+62% in four years. But three quarters of the market I never bought.

Today my trading system underwent its first real test. Four years simulated, June 2022 through end of April 2026. Once daily rebalanced. The same rules the system runs now, applied to four years of historical data.

The result: +61.81% return over four years. €10,000 starting capital became €16,181.

That's not spectacular by itself. The S&P 500 did about +30% in the same period. So my system did roughly twice the market. But that's not the whole story. In fact, the most interesting observations come when you look beyond the bottom line.

The bets beforehand

Beforehand I had bet: +75% return, Sharpe 4.0, max drawdown -8%. Harry, my AI sparring partner, had bet: +55%, Sharpe 1.8, max drawdown -18%.

It became: +61.81%, Sharpe 0.60, max drawdown -20%.

Nobody had it fully right. My return prediction was close, my Sharpe prediction was insanity. Harry's drawdown prediction was almost exact. This says something about how hard prediction remains, even when you've built the system yourself.

What Sharpe 0.60 means

Sharpe ratio is a measure of how much return you get per unit of risk. Higher is better.

For reference:

• Passive index fund (S&P 500): around 0.5
• Warren Buffett over his entire career: 0.76
• Renaissance Medallion (legendary hedge fund): 2.5

So my 0.60 sits just above passive index and below Buffett. That's not bad, but also not "I've found an edge." For a first version of a multi-factor model it's reasonable. For a production system that goes live with €10,000 of real money it's a meaningful starting point.

What it says: the system makes use of existing effects (value, quality, momentum), but it doesn't extract significant alpha from them. It's just slightly better than blindly buying the market, not much better.

Year-by-year breakdown

Here it gets interesting. Averages hide stories.

| Year | End value | Return | Market context | |------|-----------|--------|----------------| | 2022 (Jun-Dec) | €8,336 | -16.6% | Bear market | | 2023 | €9,130 | +8.7% | Market did +24% (underperformance) | | 2024 | €12,761 | +39.7% | Bull market, multi-factor worked | | 2025 | €15,231 | +17.6% | Market parity | | 2026 (Jan-Apr) | €16,181 | +4.0% | Mixed |

2024 was the miracle year. +40% in 12 months, S&P 500 did +23%. That's where the real outperformance was.

2023 is the worrying year. The market recovered strongly (+24%), but my system did only +9%. That's an underperformance of 15 percentage points. In a recovery market where almost everything goes up, my system probably picked the wrong stocks. Or the right stocks at the wrong moments.

The silent engine: how you handle losses

My system has multiple ways to exit a position. What the numbers show is something fascinating: half of all exits are stop losses, and they lose money. That sounds bad, but it's exactly how trend-following works.

You accept small losses often, in order to be able to hold on to the rare big winners. Another part of the exits are trailing stops, and those work almost always. Stocks the system picked up, let run, and only sold when they started declining. Those are the moneymakers.

The difference between loss and gain doesn't sit in the buying. It sits in whether a stock gets the chance to trend, or whether it gets shot down too quickly. That's the real craft in a trend system, and that's where my V2 research will focus.

Three quarters of the market I never bought

Here comes the real surprise. My system traded only 141 unique stocks in four years, out of a universe of 554. That's 25.5%.

What happened to the other 75%? They were never bought because they were too expensive for my portfolio. My system targets 14 positions at once. With €10,000 starting capital that's about €700 per position.

Stocks above €700 were impossible to buy:

• BKNG (Booking): $4,000+ per share
• MELI (MercadoLibre): $2,000+
• AVGO (Broadcom): $1,500+
• NVDA (post splits): often $700+

This is a structural problem that no factor could solve. You can't take a position in BKNG for €700 if a single share costs $4,000. So those got skipped, regardless of how good their scores were.

That fundamentally changes the story. My system made +62%, but did it with one hand tied behind its back.

€50,000 as V2 experiment

When I realized this, I ran the same backtest again with €50,000 starting capital. Four years, same period, same rules. But now €3,500 per position, enough to buy all stocks except a few extreme outliers (BRK.A).

The result: €50,000 became €86,952. A return of +73.9% over four years. The Sharpe ratio rose from 0.60 to 0.75. Drawdown improved from -20% to -19%. Annualized return from 12.7% to 14.7%.

The difference per year showed when the extra room mattered. In 2024, the bull market year, return went from +40% to +48%. In 2022, the bear market, the loss was smaller: -13.5% versus -16.6%. With more position room the system could deploy its defensive mechanisms better.

But there's a second layer underneath. With €50,000 the system traded 131 unique stocks, not 141 like at €10,000. It bought different names: ADYEN.AS, FICO, AZO, CVNA, and eleven others that weren't reachable with €10,000. It also did not buy a number of names that €10,000 did, like MSFT, GOOGL, AMGN. The bigger the portfolio, the more selective the ranker, the better the choices.

An interesting detail: NVDA was traded only twice with €50,000 instead of five times, and therefore cost -23% instead of -47%. Bigger positions mean fewer repeated whipsaws on the same name. That's not coincidence.

My provisional conclusion: €10,000 is a fine starting amount to learn with, but the strategy structurally works better with more capital. Below €15,000 you miss a third of your investable universe. Stock prices aren't designed for small portfolios.

What I do and don't read into this

What I do read:

• The system isn't broken. It works. +62% over 4 years is real return, not artifact.
• Trailing stop is the real engine. Letting stocks run works.
• Risk management works. -20% drawdown in a 4-year window with 2022 included is acceptable.
• 2024 showed multi-factor can have real edge in bull markets.

What I don't read:

• Not "I've discovered an edge." Sharpe 0.60 is no edge.
• Not "this is a proven strategy." Four years of data is statistically too short.
• Not "I know what 2027 brings." These are historical numbers, not predictions.

What I don't know (and honestly can't know)

Survivorship bias: my universe contains only companies that still exist now. Companies that went bankrupt or were acquired in these 4 years aren't in my dataset. That makes my numbers 5-10% per year too optimistic. With survivorship correction +62% might have been +35%. Not certain, but realistic.

Look-ahead bias: I've worked hard to prevent this (see the blog about the Sharpe 7 bug). But there can remain subtle forms, like that scores on date X make use of closing prices that on date X-1 weren't yet known.

Forward-predictive power: this is a backtest, not a prediction. What worked in 2022-2026 doesn't have to work in 2026-2030. Markets evolve. Edges disappear.

What now

On June 22, 2026, this system goes live with €10,000 of real money. Between now and then:

• Paper trading from June 1
• Definitive position sizing tuning
• Monitoring infrastructure
• A final check on everything

In five years I'll know whether this was edge, luck, or survivorship bias. Until then it's all hypothesis.

What I do know: four years of simulation has taught me that I've built a system that at least isn't broken. It works as designed. It loses money in bear markets, makes money in bull markets, and cuts losses quickly while letting winners run.

That's enough basis to start with. The rest I'll learn along the way, with real money.