Zacks advertises 23.89% a year since 1988. Its own disclaimer explains how.

A portfolio rebalanced every month buys and sells constantly. Every trade pays a commission, pays the spread between the bid and the ask, and pushes the price against itself a little. That toll is not an accounting detail: in concentrated, high-turnover portfolios it is the difference between an idea and a business.

We know because we measured it on ourselves, and it stings. The 20-stock multi-factor portfolio we publish returns 16.75% a year gross since March 2011. Charging every trade in the cheapest scenario we know how to model — a €100,000 portfolio at an electronic broker paying €0.30 a trade — the toll takes 160 to 200 basis points a year, and what is left sits between 14.41% and 14.88%.

Between 1.9 and 2.3 points of annual return disappear into execution alone, and that is the best version. With €10,000 at a traditional bank charging €20 a trade, the same strategy stops making money at all: −17.9% a year. Commissions are not a discount on the result; sometimes they are the entire result. A backtest that books that toll at zero is not being optimistic: it is measuring something else.

Our figures come from carteras.json (the file that paints the site) and costes_estimados.json, generated by medir_costes.py, which rebuilds the portfolio across all 186 rebalances and charges every trade with the estimated spread of its company on its day. They are published, with the method written out, on the methodology page, and the full portfolio has its own page.

Zacks also runs exchange-traded funds with real money. The 2026 Form 485BPOS prospectus publishes what they returned and what they measure themselves against — the Russell 3000, their choice:

Two of three behind their own yardstick. And the one that wins has only eleven months of history: eleven months does not separate skill from luck anywhere, least of all here.

There are also two Zacks-branded ETFs that no longer exist: they were liquidated. That is not a reproach — closing a fund that does not gather assets is ordinary in this industry — but it is exactly why any track record of surviving funds reads better than what actually happened: the ones that went badly are no longer in the table. We know something about that, and we wrote it up in the second piece in this series.

That paper used the Zacks database to study whether following analyst recommendations made money. It found a gross edge of more than 4% a year — the signal was there — and once they paid for the turnover required to capture it, the net result was indistinguishable from zero.

That is exactly the ending of section 01, written twenty-five years earlier and peer reviewed. The signal was not fake. The toll ate it.

And there is a gap worth saying out loud: there is no independent academic verification of the Zacks Rank. Not that it exists and is unflattering. It does not exist.

The Zacks Rank is built on analyst estimate revisions: whether consensus for a company is moving up or down, and when. Reproducing it would require storing that consensus with its date, day by day, for decades. We do not have it: all that is within reach is the last four quarters from Yahoo, which get overwritten every night and therefore cannot be used to look backward without cheating.

That limit is written into our own measurement file, in the “what cannot be measured with this data” section, and we did not write it for this article: it was already there.

Buying the deep history is not an option for a retail investor either. The honest path left is to pay for today's consensus and archive it daily: that gives clean point-in-time data from the day you start and zero history going back. Which leaves the 1988 claim where it was: on their word.

The limitation is declared in medicion_pead.json, field no_se_puede_medir_con_estos_datos, written by medir_pead.py. Literally: “Without that, the Zacks Rank is not reproducible.”

Our 20-stock multi-factor portfolio publishes 16.75% a year gross against 11.89% for the equal-weighted S&P 500 benchmark over the same window. Three things we say on the same page where we say that:

• Costs are stated separately and measured. Between 160 and 200 basis points a year in the cheapest scenario, plus eight more scenarios in case yours is different. We did not pick that number: it comes from reconstructing every trade.
• The edge is observed, not proven. Across 186 rebalances the t-statistic is 1.11, and the Fama-MacBeth information coefficient gives t = 0.72. Fifteen years of data cannot establish that those four points are skill. We say it in those words, on the site.
• When we find a bias in our own favor, we lower the number. On September 1 we lowered all five published curves. The story, with the script, is in the next piece.

None of this says our portfolio is better than theirs. It says ours can be audited: the simulator rebuilds the curve from 2011 using only what was known on each day, the ranking shows all 500 scores for today, and the series download as CSV with no account.

• We lowered our own published returns, and here is why
• Five things we tested that do NOT work