We backtested 79 finance influencers and newsletters. 3 passed.

Anyone who reads market commentary eventually asks the same question: is this person actually good, or just confident? We stopped guessing. Every source we follow — paid newsletters, Substacks, X accounts, Discord rooms — gets its full posting history pulled, every dated and tickered statement extracted, and the record scored against the market. This page is the scoreboard as of 2026-07-30, including the failures.

The method, in plain terms

The audit treats “should I listen to this source?” as a measurement problem with three stages. Each stage exists because a shortcut at that point once produced a wrong verdict that a fuller look reversed.

• 1. Species check. Does the source make explicit, dated, tickered directional statements — “bullish NVDA, here is why” — or does it publish coverage and context without ever committing? Only the first kind can be graded at all. The second kind is classified as radar: potentially useful for surfacing names and catalysts, but unscoreable, and never a reason to follow anyone into a position. Most sources either fail here or land in the radar bucket.
• 2. Full archive, never the feed. RSS typically exposes only the ~20 most recent posts, and a recent slice is a systematically flattering sample — you find a source because it is currently hot. Grading requires the complete history: sitemap crawls for Substacks, full-timeline pulls for X, room exports for Discord. The mirage this rule prevents is quantified below.
• 3. Backtest the statements. For each ticker the source has been explicitly bullish on, a hypothetical position is opened at the first (or, for sources that build conviction slowly, a later) bullish statement, and closed either when the source turns bearish or held to the end of the archive. Each position is scored as the stock’s return minus QQQ’s return over the identical period — a source riding a sector-wide rally scores zero, not a win. The scoreboard for a source is the fraction of its positions that beat the benchmark, and the median excess return.

Two standing rules sit on top. Self-reported track records are never counted — a published “my 2025 results” post is a survivorship exercise, and the audit only grades statements it can date independently. And every source is checked for echo: if its first bullish mention of a name tends to land after the stock has already outrun the benchmark by double digits in the prior ten sessions, that is momentum narration, not foresight — two otherwise plausible sources measured at +10.3% and +16.7% of pre-mention run-up and were graded accordingly.

The 3 that passed

“Statements” is every extracted directional or mention row in the archive; “n” is the number of ticker positions the graded rule actually produced. “Hold” means held to the latest price in the archive.

Two different species pass. SemiAnalysis and Irrational Analysis are first-call sources: the initial bullish statement is already informative. FundaAI is a rhythm source: its first mention of a name is exploratory, and the signal concentrates by the fourth bullish statement — conviction you can watch building across posts. Knowing which species a source is matters as much as knowing that it passed: reading a rhythm source as if it were a first-call source would grade, and trade, the wrong statements.

The one that passed alone and failed the team

A fourth source, Citrini Research, produced the single strongest standalone record in the audit: 2,416 statements across 908 tickers (2022-05 → 2026-06), with its graded rule at 64% / +21.6% median excess (n=44). Followed mechanically on its own, that rule returned +373% against QQQ’s +222% over the same span, through a -35% drawdown.

Added to a simulated portfolio already following the three sources above, it made the result worse: the combined portfolio finished at $189k against $202k without it. Its coverage overlaps the incumbents heavily, so its statements mostly collided with positions the portfolio already held — crowding capital rather than adding information. It is tracked separately rather than discarded, but the lesson generalises: a source’s value is marginal, not absolute. “Is this person good?” is the wrong question if you already follow someone who covers the same ground.

The failure gallery

25 sources were cut. The numbers below are real measurements from those audits; the names are withheld — the point is the failure modes, which repeat, not the individuals.

• The recent-slice mirage. A value-oriented generalist graded 83% / +12% on its recent three months — genuinely promising. The full 18-month archive, benchmark-adjusted, read 37% with a -5% median in 2025 (n=19) and 45% / -1% in 2026 (n=11). The hot streak was the anomaly. This single case is why stage 2 above is a hard rule.
• Popularity is not signal. A newsletter with more than 11,000 subscribers graded 36% on 58 first-mention statements, with a median excess of -10.9% — the audience had voted, and the market disagreed.
• The wrong species entirely. A trading personality’s recent 8 calls graded 75–88% and looked like a find. The full 968-trade history showed 71% options trades with a median holding period of 1 day: a day-trader, whose record cannot mean anything for someone holding stocks for weeks. The slice told a story; the archive told the truth.

The third exam: event days

Backtests grade what a source said. They cannot grade what a source failed to say — and silence in the source’s home domain is often the most informative datapoint. So on days when the market moves hard, the audit runs an open-book exam: for each source whose beat covers the event, did it flag the mechanism before the move (lead), narrate it after (explain), or say nothing at all (silent)?

Across 3 exam days so far — a foreign-leverage-led semiconductor crash, a single stock repricing 8% on a restructuring, and a hawkish Fed day — 9 source-event pairs have been graded: 5 lead, 2 explain, 2 silent. A trial cohort of 17 macro and sector commentators is currently being graded on this exam alone, under a standing rule: two leads and a source is promoted into the tracked set; three silences on home-field events and it is cut. No backtest required — the market writes the questions.

What this page is not

It is not investment advice, and it is not a recommendation to subscribe to, follow, or trade on anyone named above. It reports historical measurements of published statements — descriptive statistics with every known bias stated. A source that passed can decay, drift off its beat, or simply have been lucky in a regime that flattered it; the measurements here cannot tell those apart, and neither can you from this page alone.

It is also not reproducible by you, yet — the underlying archives include paid publications we cannot redistribute, so unlike the rest of this site’s numbers, these grades rest on data you cannot pull for free. We publish the method, the rules, and the failures so the process can be argued with, and corrections get published: quantdata@quantdata.uk.

Quant Data. "The finance influencer audit: 79 commentary sources graded against their full posting history." quantdata.uk. https://quantdata.uk/research/finance-influencer-audit

@misc{quantdata-influencer-audit,
  author       = {Quant Data},
  title        = {The finance influencer audit: 79 commentary sources graded against their full posting history},
  year         = {2026},
  howpublished = {\url{https://quantdata.uk/research/finance-influencer-audit}},
}

https://quantdata.uk/research/finance-influencer-audit