How Acubic Handles Market Volatility

Key takeaways

• Acubic runs its construction process on a fixed schedule; it does not trigger extra trades in response to market falls.
• No order reaches your broker without your involvement unless you have specifically enabled full-auto mode on a supported account.
• Switching to monitoring mode keeps the connection active and stops all automated orders, at any time.
• The methodology page publishes five model limitations; one applies directly to volatile periods: correlations between risky assets tend to converge in drawdowns.
• Acubic does not predict short-term market moves and does not attempt to time exits or re-entries around price events.

The most common concern investors raise before connecting any AI portfolio tool to a live account is this: what happens when markets drop? The question is reasonable. Handing over any degree of automation to a system creates real questions about what it will do in an extreme situation, when the cost of an unwanted trade is highest. This article answers those questions precisely, from how the construction cycle runs to what the approval step actually does and what monitoring mode is for.

What Acubic does not do when markets move

Acubic does not monitor price feeds and react to market events in real time. It does not reduce positions when a market falls, move to cash when volatility rises, or increase allocations when conditions look favourable. None of those behaviours exist in the system. The article on why Acubic does not predict the market explains this at length, but the short version is structural: the construction method uses historical price data to build a risk-calibrated allocation, not to forecast where any asset is going. A system that does not forecast cannot sensibly react to price signals.

In practice, this means a market sell-off does not trigger any additional activity in Acubic beyond what was already scheduled. If your next scheduled rebalance is three weeks away, nothing happens in those three weeks regardless of how markets move. If a rebalance runs during a period of volatility, it runs the same process it runs in any other period: screen the universe, rank by the method you chose, allocate using the optimizer, and propose the smallest set of trades that closes the gap between current positions and the resulting target.

How the construction cycle runs

The construction process runs as three stages: screen, select, and allocate. Screening removes assets without sufficient price history. Selection ranks the eligible universe by market capitalisation, trailing Sharpe ratio, or trailing momentum, depending on your setting. Allocation assigns weights using the optimizer you chose, subject to your position bounds and risk constraints. This is the same process every rebalance runs, on the schedule you configured, whether markets have been calm or turbulent.

The key property is that it reruns from the current state of the portfolio each time, not from a fixed static target. A scheduled rebalance during a downturn reflects the current price levels and the current correlation structure of the holdings. What triggers a rebalance is the schedule, not market conditions.

Portfolio drift is the gap between where your weights are now and where the construction method says they should be given the current data. In a period of sharp price moves, that gap grows faster than normal because the relative prices of holdings diverge quickly. The next scheduled rebalance corrects that drift toward the target the method derives from the current data, not toward a target frozen at a better moment.

The model limitation that applies to volatile periods

The methodology page lists five model limitations explicitly. One of them applies directly to volatile periods: in severe drawdowns, correlations between risky assets tend to converge toward one. This is the condition under which diversification is expected to matter most, and it is the condition in which the diversification assumption in the model is most tested.

Hierarchical risk parity, the default optimizer, distributes risk across groups of assets that tend to move differently from each other under normal conditions. When a sharp market event causes most risky assets to fall together, those historical correlation differences narrow. The model does not anticipate this: it adapts as new price data arrives and the rolling lookback window reflects the new correlation structure. The methodology page names this as a lag, not a forecast, and names it as a property of this class of model rather than a defect to be tuned away. It applies to hierarchical risk parity, mean-variance optimisation, and CVaR minimisation alike.

Reading that limitation before connecting real money is the intended path. It is there because a user who understands it is better positioned to judge the output appropriately than one who is not aware of it.

Staying in control

Whatever the market is doing, you have two tools that keep you in direct control of what Acubic does.

On a standard eToro connection, every rebalance requires your approval before any order reaches the broker. The scheduler prepares the trade set and stops. Nothing executes until you act. This is not a default you change in settings: it is the only mode available on a main eToro account. If you want to skip a rebalance because the timing does not feel right, you decline or simply leave it pending, and the next scheduled run is the next opportunity. The article on one-click vs full-auto rebalancing covers what the approval step does and when it applies on each broker.

Monitoring mode takes this further. In monitoring mode, Acubic tracks your live positions and calculates drift, but does not prepare any rebalance and does not send any orders. You stay connected and you can observe how the portfolio is behaving, but no automated activity takes place. You can switch to monitoring mode at any time, from any other mode, and switch back when you are ready. The article on pausing Acubic rebalancing without disconnecting explains this path in full.

Disconnecting entirely is also always an option, and it does not liquidate anything. Every existing position stays exactly where it is. Acubic stops sending orders, and that is all.

What full-auto means during a volatile period

Full-auto mode removes the per-rebalance approval prompt. Scheduled rebalances execute without requiring your presence. On a Trading 212 Invest account, all three modes including full-auto are available directly. On eToro, full-auto requires an eToro Agent Portfolio, which is a ring-fenced sub-account you fund separately from your main balance.

In full-auto, the construction process runs on schedule exactly as it does in one-click mode. The safety guards are the same: orders are submitted one at a time, position sizing rounds down, and every event is recorded in an audit log. What changes is that the trade set is not staged for your approval before going to the broker. During a period of sharp price movement, a scheduled full-auto rebalance executes the trades the method derives from the current data. That may include buying assets that have fallen in price if their current weights are below target, or trimming assets that have risen above their target weight. Those trades follow from the method; they are not predictions about what will happen next.

If you are in full-auto and want to stop automated activity for a period, switching to monitoring mode is the direct path. The setting can be changed at any time and takes effect before the next scheduled run.

What this means in practice

The practical picture is this: when markets drop, Acubic does not take any autonomous protective action. It waits for its scheduled run. When that run arrives, it applies its method and either proposes a trade set for your review or executes it, depending on your mode. The model limitation on correlation convergence means the diversification assumption in the method is tested most precisely during the events that matter most. That is the honest answer, grounded in the five limitations the methodology page publishes before you connect anything.

A user who understands that Acubic is a construction method running on a schedule, not a prediction system reacting to price events, has a bounded set of failure modes to consider: the historical correlations may not hold forward, a regime change may behave differently from any period in the training data, a rebalance may execute during a drawdown rather than after it. Those risks are named. They are not hidden behind a performance claim. Acubic does not promise returns and cannot predict short-term market moves.

For a full walk-through of what Acubic does from first use to connected broker, the article on how Acubic works covers the complete cycle. The methodology page publishes the five model limitations, the three optimisers, and the three ranking methods before you commit to any configuration.

Getting started

Understanding what the system does in a difficult market before connecting real money is the right question to ask. Acubic does not promise a return and does not predict short-term market moves. It applies a stated method on a schedule you set, with a human approval step in the loop unless you specifically remove it. The AI portfolio builder is where to begin if you want to see what a portfolio looks like before connecting a broker.