What my AI missed at Wolters Kluwer

Friday May 15, 12:00 NL. My system has just submitted 14 buy orders to IBKR. One of them is Wolters Kluwer. 49 shares, €58.34 each, total position €2,860. The factor score is high, the fundamental numbers are solid, the rule is clear: top 14 ranks get bought.

But something is nagging. Wolters Kluwer has dropped 63 percent in twelve months. From €166 to €58. While in that same period the operational numbers have just kept growing: 6 percent organic revenue growth, EPS guidance €6.71 for 2026, recurring revenue at 83 percent of total and growing 7 percent per year.

How can a company with this profile sit in the top 3 percent of my universe by factor score, while the market has punished it 63 percent? And more importantly: what is my system not seeing?

The pattern I hadn't caught

Same week I started an exploration. Not to change my V3.1b factor model before the June 22 launch. There's no time, no backtest, and no discipline for that. But to understand where the gap is.

Four tickers as testbed: Wolters Kluwer (punished by market despite fundamentals), SanDisk (rocketed on AI memory thesis), plus Micron and Seagate as peer cohort for SanDisk.

The numbers were unforgiving:

SanDisk: between May 2025 and May 2026 the price went from $35 to $1,383. A return of 3,854 percent in 12 months. Five consecutive earnings surprises, one with an EPS beat of 522 percent. Revenue growth from $1.9B per quarter to $5.95B. Classic post-earnings-announcement-drift in every direction the literature predicts.

Wolters Kluwer: in the same 12 months -63 percent. Despite six consecutive earnings beats, one with a 31 percent beat. A fundamentally strong company in a market pricing in its structural future through the lens of AI disruption fears.

Micron and Seagate: similar SanDisk pattern from mid-2025. Before that: surprises delivered little. From June-August 2025 onwards: surprises got a drift of 30 to 85 percent at T+90 days.

What makes the difference

Not the surprise alone. Wolters Kluwer had surprises too. Some bigger than Micron's. Yet they did nothing to the price.

The difference sits in something I've started calling "sector regime". Storage and memory was in a demonstrable momentum regime from mid-2025. The whole sector outperformed the broader market. Three tickers in the same sector reinforced each other. Earnings surprises got rewarded by the market in that regime.

Information services where Wolters Kluwer sits did the opposite. The whole sector got the bear lens through AI threat narratives. Earnings beats got no credit. Morningstar even downgraded the moat from Wide to Narrow for both Wolters Kluwer and Thomson Reuters in March 2026. Sector in negative regime, surprises mean nothing.

That's a factor my current V3.1b model doesn't explicitly measure. I have price momentum (6 and 12 month returns). I have earnings catalyst (recent surprise). But I don't have sector regime detection saying "the whole segment is running" or "the whole segment is punished".

The six criteria I'm researching

Based on four tickers of data I formulated a concept-filter. Not to implement today. Not for June 22. But to research from July in a serious backtest.

Six criteria that must fire simultaneously before a ticker qualifies for a separate "earnings catalyst sleeve" alongside the core 14 positions:

One: positive earnings surprise of at least +5 percent.

Two: revenue growth sequential (Q over Q more than 3 percent) and YoY (more than 10 percent).

Three: EPS year-over-year growth above 25 percent.

Four: sector in top quartile of universe by 60 day and 90 day return.

Five: at least two peer tickers in the same GICS sector level 2 also with positive surprise most recent quarter.

Six: operating cashflow positive.

Especially criteria three and six are the "actually making money" guard. Cannabis 2019 would fail on three (operating losses). SPAC-mania 2020 would fail on six (negative cashflow). Those are the patterns I want to filter out, not buy into.

Why this isn't a feature yet

Honest disclosure: I'm looking at seven datapoints across three tickers in the same sector. That's not statistical evidence. That's exploratory observation.

For this to really become something there needs to be a backtest over my whole universe (503 plus 51 tickers) and over my whole OOS period (2018 to 2024). With out-of-sample validation on a separate 2025 period not used in tuning. With walk-forward analysis. With Sharpe incremental calculation versus V3.1b baseline.

If that backtest yields Sharpe incremental above 1.2: then we're talking about a feature for V3.2. Maybe August deployment. Maybe as a parallel sleeve with its own capital allocation of 5 to 10 percent of portfolio.

If the backtest comes in below 0.8 incremental: idea gets parked. Was then probably just sector momentum we observed, not earnings power. Both outcomes are valuable. Research that only yields positive results is usually curated, not validated.

What does Wolters Kluwer keep doing in my portfolio

Hold Friday's position. No manual adjustments. V3.1b has a mean reversion exit rule that ensures underperforming positions get closed at some point as factor scores drop. That handles itself, without my intervention.

The idea of building a new factor now to remove Wolters Kluwer from selection is exactly the psychological trigger systematic trading protects against. I'd be curve-fitting on one observation. A mistake that only works as long as the pattern keeps repeating, and that is vulnerable to regime shifts.

Wolters Kluwer could be a mean reversion winner in 2 or 3 years. Solid fundamentals, structurally large market, recurring revenue. Or it could remain a value trap if AI disruption becomes substantive. Both are possible. My portfolio diversification ensures one position doesn't become a disaster.

The take-away

A high factor score doesn't mean a position will rise. It means a position historically had a good chance of rising, given the factors my model measures. Sector regime is a factor my V3.1b model doesn't measure. That's a gap. It's being researched for V3.2, not hacked into V3.1b for June 22.

SanDisk rose 3,854 percent because all three were on at once: sector momentum, earnings power, and peer confirmation. Wolters Kluwer fell 63 percent because the sector was against, despite continued earnings beats. Between those two sits the kind of signal I'm going to try to isolate in July. With n much larger than seven, and with the discipline to accept the outcome or park the idea.