Archive · 16 Dec 2025
Complexity vs Simplicity in Investing Strategies
Some smart people are arguing about whether complicated computer models are better than simple ones for predicting how money markets will behave.
A study found that a very complex model with lots of settings could predict the future better than simpler models.
This is surprising because usually, simpler models are preferred.
Some people are excited because this could mean making more money in investments.
But others are skeptical and think the complex models might just be lucky or might not work well with messy data.
If the complex models are indeed better, it could change how investment firms work, making them hire more tech experts and spend more on data and computers.
But there are risks too, like not understanding how these complex models work and relying too much on them.
New research suggests complex machine-learning models may outperform simpler ones in predicting financial returns.
A study by Bryan Kelly and colleagues used a model with 12,000 parameters to analyze 12 months of data, challenging the traditional preference for simplicity.
Critics argue that the benefits of complexity may not hold if the data is noisy or if the models mimic momentum-trading strategies.
If complex models prove superior, investment firms may need to hire more machine-learning engineers and invest in better data and computational power.
Potential risks include increased reliance on black-box algorithms, reduced competition, and difficulty in interpreting complex models.
- Who
- Researchers Bryan Kelly, Kangying Zhou, and Semyon Malamud, along with critics Álvaro Cartea, Qi Jin, Yuantao Shi, and Stefan Nagel
- What
- A debate on the effectiveness of complex versus simple investing strategies using machine-learning models
- When
- The debate began in 2021 and has intensified recently
- Where
- The research and debate are centered around academic institutions and investment firms
- Why
- To determine if complex models can better predict financial outcomes and potentially revolutionize investing methods
- Researchers
- Bryan Kelly, Kangying Zhou, Semyon Malamud
- Institution
- Yale University, Swiss Federal Institute of Technology
- Model Parameters
- 12,000
- Data Period Analyzed
- 12 months
- Potential Impact
- Higher importance of machine-learning engineers, data acquisition, and computational power in investing
- Critics
- Álvaro Cartea, Qi Jin, Yuantao Shi, Stefan Nagel
- Criticism
- Complex models may not hold if data is noisy or mimic momentum-trading strategies
- Potential Risks
- Increased reliance on black-box algorithms, reduced competition, and difficulty in interpreting models