How can machine investors beat humans? In the introductory chapter of his January 2018 book entitled “Financial Machine Learning as a Distinct Subject”, Marcos Lopez de Prado prescribes success factors for machine learning as applied to finance. He intends that the book: (1) bridge the divide between academia and industry by sharing experience-based knowledge in a rigorous manner; (2) promote a role for finance that suppresses guessing and gambling; and, (3) unravel the complexities of using machine learning in finance. He intends that investment professionals with a strong machine learning background apply the knowledge to modernize finance and deliver actual value to investors. Based on 20 years of experience, including management of several multi-billion dollar funds for institutional investors using machine learning algorithms, he concludes that:
- Discretionary portfolio managers cannot naturally work as a team and, in fact, generally work separately. Employing quants in this way fosters bad strategies derived either from backtest overfitting or overcrowded factor investing.
- Successful quantitative firms instead set up research factories, with each quant specializing in a particular task, while having having a holistic view of the full process.
- Individuals searching nowadays for macroscopic alpha face overwhelming odds. The only true alpha left is microscopic, and finding it requires capital-intensive factory methods, with the following key team roles:
- Data curators collect, clean, index, adjust and store inputs for subsequent analysis.
- Feature analysts apply information theory, signal extraction, visualization, labeling, weighting, classifying to screen data features that have some power to predict financial variables.
- Strategists apply deep knowledge of financial markets and the economy to transform predictive data features into investment algorithms (strategies to be tested), including theories that explains them.
- Backtesters apply deep understanding of experimental methods to assess profitability of investment strategies under various scenarios, including historical performance, with focus on probability of backtest overfitting.
- Deployment specialists design and program code to ready strategies for production with minimal latency, reusing code when applicable.
- Portfolio oversight specialists guide each strategy through a 5-step life cycle:
- Embargo – confirm strategy backtest results on previously reserved data or new data accumulated during deployment.
- Paper trading – if passed through embargo, further confirm strategy performance on live data to ensure real-time executability.
- Graduation – if passed through paper trading, implement the strategy and track returns, risks and costs precisely.
- Re-allocation – continually reassess the allocation to the strategy in the context of a
diversified portfolio, generally starting small and then increasing or decreasing the allocation according to actual performance. - Decommission – discontinue the strategy if it performs below expectations long enough to conclude that evidence no longer fits the supporting theory.
- Careless users of machine learning accelerate false discoveries because of low signal-to-noise ratios in finance and poorly trained algorithms that confuse statistical flukes with patterns.
- Discretionary portfolio managers are slow learners with an inherent disadvantage when betting against a machine learning algorithm, but it is possible that combining discretionary managers with machine learning algorithms achieves the best results.
In summary, investors should understand the potential (as competitor or ally) of machine learning as a disruptive technology that may transform how everyone invests for generations.
Cautions regarding conclusions include:
- As noted, the book requires considerable background.
- Investors seeking the benefits of well-executed machine learning may bear material fees.
See also:
