fxcryptobots
☰
Base
📊 Rankings 📅 Data Releases 🤖 Trading Bots 📈 Paper Trading BETA 🏦 Recommended Brokers NEW 📑 Research Reports NEW
Pricing Login Get Started
Research Report · Twitter Alpha

Zephyr (@zephyr_z9) Backtest v3: 35 BULL Picks vs XSD Sector ETF

A strict-sentiment, first-mention-hold backtest of every BULL stock Zephyr (@zephyr_z9, 146K followers, "AI & Chips") posted on X between Dec 30, 2023 and May 18, 2026. 136 valid mentions across 35 unique BULL tickers, equal-weighted, held to today, benchmarked against XSD (S&P Semiconductor ETF) and SPY over identical windows.
Published 2026-05-18 Period: Dec 30, 2023 → May 18, 2026 870 days · 2.4 years Methodology v3 (explicit BEAR only)
VERDICT: SELECTIVE FOLLOW · net alpha +13.17% vs XSD, +63.59% vs SPY
35
Unique BULL tickers
136
Valid mentions (125 BULL / 11 BEAR)
128d
Avg hold duration
+74.98%
Mean hold return
+13.2%
Mean alpha vs XSD
37.1%
XSD alpha win rate
+63.6%
Mean alpha vs SPY
80.0%
Raw win rate
Prefer the print version? Download the full 6-page PDF with every table, methodology notes, and limitations. Same data, paginated for offline reading and sharing with your trading desk.
Download PDF ↓

Headline Findings What we found when we held every BULL pick to today.

The equal-weighted Zephyr BULL portfolio beat SPY by 63 points and beat XSD by 13 points (mean).

Buying every BULL pick on the day of its first bullish mention — at equal weight, holding to today — returned a mean of +74.98% (median +27.37%). Over the same period, simply buying XSD on the same entry dates returned +61.81%, and SPY returned +11.39%. Net alpha: +13.17% vs XSD, +63.59% vs SPY. Raw win rate: 80%.

This makes Zephyr the only account in our v3 series with a positive mean alpha against its own sector ETF.

But the mean is carried by a handful of memory winners. Median alpha vs XSD is −18.81%.

Alpha win rate vs XSD is only 37.1% — 63% of BULL picks underperformed the semi sector. The positive mean comes from 3 super-winners that drag the average up: SNDK +1,145% (alpha +1,071), MU +283 (alpha +217), WDC +210 (alpha +154). Strip those out and the alpha distribution looks like everyone else: noisy, slightly negative.

Memory supercycle = real skill.

Zephyr's flagship thesis — memory / storage / advanced packaging — produced the entire alpha stack: SNDK +1,145% (alpha +1,071, 157d), MU +283% (alpha +217, 152d), WDC +210% (alpha +154, 135d), LITE +154% (alpha +88, 100d), ASX +116% (alpha +50, 105d). Early and correct calls, repeated over many tweets. This is the domain to follow.

Mega-cap and China ADR picks dragged alpha ≥100 points negative.

MSFT −135 alpha (595d), META −135 (445d), BABA −96 (223d), PSIX −110 (134d), CRWV −75 (134d), IREN −74 (153d). Zephyr opines on the megas but produces no edge there — XSD or SPY does the same job standalone. The AI compute pure-plays (IREN / CRWV / NBIS) are the weakest single category.

Per-Mention Horizon Returns Each mention's tweet day → D+1 open entry, exit at 1 / 5 / 30 / 90 trading days. Alpha = mention return minus same-period XSD/SPY return.

SentimentHorizonnRaw meanRaw WRα XSD meanα XSD WRα SPY meanSharpe
BULL1d125+0.59%60.0%+0.17%48.0%+0.56%+2.58
BULL5d116+3.27%65.5%−1.14%42.2%+2.28%+2.33
BULL30d46−0.46%43.5%−0.10%43.5%−1.67%−0.04
BULL90d46+19.54%41.3%+11.80%30.4%+17.25%+0.40
BEAR1d11+2.96%36.4%+2.57%45.5%+2.90%+5.24
BEAR5d11+1.01%54.5%+4.39%45.5%+1.99%+0.54
BEAR30d9−7.79%33.3%−4.82%33.3%−7.57%−1.20
BEAR90d9−7.93%22.2%−9.66%22.2%−6.63%−0.67

BULL 1–5 day windows show small positive raw returns and roughly flat sector alpha. The 90-day BULL horizon delivers mean +19.5% raw / +11.8% alpha vs XSD, but the win rate is only 30% — a few super-winners (SNDK, MU) are pulling the average. BEAR sample is small (n=11) but directionally correct: 30d/90d windows return −5% to −10% alpha. Statistical significance is low; treat as a signal, not a proof.

First-Mention Hold — Aggregate Stats Buy each unique BULL ticker on its first BULL mention day, hold to May 18, 2026 close.

MetricValue
BULL pick count (unique tickers)35
Mean hold duration128 days (4.3 months)
Median hold duration134 days
Mean hold return (raw)+74.98%
Median hold return+27.37%
Win rate (return > 0)80.0%
Mean alpha vs XSD (semi ETF)+13.17%
Median alpha vs XSD−18.81%
Win rate alpha XSD > 037.1%
Mean alpha vs SPY (broad)+63.59%
Win rate alpha SPY > 068.6%
3×+ winners (>200% return)3
50%+ losers (<−50% return)1
Beat sector by 100%+ alpha3
Trailed sector by 100%+ alpha3

Top 15 Winners Entry on first BULL mention day → close of May 18, 2026. # BULL = total BULL mentions for the same ticker.

TickerFirst BULLDaysEntry $Returnα vs XSDα vs SPYMax DD# BULL
SNDK2025-10-01157113.06+1,145.01%+1,071.08%+1,133.55%−31.34%5
MU2025-10-08152188.99+283.44%+217.14%+273.16%−30.31%7
WDC2025-10-31135155.50+209.98%+154.14%+202.08%−20.59%1
LITE2025-12-22100382.65+153.68%+88.29%+145.60%−28.70%1
GOOG2024-08-07445163.24+140.95%−15.78%+101.08%−29.43%1
ASX2025-12-1510515.65+116.04%+50.36%+108.25%−16.81%1
AMD2025-11-20121232.70+82.25%+4.21%+72.41%−26.47%9
INTC2026-04-152363.77+70.57%+29.43%+64.25%−15.97%4
NBIS2025-11-03134134.00+64.13%+9.37%+56.33%−38.68%1
WOLF2026-05-05938.69+60.58%+51.53%+58.17%−11.19%2
QCOM2026-04-1325128.00+57.41%+8.95%+48.30%−15.77%3
TSM2025-10-08152295.00+37.07%−29.23%+26.79%−18.37%9
CSCO2026-04-211987.92+34.45%+6.45%+30.38%−3.27%2
NOK2026-04-211910.60+31.60%+3.60%+27.54%−7.97%2
GOOGL2025-11-20121304.54+30.29%−47.76%+20.44%−20.42%7

Top 15 Losers Lowest raw hold return. Many of these are positive nominal returns but underperformed the sector by a wide margin — see the next table for the alpha view.

TickerFirst BULLDaysEntry $Returnα vs XSDα vs SPYMax DD# BULL
PSIX2025-11-0313486.55−55.23%−110.00%−63.03%−61.53%1
CRWV2025-11-03134133.84−19.83%−74.60%−27.63%−48.90%1
IREN2025-10-0715360.22−12.10%−73.54%−22.00%−58.62%9
IONQ2025-09-1616858.98−11.92%−92.76%−23.67%−67.61%1
AEIS2026-04-0828363.21−10.94%−63.85%−20.23%−16.86%1
CDNS2026-05-077358.00−3.01%−5.75%−3.57%−4.66%1
CRDO2025-10-31135172.85−0.39%−56.24%−8.30%−53.59%1
IFNNY2026-05-15175.03+0.67%−0.40%+1.02%0.00%5
AMZN2025-10-31135250.10+5.61%−50.23%−2.29%−21.74%2
CIEN2026-04-2119507.77+9.20%−18.81%+5.13%−9.05%1
ON2026-05-068102.62+10.22%+5.46%+8.71%−4.88%1
MSFT2024-01-02595373.86+12.86%−134.77%−43.70%−34.18%5
AAPL2026-04-1325259.73+15.59%−32.87%+6.48%−2.52%4
BABA2025-06-27223114.37+15.93%−96.46%−4.68%−36.77%8
NVDA2025-10-13149187.97+19.87%−47.84%+7.98%−20.22%25

Top 15 Alpha Losers (vs XSD) Raw return can be positive but the sector was much stronger → net alpha is negative. Mega-caps and China ADRs dominate this list.

TickerFirst BULLDaysHold retXSD retAlpha# BULL
MSFT2024-01-02595+12.86%+147.62%−134.77%5
META2024-08-07445+22.08%+156.72%−134.64%3
PSIX2025-11-03134−55.23%+54.77%−110.00%1
BABA2025-06-27223+15.93%+112.39%−96.46%8
IONQ2025-09-16168−11.92%+80.84%−92.76%1
AEO2025-07-24205+27.42%+105.12%−77.71%1
CRWV2025-11-03134−19.83%+54.77%−74.60%1
IREN2025-10-07153−12.10%+61.44%−73.54%9
AEIS2026-04-0828−10.94%+52.91%−63.85%1
CRDO2025-10-31135−0.39%+55.84%−56.24%1
AMZN2025-10-31135+5.61%+55.84%−50.23%2
NVDA2025-10-13149+19.87%+67.71%−47.84%25
GOOGL2025-11-20121+30.29%+78.04%−47.76%7
AVGO2025-10-08152+27.37%+66.30%−38.93%4
AAPL2026-04-1325+15.59%+48.46%−32.87%4

High-Conviction Names (≥3 BULL mentions) The names Zephyr talked about most. # BULL = total bullish mentions. Verdict reflects alpha vs XSD bucketed by magnitude.

Ticker# BULL# BEARFirst BULLHold retXSD retAlphaVerdict
SNDK502025-10-01+1,145.01%+73.93%+1,071.08%Big winner
MU702025-10-08+283.44%+66.30%+217.14%Big winner
INTC402026-04-15+70.57%+41.14%+29.43%Beat the sector
QCOM302026-04-13+57.41%+48.46%+8.95%Beat the sector
AMD902025-11-20+82.25%+78.04%+4.21%Beat the sector
IFNNY502026-05-15+0.67%+1.07%−0.40%Trailed the sector
TSM902025-10-08+37.07%+66.30%−29.23%Trailed the sector
AAPL402026-04-13+15.59%+48.46%−32.87%Trailed the sector
AVGO402025-10-08+27.37%+66.30%−38.93%Trailed the sector
GOOGL712025-11-20+30.29%+78.04%−47.76%Trailed the sector
NVDA2522025-10-13+19.87%+67.71%−47.84%Trailed the sector
IREN922025-10-07−12.10%+61.44%−73.54%Devastating loss
BABA802025-06-27+15.93%+112.39%−96.46%Devastating loss
META312024-08-07+22.08%+156.72%−134.64%Devastating loss
MSFT502024-01-02+12.86%+147.62%−134.77%Devastating loss

Portfolio Simulation: Zephyr BULL vs XSD vs SPY $10K starting capital, equal-weighted into every BULL pick on first-mention day vs the same dollars into XSD or SPY on the same dates.

ScenarioFinal valueReturnDescription
$10K equal-weight Zephyr BULL (35 picks)$17,498+74.98%~$286 into each BULL pick, first mention day
$10K equal-weight XSD same dates$16,181+61.81%Buy XSD on the same entry dates
$10K equal-weight SPY same dates$11,139+11.39%Buy SPY on the same entry dates
Net alpha vs XSD+$1,317+13.17%How much Zephyr beat the sector
Net alpha vs SPY+$6,359+63.59%How much Zephyr beat the market

If you had started with $10K and put equal weight into every Zephyr BULL pick, you would have $17,498 today (+74.98%). Same dollars into XSD on the same dates: $16,181 (+61.81%). Same dollars into SPY: $11,139 (+11.39%). Net alpha: +13.17% vs XSD, +63.59% vs SPY.

SPY-beat is decisive but most of the XSD outperformance comes from a single ticker (SNDK +1,071 alpha). Strip SNDK and the alpha vs XSD turns mildly negative. The honest interpretation: Zephyr beats the broad market because he picks semis; he beats his own sector because of one outlier call.

Comparison: Zephyr vs Aleabit vs Jukan05 Three semi-adjacent X accounts run through the same v3 backtest pipeline.

Metric@aleabitoreddit@jukan05@zephyr_z9
Account age10 months20 months33 months
Followers302K117K146K
StyleStock rec + traderNews / supply-chain analystSemi / AI commentary + trader
Mention rate92.5%7.1%~23% (148/644)
Total mentions (analysed)5,325222147
BULL / BEAR split5,429 / 205 (v3)1 / 6 (sample)137 / 11
Unique BULL tickers3833035
Mean hold alpha vs XSD−37.32%+21.24% (sector beta)+13.17%
Median hold alpha vs XSD−51.72%n/a−18.81%
WR alpha XSD21.1%n/a37.1%
Top winnerAXTI +711% alphan/aSNDK +1,071% alpha
Strategy verdictPhotonics follow + fade restInfo source, not a signalMemory follow + mega-cap fade

Limitations & Honest Caveats

  1. Cashtag density is LOW. Only ~0.15% of Zephyr's 30,640 tweets use a $TICKER cashtag. The rest of his mentions were extracted via a COMPANY_MAP regex (NVDA, MU, IREN, AVGO, etc., ~50 tickers). Niche Asian / Chinese / Japanese names he discusses frequently (SICC, ACM Research, Hua Hong, FANUC, Sumitomo, Wartsila) are either not US-listed or unavailable in yfinance — they sit outside this dataset.
  2. Tweet sample is 644 of ~30,640. We pulled the full cashtag set (46 tweets) plus the dense recent ~35-day window (598 tweets). Expanding to the full 33 months would cost ~700 additional API calls; ROI is low because the mention-rate is so thin in non-recent windows.
  3. Sentiment classifier is imperfect on IREN. Zephyr is actually mixed-to-bearish on IREN ("stoopid", "foolish", "retarded" language). Our regex captured 11 BEAR mentions but missed enough cases that 9 of his IREN mentions still landed in the BULL bucket. The IREN −74% alpha is likely a misclassification — if anything it should reinforce his BEAR signal accuracy, not drag his BULL stats.
  4. Survivorship bias. Zephyr does not appear to delete tweets, but he does not always tweet position-size changes. "It was a good ride" (PSIX exit) was captured, but most positions are modelled as buy-and-hold — a real holder would have trimmed or rotated.
  5. Bull-cycle window. The 2.4-year period spans a memory + AI rally with one major drawdown leg (Sep 2024, Feb–Apr 2025, Mar 2026). XSD took the same drawdowns, so the alpha analysis is sector-neutral — but absolute returns would look very different in a bear cycle.
  6. Entry timing. Tweets posted after 16:00 ET use the D+1 open. In practice a tweet to 146K followers can move price the moment it goes out; the follower's actual fill is typically higher than our model assumes.

Final Verdict

+ Memory & storage thesis = real skill

SNDK +1,071% alpha (157d), MU +217% (152d), WDC +154% (135d), LITE +88% (100d), ASX +50% (105d) — the memory supercycle calls were early and correct. AEHR (test/burn-in) and CRDO (high-speed networking) sit at the periphery of the same thesis. This is the cluster to follow.

× Mega-cap & China ADR — no edge

MSFT −135% alpha, META −135%, BABA −96%, GOOGL −48%, AAPL −33%, AMZN ~0. Zephyr opines on the megas but produces no edge there — XSD / QQQ / SPY do the same job standalone. Don't follow him into this bucket.

! AI compute pure-plays are weak

IREN −74% alpha, CRWV −75%, NBIS only +9%. Zephyr is actually skeptical of this category in his prose ("stoopid", "foolish") — our BULL classifier under-counted his BEAR sentiment. Practical result is the same: a buy-and-forget IREN/CRWV hold is the worst outcome in the dataset.

Pragmatic strategy: follow the memory bucket, fade the mega-caps, mind the BEAR calls.

1. Follow the memory / storage cluster (SNDK / MU / WDC / Lumentum / SK Hynix proxy) — this is Zephyr's core thesis, repeated across many tweets, with +500% to +1000% moves over 2 years.

2. Ignore mega-cap mentions (NVDA / MSFT / META / GOOGL / AAPL / AMZN) — QQQ or XSD does the same work without single-name risk.

3. Respect the BEAR calls ("Shorting $SNPS", "PSIX shipments down", "$WOLF Chapter 11"). The sample is small (n=11) but the direction is right: 30–90 day alpha of −5% to −10%.

4. Niche Asia names (SICC, FANUC, Sumitomo, Wartsila) are not US-tradable but valuable for thesis building — treat Zephyr as an info source there, not a signal.

5. Paper trade for 60 days before committing real capital: validate that you can capture his next SNDK / MU-style early call in real time, not just in hindsight.

Full 6-page PDF report Every table, methodology notes, account comparison, limitations and final verdict in one file. Free, no signup.
Download PDF ↓
Share this report: Share on X Share on Telegram ← All research reports

Report generated 2026-05-18 by the fxcryptobots research desk. Source data: X archive of Zephyr (@zephyr_z9) from Dec 30, 2023 to May 18, 2026 (644 unique tweets sampled from a 30,640-tweet archive; 136 valid (tweet, ticker) mentions extracted via $TICKER cashtag plus company-name regex). Price data: Yahoo Finance close-to-close. XSD = SPDR S&P Semiconductor ETF; SPY = SPDR S&P 500 ETF; SOXX = iShares Semi ETF. This is research, not investment advice; see our risk disclaimer.