[WEBINAR] Designing Universes and Ranking Systems - Yuval Taylor

Summary

This webinar delves into designing universes and ranking systems for investment strategies. It distinguishes between hard rules, like price and market cap, suitable for universe definitions, and flexible rules like growth and margins, better for ranking systems. The presenter advocates for a refined universe to enhance ranking accuracy, illustrating the concept with examples and discussing the implications of including unwanted stocks in rankings. The session also touches on strategy implementation, backtesting principles, and adjusting for market capitalization changes over time.

Takeaways

• 🌌 **Webinar Series Overview**: The webinar series covers designing universes and ranking systems, simulations, principles of backtesting, and strategy implementation.
• 📈 **Strategy Implementation**: The placement of rules in strategies is crucial; they can be placed in universes, ranking systems, or screens/simulations.
• 🔢 **Rule Types**: There are two types of rules - hard and fast rules (e.g., price > $1, market cap > $50M) and flexible rules (e.g., low PE, high growth). Hard rules are better suited for universes, while flexible rules fit ranking systems.
• 📉 **Ranking System Accuracy**: Ranking systems are more accurate with a limited, well-defined universe rather than a large one with subsequent exclusions.
• 📑 **Experimentation**: The presenter conducts experiments to show how different universe definitions (like excluding over-the-counter stocks or those with market cap under a billion) affect stock rankings.
• 🚫 **Noise Reduction**: Excluding unwanted stocks from the universe reduces noise in the ranking system, leading to a cleaner and more accurate ranking.
• 💧 **Liquidity Considerations**: Liquidity constraints, like minimum prices and volume, are important, especially for microcap investments.
• 📊 **Backtesting Flexibility**: Using flexible restraints, such as market index multiples or ranks, can improve the accuracy of backtesting over time.
• 🛠️ **Ranking System Nodes**: Various nodes (composite, conditional, boolean, factor, function, industry) are used to build complex and nuanced ranking systems.
• 🔄 **Adjusting for Market Changes**: Adjusting hard and fast rules over time, such as market cap filters, is important to keep strategies relevant as market conditions evolve.
• ✂️ **Text Editing for Ranking Systems**: Advanced users can use text editing to manipulate ranking systems, such as removing composite nodes or transferring factors between systems.

Q & A

What are the main topics covered in the webinar series?

The webinar series covers designing universes and ranking systems, simulations, principles of back testing, and strategy implementation.

What is the difference between hard and fast rules and flexible rules in strategy design?

Hard and fast rules are strict criteria such as price being greater than a dollar or market cap being over 50 million. They are better suited for universes. Flexible rules are more subjective, like avoiding companies with high banished scores or preferring companies with low PE ratios, and are better for ranking systems.

Why are hard and fast rules more suitable for universe design?

Hard and fast rules are more suitable for universe design because ranking systems work more accurately on a delimited universe. This helps to reduce noise and focus on a smaller, more relevant set of stocks.

How does including unwanted stocks in a ranking system introduce noise?

Including unwanted stocks in a ranking system introduces noise because it adds irrelevant data to the analysis, which can skew the results and make the ranking less accurate for the stocks of interest.

What is the core combination ranking system mentioned in the script?

The core combination ranking system is a method that combines six core ranking systems into one, used for ranking stocks based on various criteria without including over-the-counter stocks or those with a market cap under a billion dollars.

Why might someone want to exclude certain types of stocks like MLPs or REITs from their universe?

Someone might want to exclude certain types of stocks like MLPs or REITs from their universe due to specific investment preferences or because they believe these stocks behave differently and may not fit well with their investment strategy.

How can one adjust for market capitalization filters as market conditions change over time?

One can adjust for market capitalization filters over time by using flexible restraints, such as setting limits as multiples of a stock market index, CPI, or GDP, or using ranking methods that adjust based on market conditions.

What is a boolean node in a ranking system and when should it be used?

A boolean node is used for yes or no factors in a ranking system, representing true or false values. It should be used when you want to include a factor that significantly impacts the ranking, such as excluding over-the-counter stocks or ADRs.

How can you ensure that your ranking system is not overly influenced by stocks with stale fundamentals?

To ensure that your ranking system is not influenced by stocks with stale fundamentals, you can use a command to exclude stocks with stale statements or recent corporate actions that make their fundamentals irrelevant.

What are the different node types used in ranking system design and what are their uses?

The node types used in ranking system design include composite nodes for grouping related factors, conditional nodes for different stock behaviors, boolean nodes for yes/no factors, factor nodes for shorthand factors, function nodes for custom formulas, and industry nodes for industry-based rankings.

How can you deal with rankings so that middling numbers are best?

To deal with rankings so that middling numbers are best, you can use methods like parabolic functions, absolute value differences from a target rank, or evaluation formulas to assign higher ranks to stocks that are neither too high nor too low on a given factor.