When 8% looks like a crash

Today, while building my prices ingest, I asked Claude a simple question: "if the dividend is actually 8%, does this still work?"

Context: I had to choose between two endpoints at my data provider. An endpoint, for those who missed it in a previous blog, is a specific question you can ask a data source. Fixed address, fixed answer format. A menu item. One endpoint delivered raw prices, the other delivered "dividend-adjusted" prices.

Claude had recommended the adjusted version. Argument: dividends are typically 1 to 3 percent per year, so marginal within a swing trading horizon.

Logical. But it didn't sit right.

What is a swing trading horizon?

Quick explanation, otherwise nobody can follow: my system buys stocks for several weeks to months. Not day trading (within one day), not long-term investing (years). Between those two. For a multi-month position, a 1-3% annual dividend indeed doesn't matter much. A few tenths of a percent on your position.

But some stocks pay much more. ING sits at 7 to 8 percent. Energy stocks similar. Real estate sometimes above 10 percent.

At those numbers, "marginal" doesn't work as an argument anymore.

What raw close does on a dividend payment

Say: a stock is at 100 euros. The company pays 8 euros in dividend. On the ex-dividend day the price opens at 92 euros. Not because the market thinks badly of the company. Because that 8 euros is now in your account.

Raw close shows:

• Day 1, before dividend: 100 euros
• Day 2, ex-dividend: 92 euros
• Day 3: 92.50 euros

For the market this is a normal day. For your portfolio value: nothing changed. You had 100 euros of stock. Now you have 92 euros of stock plus 8 euros of cash.

But my trading system would read this as an 8 percent drop. Because it only sees the price.

Two things that break

Stop loss. I have a rule "sell at losses over 8 percent". With raw data a normal dividend payment would automatically trigger my stop loss. Selling at what isn't actually a loss. Classic case of a system that thinks it's smarter than you, and reacts dumber as a result.

Momentum factor. One of my signals for "is this stock in an uptrend" would be systematically wrong for high-dividend stocks. Energy, banks, real estate. Exactly the sectors where I expect positions. The momentum z-score would persistently come out too low. Banks would appear in the top ten less often than they deserved.

What dividend-adjusted does

Dividend-adjusted close recalculates all historical prices retroactively. The series becomes:

• Day 1: 92 euros (adjusted)
• Day 2: 92 euros
• Day 3: 92.50 euros

No false drop. My system correctly reads it as "flat price". Stop loss doesn't trigger unnecessarily. Momentum is correct.

Where it could have gone wrong

If I had accepted Claude's first argument (marginal, 1 to 3 percent), I would have used raw data. Then on every ex-dividend day of a Dutch bank stock my system would have generated wrong signals. Sell orders on what wasn't a loss. Momentum scores systematically wrong for my dividend-heavy Dutch universe.

And the nasty part: it would work for months without me noticing. The errors wouldn't be large. One position sold that didn't need to be. A momentum ranking slightly lowered for a bank. Only if I specifically looked at a dividend announcement, or compared Sharpe ratios across sectors, would it stand out.

In the meantime I'd be trading on a system that was structurally doing the wrong thing.

Good sparring is good questions

Good sparring with AI isn't that AI always gives the right answer. It's that a good question from me forces Claude to think more carefully.

My question "what if it actually is 8 percent" made explicit that the earlier reassurance had been too easy. Claude acknowledged it directly, gave a stronger answer with reasoning, and the choice now sits in my project documentation. In six months I can still trace why raw data was rejected and adjusted won.

The throughline from earlier blogs: structure and discipline make an AI tool valuable. Not the tool itself. A variant today: the right question at the right moment is the difference between a working system and a system that quietly does the wrong thing for months.

Silent mistakes are the most expensive mistakes. Especially when there's real money in the game.