We Sold Our Best Stock Picks Too Early: The Full Record

Between 22 February and 18 April 2026 we published ten stock picks and closed all ten by 24 June. As closed, they returned an average of +14.4%. Had we done nothing and held all ten to 23 September, they would be up +34.1%. The S&P 500 gained +15.4% over the same windows. We picked well and sold badly. This is the full record, including the numbers that weaken our own case.

All ten picks, not just the winners

As closed, the ten beat the S&P 500 over their holding periods, +14.4% against +9.0%. Seven of ten made money. Held, eight of ten would be up, and the average would be 18.7 points ahead of the market instead of 5.4.

Where the gap comes from

Three sales explain it. We closed Magnite at +15.9% on 23 April. It is +112.4% today. We closed Amplitude at -5.9% the same day. It is +79.9%. A stop took us out of Freshworks at +1.1% on 24 April. It is +57.9%.

Those three exits cost 239 percentage points. The whole gap across the ten picks is 197 points.

Where selling was right

Not every exit was a mistake. A trailing stop closed indie Semiconductor at +35.4% in June. Held, it would be +4.4%. We sold Snap in two halves for +40.3%. It would be +29.7% today. IMCD and Opera were each 3.8 points better sold than held. Together those four exits saved 49 points.

So the lesson is not "never sell". Selling on price alone gave back far more than it protected.

What happened after April

From May to September we published 21 more picks that are now closed. They returned +0.3% as closed. Held, they would be -2.6%, against +2.7% for the S&P 500 over the same windows. Here our exits helped slightly.

The bigger problem was the picks themselves. They drifted towards larger, widely followed companies, and the upside we could justify shrank with them. The median target was +29% above entry for the later 21 picks, against +55% for the first ten. A +20% target cannot become a Magnite, however long you hold it. Fixing our exits alone would not have fixed that.

What we changed

1. Every pick now states two or three facts that would prove the thesis wrong, such as a margin level, a contract or a guidance number. Those facts close a position. The chart does not.
2. One disaster stop, around -35% or the business's downside floor. It needs five closes in a row below it. A one-day drop on a headline never triggers it.
3. No trailing stops, no selling at a fixed target, and no selling because a date passed.
4. We only buy what we expect to return 30% or more a year, and we keep holding while that is still true from today's price.
5. No quota. Some weeks bring five ideas, some months none.
6. Our screen starts from our own data: fair value from several valuation models, valuation against the company's own history, insider buying, fund manager filings and congressional trades. The web search comes second.
7. We keep tracking every pick after we sell it, so our exits get scored too.

Every open and closed pick stays on our public track record, including the ones that went wrong.

How we measured this

• Returns are price returns. Dividends are excluded.
• Entry price is the price published with the pick.
• IMCD, Coloplast and Interpump are measured in their local currency (EUR, DKK, EUR).
• "If held" uses closing prices on 23 September 2026. That number moves every day and will be different when you read this.
• S&P 500 figures use SPY closing prices on the same dates.
• Opera and Snap were closed in two halves. Their return at close is the average of both halves.
• Targets are the last target we set for each pick, measured from its entry price.
• Ten picks is a small sample, and hindsight makes every sale look worse than it felt at the time.

This is research and education, not personal investment advice. Past performance does not predict future results.