A few big winners: why owning everything catches them

Look at companies one by one over a long period and a surprising picture appears. In the Zapas model of 200 synthetic companies over 20 years:

• 7 went to zero;
• 86 more lost money;
• so 93, or 46.5%, ended below where they started in real terms;
• only 12 grew more than tenfold.

Yet the index of all 200 rose 235.4%. Both facts are true at once because the winners grew by far more than the losers could lose: a company can fall at most 100%, but it can rise 1,000% or more.

For an investor holding a few companies this is uncomfortable. Pick five at random and there is a real chance most of them end up as losers, even while "the market" did well.

Skewed returns means the outcomes are lopsided: many small or negative results and a long tail of very large ones. Single company returns over long periods look like that.

The reason is simple arithmetic. A share can lose at most 100% of its value, but it can gain without limit. Over 20 years, in the Zapas model, the best company returned +4,788.7%. One such company adds as much to the total as dozens of ordinary ones.

In a skewed set the median (the middle value) and the average split apart. Here the median company made 11.5% and the average, which is what the equal-amount index earns, made 235.4%.

So "the typical stock" and "the market" are different things. The market's return is carried by its tail.

Long-period returns are often quoted as a total: +235.4% over 20 years. To compare it with yearly numbers, convert it to a yearly rate that, compounded, gives the same total.

Add one, take the N-th root, subtract one: (1 + 2.354)^(1/20) − 1 = 6.2% a year.

Don't divide the total by the number of years. 235.4 ÷ 20 = 11.8% looks twice as good, because it ignores that each year's growth builds on the previous ones. The same mistake runs the other way: 6.2% a year sounds modest, yet over 20 years it more than triples the money.

When a product shows a long-run figure, say both forms: "+235% over 20 years, about 6.2% a year".

If a small group of companies drives most of the market's gains, then the most important thing is to own that group. The trouble is that nobody can reliably name it in advance.

In the Zapas model, the top 10 of 200 companies (5%) produced 50.8% of all the net gains over 20 years. Only 53 companies (26.5%) beat the index at all. A short list chosen at the start is more likely to hold ordinary companies than tail winners, so its likely result is closer to the median company (11.5%) than to the index (235.4%).

A broad fund solves this without any forecasting: it owns every company, so it owns the future winners too. In real markets broad funds use market-cap weighting (bigger companies get a bigger share), so as winners grow, their weight grows with them.

One way to measure how lopsided returns are is to ask: what share of all the gains came from the top few?

Add up the 20-year returns of all 200 companies, losers included, to get the net total. Then add up only the top 10. In the Zapas model the top 10 come to 50.8% of the net total. Ten companies did as much as the other 190 together.

This is why concentration cuts both ways. A portfolio that happens to own two or three of the tail winners can beat the market by a lot, and those stories get told. A portfolio that owns none of them lags badly, and those stories are rarely told. Because you can't tell in advance which portfolio you'll have, the expected result of a short list is worse than its best-case story suggests.

A model is useful only if you read it carefully. The 200-company data supports three claims:

• many single companies lose money over long periods even when the index rises (93 of 200);
• most companies trail the index; only 53 (26.5%) beat it;
• a small group makes a large share of the gains (the top 10 made 50.8%).

It does not show that picking is hopeless for everyone, that the same companies will win next time, or that real markets have exactly these numbers. It is a synthetic universe built to show a pattern, not a record of history.

In product copy, claim what the model shows and stop there. "In our model" is a short phrase that keeps the explainer honest.

A newsletter paragraph is not a report. Users skim, so lead with the surprise, back it with one or two numbers, and end with what it means for them.

For this finding a strong order is:

1. The surprise: almost half the companies lost money, and the middle one made only 11.5% in 20 years.
2. The cause: a few companies, 10 of 200, made about half of all gains.
3. The consequence: nobody can name them in advance, so the app owns everything, and therefore owns them.

Say the data is a synthetic model. Keep the tone calm: no "beat the market", no "guaranteed", and no hint that Zapas knows which companies will win. The point of the paragraph is that nobody does.