Should You Trust ETFs That Use AI to Pick Stocks?

The rise of ETFs that use AI to pick stocks is something investors looking at ETFs need to ponder. Now by ‘use AI’, we mean use it as the core engine that selects stocks, constructs portfolios and adapts to changing macroeconomic conditions.

ETFs That Use AI to Pick Stocks Are Here! And Here To Stay!

The most visible example in Australia is VanEck’s GOAT ETF, which has recently told investors that it will track a new index built using generative reinforcement learning. The index was developed jointly by VanEck and Seoul‑based Akros Technologies, an AI‑driven quantitative index specialist.

The model sifts through more than 1,200 of the world’s largest companies each month, simulates millions of market scenarios, evaluates over 10,000 signals — fundamentals, technical factors, macroeconomic indicators — and autonomously selects the 150 stocks with the highest probability of success.

In other words, this is an AI agentic system that learns from market conditions, adapts to new information and attempts to optimise future outcomes rather than simply replicate historical patterns. Other managers, including Minotaur and several global quant shops, are building similar systems.

The questions we seek to answer are should investors trust them? And if so, what do investors need to consider? Can they forgo due diligence because “the AI will handle it”? Or does the rise of AI‑driven ETFs make due diligence more important, not less?

In our view, the answer is nuanced. AI‑driven ETFs are different, but the question of whether they are inherently better or worse than traditional strategies is harder, and might even be a ‘dumb question’. Hear us out.

The promise of AI‑driven investing

The appeal of AI‑driven ETFs is that humans will struggle to consider thousands of variables – at least not all of them; but AI can.

Now, traditional quant models can process more than humans, but they are constrained by fixed rules and historical relationships. AI models, particularly generative reinforcement learning systems, are a further step ahead because they can process vast datasets, identify non‑linear relationships, adapt to new conditions and simulate millions of possible futures.

The GOAT index is built on that premise. Each month, the AI model evaluates more than 10,000 signals across 1,200 stocks. It simulates millions of market scenarios, scores each stock based on its probability of success and selects the top 150.

The model is designed to adapt to macroeconomic shifts — inflation cycles, rate changes, geopolitical shocks, liquidity conditions — and adjust the portfolio accordingly.
In theory, this solves several problems that plague traditional index construction in reducing human bias, backward-looking assumptions, rigidity and the lag between macroeconomic change and portfolio adjustment.

It also introduces forward‑looking optimisation. Traditional indices are built on historical data. AI‑driven indices attempt to optimise for future outcomes. This is the promise, but promises are not guarantees.

The risks investors need to understand with ETFs that use AI

AI‑driven ETFs introduce new risks that investors must understand before allocating capital. The first is model opacity. Traditional indices are transparent: investors know the rules, the factors and the weighting methodology. AI‑driven indices are not. The model may be explainable at a high level, but the specific decision logic is opaque. Investors must trust the model without fully understanding how it works.

The second risk is overfitting. AI models are powerful, but they can learn patterns that are statistically significant but economically meaningless. They can optimise for noise. They can overweight signals that worked in the past but will not work in the future. Reinforcement learning reduces this risk, but it does not eliminate it.

The third risk is regime dependency. AI models can adapt to new conditions, but they still learn from data. If the market enters a regime that has no historical analogue — a geopolitical shock, a liquidity freeze, a structural inflation shift — the model may misinterpret the signals. Human judgement can adapt instantly. AI judgement may lag.

The fourth risk is crowding. If multiple AI‑driven funds use similar signals, they may converge on similar stocks. This creates crowding risk, particularly in mid‑cap names. Crowding reduces alpha and increases volatility.

The fifth risk is accountability. When a traditional manager underperforms, investors can evaluate the decisions. When an AI model underperforms, investors cannot interrogate the logic. Accountability becomes abstract.

We’re not saying that these risks make AI‑driven ETFs untrustworthy, but we are saying that investors must understand what they are buying.

Can investors forgo due diligence?

The short answer is no. The longer answer is that due diligence becomes more important, not less.

AI may change where judgement is applied but does not eliminate judgement. Investors must evaluate the model architecture, the data sources, the training methodology, the reinforcement logic, the rebalancing frequency, the risk controls and the governance framework. They must evaluate the manager’s ability to monitor, validate and override the model when necessary.

The GOAT index uses generative reinforcement learning — a sophisticated architecture that allows the model to simulate millions of market scenarios and optimise for future outcomes. But investors still need to understand how the model is trained, how it avoids overfitting, how it handles regime shifts, how it interprets macroeconomic signals and how it manages risk.

What investors should consider before trusting AI‑driven ETFs?

Investors should consider three things. The first is whether the model has a clear economic rationale. AI models can identify patterns, but patterns are not always meaningful. Investors must understand whether the model’s signals have economic logic, not just statistical correlation.

The second is whether the model has been tested across multiple market regimes. Reinforcement learning is powerful, but it must be validated across inflation cycles, rate cycles, liquidity shocks and geopolitical events. Investors should look for evidence that the model performs in both calm and volatile markets.

The third is whether the manager has strong oversight. AI models are not autonomous. They require governance, monitoring and intervention. Investors should evaluate whether the manager has the capability to supervise the model, validate its decisions and override it when necessary.
AI‑driven ETFs are not “set and forget”. They are “trust but verify”.

Will AI‑driven ETFs outperform traditional strategies?

The honest answer is that no one knows. AI‑driven ETFs may outperform in certain regimes and underperform in others. They may excel in environments with high data complexity and struggle in environments with low signal clarity. They may outperform traditional quant models but underperform discretionary managers during periods of structural change. In the end, they may record better returns than traditional funds and may be spruiked as better – but it may be coincidental rather than being just because of AI.

The GOAT index is an ambitious attempt to build a forward‑looking, adaptive, macro‑sensitive portfolio using generative reinforcement learning. It may succeed. It may struggle. What matters is not whether AI is inherently better but whether the specific model is well‑designed, well‑trained and well‑supervised.

Investors should not assume AI will outperform simply because it is AI. They should evaluate the model the same way they evaluate any investment strategy: logic, evidence, risk, governance and execution.

Conclusion

AI‑driven ETFs represent a new frontier in portfolio construction. They are not inherently superior to traditional strategies, but they offer a different approach — one that is adaptive, data‑rich and forward‑looking. Investors should not fear them, but they should not blindly trust them either.

The question investors need to ask is,“ should you trust this AI‑driven ETF, built by this manager, using this model, trained on this data, supervised by this governance framework?”

AI changes due diligence by changing its focus but not eliminating the need. Investors who understand that distinction will be better positioned to evaluate whether AI‑driven ETFs like GOAT (and the broader wave of AI‑powered funds emerging globally) represent innovation, noise or something in between. And we reckon investors need to be positioned because GOAT will be just the beginning, mark our words.

Are you new to ETF investing? Check out these articles before investing!