How the Optimizer Limits Concentration

Watch: How the Optimizer Limits Concentration

A short video deep dive on this topic. Prefer to read? The full post is below.

July, and what we changed

July was a difficult month for AlphaWizzard. The portfolio that entered the month was heavily focused on one theme: AI hardware, from chips and memory to the optical parts that connect them. When that theme fell, most of the portfolio fell with it. The full July numbers will follow in the monthly review, as they do every month.

We responded with two new selection rules during July, and both are already live in the portfolio. The first limits how much of the portfolio one sector can hold. The second limits correlation clusters. If you copy the strategy, the resulting trades appeared in your own account as they happened.

Sector labels can hide concentration

Every stock carries an official sector label, such as Technology or Energy. These labels are useful, but they have a blind spot. Some companies carry different labels, yet their share prices move together. Think of a chip maker, an electricity producer that powers data centers, and a construction company that builds them. They carry three different sector labels, yet all three depend on the same AI-demand story. In a sell-off around that theme, all three often fall as one group.

Our answer is to group stocks by behavior instead of labels. The grouping uses price data alone. Stocks whose prices move together land in the same cluster, whatever their label says. The chart below shows what this found in July’s candidate list.

41 of 132 companies moved with a group outside their official sector. A sector limit alone cannot see these connections. A cluster limit can. That difference is why we added a second rule instead of only tightening the first.

The new rule, in one sentence

The rule itself is simple: no more than half of our positions may come from one correlation cluster. The sector limit stays in place as an extra safety layer. The Optimizer decides which qualified names to hold, not how big each position should be. All positions stay equal-weight, so the cluster rule limits how many names can share one risk driver.

Half of the portfolio in one cluster may still sound like a lot. That is a deliberate trade-off. We run a focused portfolio of 15 to 30 names, because focus is where outperformance comes from. Spread too widely, a portfolio starts to behave like the index it tries to beat. A stricter limit, allowing fewer stocks per cluster, would lower the concentration risk further. It would also cut into the return potential we are here to pursue. We chose the level with the help of historical simulations, aiming for a sound balance of risk and return. The final judge is live performance, month after month.

What stays fully visible is the result: every trade and every position appears publicly on eToro, at all times. The illustration below shows the idea in one image.

What the limits do, and what they do not

Their job

The limits cap how hard one theme can hit the portfolio. The return potential comes from the other side of the process: selecting strong companies and holding them with discipline.

How they were built

Historical simulations informed the design of these limits. They are built for the months ahead. Every month is its own market.

Broad market declines

A cluster limit shapes the stock list. Protection in broad declines is the separate job of the Risk Overlay. Every night, the system recalculates more than 20 risk indicators across markets and sets our equity exposure for the next day.

What the end-of-month rebalance does

Our end-of-month July rebalance positions the portfolio for August, trading into the month-end close. Every name must re-qualify against the same systematic screens, on a freshly updated stock universe. Only the best opportunities stay.

If you choose to copy, eToro technically replicates our trades proportionally into your own account. You stay in control of your own account at all times. Data-driven, not gut-driven: the rules decide which names stay, and the same rules now watch the concentration. Improving the process never stops; when the data supports a refinement, we build it in.

“We cannot control what markets do. We can control how much of the portfolio moves as one block.”

Related reading: Sector Rotation: Are We an AI Portfolio? on how sector labels can mislead, Monthly Review: June 2026 for the last full monthly report, and Navigating AI Volatility: Why We Hold US Memory on the theme that tested us in July.

Past performance is not an indication of future results. Your capital is at risk.

Weekly investment newsletter, every Sunday — positioning, performance vs SPY/QQQ, and what we're watching next. No spam.