Sector Rotation Trading with Machine Learning Models

Sector rotation is one of the most studied strategies in investing: money moves between sectors as the economic cycle evolves. Energy does well when inflation is rising. Consumer staples hold up when the economy slows. Tech leads when rates are low and growth expectations are high.

The problem with most sector rotation strategies: they're rule-based and lagging. By the time the macro data officially confirms that energy is entering a bullish phase, the move has already started.

A trained ML model gives you a different kind of rotation signal — one that comes from the stocks themselves, not from macro indicators you're reading after the fact.

How Pick Counts Become a Rotation Signal

When I run sector models — Energy Top 100, Consumer Cyclical, Healthcare, Tech — I'm not just using them to find individual stocks. I'm using the pick count itself as a signal about the sector.

Here's what that looks like in practice.

The Energy model might have 2–3 picks per day for weeks. That's the model scanning 100 energy stocks every morning, looking at all its learned signals, and concluding: the setups aren't there. Confidence is high — the model isn't confused — but the stocks don't match the patterns it was trained to find. So it barely picks.

Then one morning: 28 picks. The next day, 18, then 16. The sector had pulled back. Valuations reset. Momentum conditions changed. The model found setups that match its training, and it started firing.

That spike in pick count is a rotation signal. The model is telling you something changed before the headlines explain it. You don't need to track every energy stock manually. You just check the pick count. When it goes from 3 to 28, something happened in the sector.

The Energy Model in Action

In April 2026, the Energy sector peaked in late March and sold off hard — down over 3% across the board. During the peak and the early selloff, the Energy model barely moved. 7 picks, 4 picks, 3 picks, 1 pick. Confidence stayed steady between 65–73% — the model wasn't broken. It just wasn't seeing the setups it was trained to find at elevated prices.

Then April 9th and 10th: 33 picks. Then 25, then 66, then 43, then 67.

The pullback created the exact kind of setups the model learned to find. Stocks that had pulled back to technically interesting levels, with the macro context aligning. The model recognized them. The equity curve followed — up 2.45% while the Energy benchmark was still down 3.10%. A 5.5% spread from simply letting the model tell you when to get in.

The Consumer Cyclical Model: Months of Patience

The same pattern played out over a longer timeframe with the Consumer Cyclical model.

For months — January through mid-March — the model had 1 or 2 picks per day. Sometimes zero. It wasn't broken. The sector was bleeding, and the model was accurately reflecting that it couldn't find setups worth taking.

Then late March: 43 picks. Then 43 again. Then 23. The sector had bottomed and started recovering. The model saw it in the stock-level data before the headline narrative caught up.

The equity curve was flat all the way through the quiet period — no picks, no losses. Then straight up when the model woke up. The Consumer Cyclical sector itself was still down roughly 4% at the time.

Running Multiple Sector Models Together

The real power comes from running several sector models simultaneously and using the pick counts as a dashboard.

Every morning the question is simple: which models have picks today?

• Energy at 2 picks — ignore it
• Consumer Cyclical at 1 pick — ignore it
• Healthcare at 45 picks — something is happening in healthcare
• Tech at 8 picks — normal background noise

You don't need to read every sector report or track macroeconomic indicators in real time. The models are doing that implicitly — they were trained on conditions that included all of those factors. When conditions align in a sector, the model finds setups. When they don't, it doesn't. The pick count is the summary signal.

Why This Works Better Than Traditional Rotation

Traditional sector rotation models rely on economic indicators that are reported monthly or quarterly, often with revisions. You're trading on data that describes what already happened.

ML sector models are trained on stock-level signals — price, momentum, fundamentals, relative strength — that update every day. When a sector bottoms and conditions start changing, the stocks move before the economic indicators reflect it. A model trained on those stock-level signals catches the early rotation signal at the source.

You're not rotating because a report says energy is entering a bullish cycle. You're rotating because 28 energy stocks just showed up on your model's pick list after weeks of silence. That's the market telling you directly.