A few weeks ago I posted about tracking 23 paid Substack newsletters to see which ones actually make money. That post got 7.4k upvotes and a lot of great feedback here. The three biggest criticisms:

1. ”Mixing apples and oranges." Some of those authors are macro/ETF traders, not stock pickers. Ranking them alongside individual stock pickers isn't fair.
2. "30 and 60 days is too short." Many of these authors have 6–12 month theses. Need longer time horizons.
3. "23 isn't enough." There are more newsletters worth tracking.

They all made sense so I spent a few days to update my code. The update itself was done in 2 days, but it took me more than a week to recover from Iran’s bombing on my account.

Quick disclaimer: I am not affiliated with any of these Substack authors. No one is paying me to promote them. Case in point — Global Tech Research ranked #1 in Part 1, and they haven't published a single new article since my post. If I were shilling for someone, they'd at least be active. This is just data from a guy who spends too much money on newsletters and wanted to know which ones are worth it.

UPDATE: As I’m about to publish this post, I realized Global Tech Research (who ranks #1 this time and last time) took his substack off. Not sure what happened, but I do hope he gets back in the game soon.

What Changed

1. Classification — stock pickers only.

I built a three-tier classification to separate content types:

• Macro commentary (7 authors, excluded): >50% of their high-conviction calls are ETFs, indices, or crypto. James Bulltard (SPY, TLT), 10x Research (BTC, ETH), Paulo Macro, Lord Fed, Macro Charts, Shrubstack, Eliant Capital. These need different evaluation (market timing accuracy), not stock-picking metrics.
• Sector commentary (1 author, excluded): Clouded Judgement is a weekly SaaS sector review — valuable but not actionable stock picks.
• Stock pickers (23 authors, ranked): Specific, thesis-driven calls on individual stocks with clear direction.

If you're wondering where Paulo Macro, Lord Fed, and Macro Charts went from Part 1 — they're reclassified as macro commentary and excluded from the stock-picking rankings. They're still in the portfolio, still costing me money, just measured separately.

A note on volume fairness. Among the stock pickers, output frequency varies wildly. Some authors (TMT Breakout, FundaAI) publish daily/weekly with a commitment to consistent coverage. Others only write when they have a high-conviction idea. Comparing a 529-call author head-to-head with a 33-call author isn't entirely apples-to-apples — the high-volume authors are playing a different game. That's why I added a High-Volume comparison section later in this post, to give those prolific authors a fair benchmark against each other.

2. Time horizons — now up to 360 days.

Added 180d and 360d windows on top of the existing 1d/7d/15d/30d/60d.

Unfortunately, a lot of the substacks are fairly recent so it's hard to judge them on 360d. But I'll keep watching.

Data cutoff: 2026-03-03 (before the Iran conflict distorted markets). Analysis generated 2026-03-15.

3. More authors — expanded from 23 to 32.

8 new additions. Some of these I’ve heard of, some of them are completely new (Dick Capital thanks to wsb):
– FundaAI ($1,000/yr — semiconductor/SaaS deep research)
– Citrini Research ($999/yr)
– Dick Capital ($300/yr)
– Irrational Analysis ($200/yr)
– BEP Research ($400/yr)
– ZA Stocks ($400/yr)
– Tae Kim ($480/yr)
– Total annual spend is now ~$13,400.

4. Median over average.

Switched primary metric from average to median returns. Averages get wrecked by outliers — one 200% winner makes a mediocre author look great. Median tells you what the typical pick actually does.

5. Alpha vs benchmarks. (Same as last time, but i'd like to highlight it)

Every return is also measured against sector-specific benchmarks: SOXX for semis, IGV for SaaS, KWEB for Chinese tech, EWJ for Japan, GLD for gold miners, SPY as fallback. Alpha = your pick's return minus what the relevant sector ETF did over the same period. This separates actual stock-picking skill from riding a bull market.

Total dataset: 3,101 high-conviction calls from 23 stock pickers, tracked over 1+ year.

The Rankings

Median Return (Long Calls) — Top 10 by 60d

Rank	Author	Calls	30d	60d	180d
1	Global Tech Research	33	+12.9%	+20.4%	+67.3%
2	Citrini Research	19	+8.5%	+14.9%	+30.4%
3	FundaAI	158	+5.5%	+11.2%	+23.1%
4	SemiAnalysis	45	+6.1%	+8.2%	+43.5%
5	BEP Research	52	+1.8%	+8.0%	+3.5%
6	Dick Capital	44	+0.3%	+7.6%	—
7	Irrational Analysis	41	+5.6%	+7.4%	+47.0%
8	Fabricated Knowledge	91	+3.2%	+6.0%	+17.4%
9	Altay Capital	55	+3.3%	+5.5%	+10.3%
10	TMT Breakout	529	+2.3%	+5.1%	+9.0%

I think it's interesting that in general return dilute with # of calls. Those who concentrate and do their research call less and earn more.

Median Alpha vs Benchmark (Long Calls) — Top 10 by 60d

This is where it gets interesting. High returns don't always mean skill — some authors just rode the market.

Rank	Author	Calls	30d α	60d α	180d α
1	Global Tech Research	33	+11.0%	+11.8%	+42.7%
2	Dick Capital	44	-1.1%	+8.1%	—
3	Altay Capital	55	+2.0%	+4.0%	+2.6%
4	Citrini Research	19	+4.6%	+3.0%	+14.9%
5	FundaAI	158	+1.7%	+2.6%	+9.8%
6	Fabricated Knowledge	91	+0.2%	+1.6%	+10.5%
7	TMT Breakout	529	-0.6%	+1.5%	-1.3%
8	Accrued Interest	81	-1.6%	+0.9%	-3.0%
9	Doomberg	20	-0.3%	+0.6%	-5.2%
10	The Setup Factory	81	+1.7%	+0.3%	+4.2%

Notice how the rankings shuffle compared to the return table. BEP Research was #5 by returns but drops out of the alpha top 10 (60d alpha: -1.9%) — meaning their picks actually underperformed their sector benchmarks despite being up. SemiAnalysis goes from #4 to off the list (60d alpha: +0.0%). Meanwhile Altay Capital and Fabricated Knowledge climb because their picks genuinely outperformed their sectors.

The five authors generating consistent alpha across horizons: Global Tech Research, FundaAI, Citrini Research, Fabricated Knowledge, and Altay Capital.

The Long Game — 180d and 360d

One of the biggest asks was "show me the 6-month and 1-year numbers." Here they are.

180-Day Median Return (Long) — Top 10

Rank	Author	Calls	180d Med	180d Med α
1	Global Tech Research	33	+67.3%	+42.7%
2	Irrational Analysis	41	+47.0%	+3.1%
3	SemiAnalysis	45	+43.5%	+1.3%
4	Citrini Research	19	+30.4%	+14.9%
5	FundaAI	158	+23.1%	+9.8%
6	Fabricated Knowledge	91	+17.4%	+10.5%
7	The Setup Factory	81	+15.9%	+4.2%
8	Winter Gems	40	+12.5%	+7.8%
9	Altay Capital	55	+10.3%	+2.6%
10	TMT Breakout	529	+9.0%	-1.3%

The 180d view reshuffles things again. Irrational Analysis jumps from #7 (60d) to #2 — their deep value thesis takes 6+ months to play out. SemiAnalysis goes from barely positive alpha at 60d to #3 by absolute return at 180d. The market eventually catches up to good semiconductor research.

360-Day Median Return (Long) — Early Results

Fewer authors have enough 360d data, so take these with a grain of salt. But interesting to see which theses survive a full year:

Rank	Author	360d Med	360d Med α
1	Irrational Analysis	+91.6%	+24.7%
2	Global Tech Research	+60.1%	-12.7%
3	Swiss Transparent Portfolio	+55.6%	+36.8%
4	Winter Gems	+29.4%	+22.5%
5	Fabricated Knowledge	+26.7%	+24.1%
6	TMT Breakout	+25.0%	+2.2%
7	Altay Capital	+13.3%	-10.3%
8	TicToc Trading	+12.0%	+1.8%
9	FundaAI	+11.0%	+8.4%
10	The Setup Factory	+10.0%	+4.1%

Notable: at 360d, alpha matters even more than returns. Global Tech Research has +60.1% return but negative alpha — the benchmark rose even more over a full year. Fabricated Knowledge and Winter Gems are the standouts: high absolute return AND high alpha at 360d. 

The Volume Question

Something I didn't address in Part 1: does the number of calls matter?

Yes. A lot.

An author making 15 calls can have a great record by being highly selective. But an author making 158 calls and still outperforming? That's a much harder thing to do.

Among high-volume authors (100+ long calls):

Author	Long Calls	60d Med	60d Med α	60d Win%	180d Med
FundaAI	158	+11.2%	+2.6%	70%	+23.1%
Swiss Transparent Portfolio	141	-0.1%	-4.9%	50%	+1.9%
TMT Breakout	529	+5.1%	+1.5%	62%	+9.0%
TicToc Trading	480	+0.1%	-0.7%	50%	+4.6%
Quality Stocks	300	+0.1%	-1.7%	51%	+0.0%

Most high-volume authors converge toward market returns — more calls means more noise. FundaAI seems to be a good outlier. At that volume, consistency matters more than any single call.

And at the bottom…

Michael J. Burry: 15 long calls, 60d median return -14.2%, 60d alpha -12.7%, win rate 12%. His 23 short calls aren't much better (60d: -2.8%). Then again, The Big Short took 2 years to play out. So I guess we can have a bit more patience.

What I Learned

1. Alpha > returns. Some top-10 return authors have negative alpha — they just rode the market. The five generating real skill-based returns across multiple horizons: Global Tech Research, FundaAI, Citrini Research, Fabricated Knowledge, and Altay Capital.
2. Volume and quality don't usually coexist. Most authors either pick a few winners or spray-and-pray. The rare exception that maintains quality at high volume stands out in the data.
3. Time horizon changes everything. SemiAnalysis is mediocre at 30d but incredible at 180d. Irrational Analysis barely cracks the top 10 at 60d but is #2 at 180d. If you pick newsletters based on 30-day numbers alone, you'll miss the deep value authors.
4. $5,400 of my $13,400 goes to macro newsletters (James Bulltard, 10x Research, Lord Fed, etc.) that don't even appear in the stock-picking rankings. They may be valuable for market outlook, but they're not generating trackable stock picks. Something to think about.
5. Expensive ≠ better, cheap ≠ worse. Global Tech Research ($100/yr) is #1 across nearly every metric. Altay Capital ($80/yr) is consistently top 10. Meanwhile some $500-1,000+ newsletters are mid-table. But some expensive ones (FundaAI, Citrini) also deliver.
6. Shorts are still hard. Almost every author does worse on shorts than longs. This hasn't changed from Part 1.

Cost Breakdown

~$13,400/year across 31 newsletters (USD + EUR/GBP converted):

Stock Pickers (23 authors, ~$8,000/year):

Author	Annual Fee	Author	Annual Fee
FundaAI	$1,000	Collyer Bridge	$350
Citrini Research	$999	Dick Capital	$300
TMT Breakout	$589	Doomberg	$300
SemiAnalysis	$500	TicToc Trading	$290
Tae Kim	$480	Irrational Analysis	$200
The Setup Factory	$450	Global Tech Research	$100
Best Anchor Stocks	$449	Earnings Edge	$100
Michael J. Burry	$439	Altay Capital	$80
ZA Stocks	$400	Quality Stocks	$70
BEP Research	$400	Winter Gems	$50
Fabricated Knowledge	$400	Swiss Transparent	~$43
Accrued Interest	$80

Macro & Sector (9 authors, ~$5,700/year, not ranked):

Author	Annual Fee	Author	Annual Fee
James Bulltard	$1,099	Shrubstack	$500
Lord Fed	~$1,010	Macro Charts	$400
10x Research	$948	Paulo Macro	$360
Eliant Capital	$760	The Overshoot	$330
Clouded Judgement	Free

Methodology

• Source: Crawled every article from all 31 Substack authors. 1,780+ articles (stock pickers only) analyzed.
• Extraction: Gemini AI identifies high-conviction stock picks — not casual mentions, but tickers the author dedicates real analysis, specific data, or price targets to. Spot-checked against manual reads.
• Deduplication: If an author calls the same ticker within 14 days, only the first mention counts. Different authors calling the same stock are tracked independently.
• Benchmarks: Each ticker is mapped to its sector ETF for alpha calculation. If a semi stock goes up 10% and SOXX goes up 8%, the alpha is +2%.
• Survivorship bias: Only tickers with available price data on yfinance are tracked. Delisted, some international, and typos are excluded.
• This was a bull market year. Many authors are long-biased. Absolute returns look good partly because the market went up. That's why alpha matters.
• Data cutoff 2026-03-03. Returns calculated before the Iran conflict affected markets.

The Biggest Limitation (Read This)

I want to be upfront about the single biggest weakness of this entire system: the AI extraction is only as good as the writing is clear.

Every article gets sent to Gemini 3.5 with a structured prompt that asks: "Is this a high-conviction bullish or bearish call?" The model looks for explicit directional conclusions, dedicated analysis (≥3 sentences per ticker), independent supporting arguments, and quantitative data.

The problem is that many newsletter authors write in nuanced, hedged, or indirect ways. An article might spend 2,000 words building a thesis and then conclude with "we remain cautious but see long-term potential”, is that bullish or bearish? The AI has to make a judgment call, and sometimes it gets it wrong. I manually checked a number of the posts and to be very honest, it’s hard for me to understand whether the author meant “buy now” or “this has potential but we will see”.

This affects some authors more than others. Authors who write clearly ("We are bullish on X, price target $Y") get extracted accurately. Authors who write in a more academic or hedged style get noisier results. The rankings are therefore partially a function of writing clarity, not just stock-picking skill.

I've published the exact Gemini prompt I use so you can judge for yourself how the extraction works (at the end of the post). For technicality, the model temperature was set to 0 for reproducibility, if you care to know.

A suggestion for Substack authors: if you want your calls tracked accurately (by me or anyone else), consider adding a clear summary at the top or bottom of each article — something like "Positions: Long NVDA, Short CRM." It takes 10 seconds and eliminates all ambiguity. The more explicit you are, the more fairly your track record gets measured.

Happy to answer questions. Roast my methodology. Tell me what to add for Part 3.

Positions: long several names mentioned by top authors. Not financial advice.

Gemini prompt for reference:

ROLE: You are a professional financial analyst. Deeply analyze the following newsletter article and extract investment-related information.

PHASE A – Candidate Identification

1. Only identify companies/tickers explicitly mentioned in the article
2. If a company name can't be uniquely mapped to a ticker (multi-listed or ambiguous, skip it – do not guess
3. Every candidate must have a supporting evidence sentence from the text

PHASE B – Classification & Conviction Rating

1. Assign direction: bullish / bearish / neutral
2. Conviction: only high or low
3. HIGH conviction requires ALL FOUR conditions simultaneously:

• Clear directional conclusion (long/short), not just factual description
• ≥ 3 dedicated analysis sentences about this specific ticker
• ≥ 2 independent supporting arguments
• ≥ 1 quantitative data point (valuation / growth / guidance / target / financials)

→ If ANY condition is not met → low

1. Neutral → always low
2. Same ticker in both bullish and bearish → neutral (low)

KEY RULES (Strict)

1) Determine article_type first, then extract tickers:

• macro: macro/policy/rates/geopolitics focused, stocks as examples only
• stock_analysis: core is company/stock analysis with investment conclusions
• mixed: both macro and stock-level depth

2) If article_type = macro OR title contains weekly/update/roundup →

default ALL tickers to low; only upgrade to high if all 4 conditions met

3) Companies merely "cited as examples" in sector discussion → low

4) If article is not investment-related → return empty arrays

SELF-CHECK BEFORE OUTPUT

• All neutral entries must be low conviction
• No duplicate tickers
• No conflicting directions for the same ticker
• Every high-conviction call must satisfy all four