How quantitative investing and machine learning work
- Explain what quantitative investing is
- Tell rules-based scoring from a trained model
- Know why markets are hard for models
Rules instead of hunches
Quantitative investing uses data and explicit rules to make decisions. A simple quant rule might rank stocks by 12-month price momentum and buy the top 10% each month.
Rules are consistent and testable, and they do not get scared. They also only know what their data shows.
Rules-based scores versus trained models
A rules-based score combines indicators with weights a person chooses. VEXA’s Score works this way: technical, fundamental and news signals with fixed weights. It is transparent, but the weights are judgement, not proof.
A machine-learning model learns weights from historical data, by finding patterns that best fit past outcomes in a training set, and is then checked on separate data it has not seen. VEXA does not currently use a trained machine-learning model.
Why markets are hard to model
Markets are noisy: most daily movement is unpredictable. Patterns that worked can stop working once many people trade on them. Relationships shift with interest rates, technology and regulation.
Data is also limited: there is only one history, and a few decades of monthly returns is not much data for a complex model.
Factors with long research histories
Academic research has documented a few factors, such as value, momentum, size, quality and low volatility, that have been associated with return differences over long periods, with long stretches where each failed. Alpha Finder’s ranking uses price-based factors like momentum and low volatility and publishes how its buckets did historically.
Rule: each month, rank 500 stocks by their 12-month return excluding the most recent month, and hold the top 50 equally. It is fully defined, so anyone can test it on history.
Testing it honestly means using only data available at each month’s end, charging trading costs, and including companies that later disappeared.
Key terms
- Quantitative investing
- Investing based on data and explicit, testable rules.
- Rules-based score
- A score combining indicators with weights chosen by a person.
- Machine learning
- Methods that learn patterns from historical data rather than using hand-set rules.
- Training set
- The historical data a model learns from.
- Test set
- Separate data used to check a model on examples it did not learn from.
- Factor
- A characteristic, such as momentum or value, associated with differences in returns.
On the Alpha Finder page, open ‘Validation: does it work?’ and write down one thing the validation shows and one limitation it admits.
Key takeaways
- Quant investing uses explicit, testable rules.
- Rules-based scores use human-chosen weights; ML learns them from data.
- VEXA’s scores are rules-based, not trained models.
- Markets are noisy and patterns can fade once known.