_10D_NETUD / V_12D_NETUD	5d/10d/12d net up:down vol extremes	"Extreme" (no number)	Trading Flip / IP transcript	MEDIUM
SE_PC_INTRADAY_HIST	Highest intraday put/call ratio in history	"Record"	Dec 2018 cited	HIGH event / MEDIUM number
VX_VIX_5_30_RATIO	Short-term VIX (5-8d) / longer-term (30-40d) — extreme top trigger	"Extreme"	Wealthion 2023 (Apr 21 2023 trigger)	MEDIUM
VX_VIX_RATIO_OUTCOME	Median 1y SPX decline AFTER VIX-ratio top: −23.43%; max gain in interim: +2.72% over 17d	Backtest claim	Wealthion 2023	HIGH (Berg's own backtest)
MAC_M2_VS_IP_COMM	M2 growth vs (IP + commodity prices) extreme slowdown	Qualitative	WisdomTree 2021 Wharton podcast	HIGH (frequency: ~80% of 10%+ drops since 1960)
MAC_FED_2_EASE_GOULD	Edson Gould "Two Tumbles and a Jump" — 2 consec Fed cuts (15th since 1958 fired Nov 12 2024)	2 consec cuts	@BergMilton tweet 1856393046599905395	HIGH
MAC_GOLD_GDP_WT	GDP-weighted gold breakout above 2020 highs (target +40-60%, ~$2,500-$3,000)	Breakout	Wealthion 2023	MEDIUM-HIGH
P_R2K_219D_4_99	Russell 2000 219 days from low up only 4.99% (median historic = 48.52%)	Lag vs 48.52% median	Wealthion 2023	HIGH (factual)
P_2GAP_OFF_LOW	Two consecutive unfilled upside gaps Days 1+2 off intraday low	Pattern	@BergMilton 1976313300628668429 (MERVAL example)	HIGH
SE_II_NOV21_MATCH	Investors Intelligence bull% matches Nov 2021 top reading	Match prior top	Wealthion 2023	HIGH

11.5 LOW-confidence / SPECULATIVE (do not implement before verification)

• MAC_GANN_90Y — Gann 90-year cycle (1932→2022). Post-hoc fit; LOW.
• P_ISLAND_REVERSAL_R2K — Adam Taggart Feb 2026 third-party attribution; pattern criteria not public. MEDIUM-but-flag.
• P_CLIMAX_TOP_TRIPLETOP — GDP-weighted gold triple top. MEDIUM, qualitative.

11.6 NEW combo recipes (additive to C1-C11)

Combo	Conditions	Source	Confidence
C_BULL_CONFIRM_FULL	(1) P_5D_ROC_74_CONFIRM AND (2) V_5D_VS_90D_DEV crosses +50% AND +100% AND (3) V_NYSE_UD_9_1 (9:1 on 2 consec) AND (4) B_SP500_PCT_BELOW_20DMA ≥90% AND (5) B_NHNL_60D_30 ≥30%. Failed Jun/Oct 2022 (only 2/5 met).	Wealthion 2023	HIGH
C_DEC18_BTM	(a) SE_PC_INTRADAY_HIST + (b) V_5D_NETUD extreme + (c) B_NHNL extreme + (d) day-count D+1/+2/+3	Dec 2018 low	HIGH
C_TWO_TUMBLES	MAC_FED_2_EASE_GOULD = 2 consec Fed cuts (standalone macro overlay)	@BergMilton	HIGH
C_SICK_MKT_EXHAUSTIVE	Upside gaps + heavy vol into new highs + climax in lead names (NVDA, SMCI) + banking divergence + non-bullish momentum + Europe/Japan new highs	Forward Guidance Feb 28 2024	MEDIUM (failed — pattern critique below)
C_8REASONS_DOUBT_BULL	Banking-index strange + rates deflationary + Oct '22 low divergence + China outlook (incomplete list)	Forward Guidance Sep 5 2023	LOW
C_INSTITUTIONAL_SHORT_2026	"Disturbing" Dec 2025 vol signal + R2K island reversal + gold/silver climax top + BTC speculative peak	Adam Taggart Feb 2026	MEDIUM

11.7 Pattern critique (operational guidance)

Berg's bottom-spotting > top-spotting. Track record:

• ✅ Strong: Dec 2018 (C_DEC18_BTM), Jan 26 2023 ("bear ended months ago"), Apr 4 2025 (54 indications fired; targets 6915-7320). Edge retail product is bottom-finding.
• ❌ Weak: Aug 23 2017 Bloomberg "30-year bear" → SPX +14% after; not vindicated through 2026. Feb 28 2024 "Sick Market" → market continued higher into Apr 2025. Sep 5 2023 "Stock Market In Trouble" → ranged then resumed higher. Mar 9 2020 "Fed's ammo doesn't work" → published days before literal bottom.
• Pattern: long-standing super-cycle bear bias → repeatedly early/wrong top calls.
• Operational rule for our model: wire Berg bottom signals FIRST. Treat top signals as confirmation only, not standalone triggers.

11.8 Prior-era anchors (newly documented)

• Oct 1987 trigger: Berg attributes crash to single Fed hike of +0.25% on Sep 4 1987 (first hike of cycle). Aug 11 1987 spike-volume signal as precursor. Oppenheimer position: 80% cash by Sep 1987.
• Edge model start date = Oct 1957 (corrects prior catalog's "Mar 4 1957"; both reference S&P 500 expansion). Per miltonbergedge.com/newsletter.

11.9 Outcome statistics (vetted)

• 5-day breadth thrust precedent outcomes: +20% to +47% within 12 months.
• April 4 2025 buy trigger: 54 bullish indications fired 4/9 → early May; projected median max 12-month SPX target 6,915.91 – 7,320.01 (Recapio summary of David Lin interview).

11.10 Pending actions (post-extras integration)

1. Cross-check new atoms vs existing berg_atoms.py registry — V_NYSE_5D_375 may overlap V2 (verify lookback semantics); P_5D_ROC_74_CONFIRM extends P_ROC74 with precedent set.
2. Plan-only sketch for adding new atoms to berg_atoms.py BEFORE coding (Rob approval gate).
3. Wire C_BULL_CONFIRM_FULL + C_TWO_TUMBLES as Phase 3 combo additions (gated by data availability — NYSE up:down vol + S&P 500 breadth feeds).
4. Backtest MAC_M2_VS_IP_COMM "80% of 10%+ drops since 1960" claim independently before relying on it as a filter.
5. Update finance/indicators/berg_combos.py Phase 2 status: keep dynamic-exit-rule note, suppress legacy "-8% trailing stop" semantics in Phase 4 retail-rule backtest spec.

11.11 Source URLs (vetted, raw at finance/research/berg_extras_raw.md)

Primary fully verified:

• https://www.miltonberg.com/ — institutional home
• https://miltonbergedge.com/faq/ — dynamic exit confirmation
• https://miltonbergedge.com/newsletter/ — Edge start Oct 1957
• https://wealthion.com/news/bear-market-to-get-far-worse-than-any-in-recent-memory-milton-berg — densest source for new atoms
• https://tradingflip.com/%F0%9F%94%B4-how-to-time-the-market-w-milton-berg/ — TRIN history + misattribution exclusions
• https://www.wisdomtree.com/investments/blog/2021/08/11/conversations-with-a-market-bear — M2/IP/commodity overlay
• https://x.com/BergMilton/status/1856393046599905395 — Two Tumbles 15th since 1958
• https://x.com/BergMilton/status/1976313300628668429 — MERVAL pattern
• https://recapio.com/digest/analyst-called-market-bottom-now-up-38-reveals-2026-outlook-milton-berg-by-david-lin