Machine Learning in Asset Management: What Actually Works
The honest answer to where machine learning has produced durable edge in investing — and where it has mostly produced expensive overfitting.
Machine learning has moved through three stages in finance: hype, disillusionment and, more recently, mature integration. The first stage produced spectacular research backtests that failed in production. The second produced a backlash. The third, where serious quant firms now sit, is quieter and more honest about what the technology actually does well.
Where ML adds durable value
Feature engineering on alternative data — satellite imagery, supply-chain telemetry, transcript embeddings — is where modern ML earns its keep. The models themselves are often modest: gradient-boosted trees, regularised linear models, and small transformers fine-tuned on financial text. The edge comes from the data and the discipline of cross-validation, not from architecture.
Where it fails predictably
Deep models applied to short, noisy, non-stationary financial time series are an excellent way to overfit. Walk-forward validation, purged k-fold and combinatorial purged cross-validation (de Prado) are essential. So is honest accounting of the multiple hypothesis testing problem: most published “alpha” is statistical noise.
The right stack for an institutional ML programme
A clean lakehouse for raw and engineered features; a model registry with versioned datasets and code; out-of-sample evaluation gates before any model touches capital; live performance monitoring against a champion-challenger framework. These engineering investments matter more than the choice of algorithm.
FAQ
Are large language models useful for trading?
They are useful for processing unstructured text — transcripts, filings, news — into structured signals. As standalone forecasting engines on price data, they remain unconvincing.
Does ML replace human researchers?
No. It changes their job. Senior researchers spend more time on data quality, hypothesis design and post-hoc analysis, and less on hand-coded feature transforms.
August Quants Research
The August Quants research desk publishes educational essays on systematic investing, market structure, ML in finance and portfolio construction. We write for institutional readers who value rigour over noise.