Cross-section without factors: a string model for expected returns

Hi! Here's Iván with this week's exciting newsletter, brimming with insights and discoveries on building robust investment strategies and risk models using Machine Learning.

In this edition, I am presenting the following sections:

• 🕹️ AI-Finance Insights: I summarize three must-read academic papers that mix cutting-edge ML/DL with Asset Pricing & Quant Finance:

  • Cross-section without factors: a string model for expected returns

  • Three Types of Backtests

• 💊 AI Essentials: The section on top AI & Quant Finance learning resources: Today, I'm excited to share a 1-hour video that explores how finance teams can significantly enhance their efficiency using AI and automation!

• 🥐 Asset Pricing Insights: In this edition, I recommend a paper that combines the well-known volatility timing strategy with ML volatility prediction models, which are simple but effective.

Today’s Sponsor

Don’t just Copy, Create your Strategy

Unlock hedge fund-level strategies with Surmount's platform. Craft, analyze, execute instantly. Full Code or No-code builders, advanced analytics elevate your portfolio precision. Experience the future: advanced tools, comprehensive data for market conquest.

Start elevating your trading today!

“Cross-section without factors: a string model for expected returns”

The paper introduces a new string model, shifting the paradigm in asset pricing by focusing on correlations among assets instead of traditional common factors.

This string model refines traditional approaches to provide a more nuanced understanding of asset returns, leading to new insights into the correlation premium.

The main contributions and findings are as follows:

👉 The string model eliminates the need for common factors, explaining asset returns through granular exposure to other assets' returns. This correlation premium captures the detailed co-movements among asset returns.

👉 Unlike traditional models, the string model directly models correlations, leading to more accurate predictions of expected returns. This approach simplifies the complexity associated with factor models.

👉 Empirical tests show that the string model performs on par with established models like the Fama-French, especially in predicting the cross-section of expected returns. The model is validated using portfolios sorted by book-to-market, momentum, and other characteristics.

👉 The model reveals that big stocks act as correlation hedges, contributing negatively to the correlation premium. Portfolios with higher exposure to these stocks command lower premiums, especially in turbulent times.

👉 Time-varying correlations are a key feature, with the model predicting higher correlations in bad times. This makes the expected returns predictable and highlights the importance of understanding correlation dynamics.

👉 The model distinguishes between the correlation premium and the premium for correlation risk. The latter compensates for the randomness in correlations, providing additional insights into the behavior of returns under different market conditions.

“Three Types of Backtests”

The paper analyzes role of backtesting in developing systematic investment strategies by exploring three primary backtesting methods: walk-forward testing, resampling, and Monte Carlo simulations.

Here are the main contributions and findings:

👉 Walk-Forward Testing:

🔎 Benefits: Simple to apply, easy to analyze, and interpret.

🔎Challenges: Path dependency and potential overfitting, assuming past processes will repeat in the future. Tests only a single path.

👉 Resampling Methods:

🔎Techniques: Includes cross-validation and bootstrapping.

🔎Benefits: Provides multiple evaluation paths, reduces overfitting risk, and offers robust performance metrics.

🔎Challenges: May not accurately represent future performance and some data may be unsuitable for bootstrapping without modification.

👉 Monte Carlo Simulations:

🔎Benefits: Generates additional data resembling real data, offers a solid foundation for analysis, and models future paths using theoretical constructs or causal relationships.

🔎Challenges: Complex creation of accurate data generation processes and difficulty replicating extraordinary events. Relies on the correctness of the data generation process.

👉 The paper emphasizes the importance of enhancing backtest quality by focusing on:

🔎Data Quality: Avoiding survivorship bias, ensuring point-in-time accuracy, correcting errors, and managing missing data.

🔎Statistical Integrity: Avoiding data snooping, addressing selection bias, and ensuring rigorous statistical tests.

🔎Modeling and Generalization: Avoiding look-ahead bias and incorporating an embargo period.

🔎Costs and Constraints: Accounting for transaction costs, short-sale constraints, and liquidity constraints.

🔎Performance Evaluation: Using causal graphs, holistic performance metrics, and peer review to validate findings.

👉 To mitigate the negative consequences of selection bias under multiple testing, the paper introduces methods like the Deflated Sharpe Ratio and controlling for the family-wise error rate (FWER). These techniques help in avoiding overfitting and making more reliable discoveries.

Sponsor 👇

AI & ChatGPT Mini Crash Course - Eliminate workplace burnout & save 16+ hours every week. Learn 20+ AI tools, prompting techniques & hacks for free.

Save your seat here (first 100 people only) ⏰

AI-Essentials

Dive into an insightful 1-hour video that explores how finance teams can significantly enhance their efficiency using AI and automation! This presentation, featuring interviews and discussions, is particularly helpful for those new to the practical applications of artificial intelligence in the financial sector. Whether you're just beginning to learn about AI or looking to understand its impact on financial operations, this video provides the perfect overview. 👇

Asset Pricing Insights

“How to Incorporate Market Timing with ML-Predicted Volatility?”

Volatility timing is one of the best-known methods for timing investments. In this paper, they enhance this idea by simply improving volatility prediction with ML.

Here my takeaways: 👇

➡ The paper employs ML, notably model averaging and the LASSO method, to predict stock market volatility more accurately than traditional models. This enhanced forecasting capability is crucial for effective asset allocation.

➡ By integrating predictive information from various drivers, the research effectively forecasts market volatilities using high-dimensional models, which are shown to outperform standard volatility models in predicting performance.

➡ The paper highlights the construction of volatility timing portfolios that generally yield higher Sharpe ratios and risk-adjusted returns compared to static market portfolios, especially those timed with LASSO-based forecasts, across various forecasting horizons.

➡ It presents empirical results where volatility timing strategies surpass the static market portfolio in terms of average returns, with LASSO-driven portfolios achieving the best overall investment performance, indicating at least 24% to 42% higher Sharpe ratios than the market.

➡ Finally, the findings are summarized with visual evidence of cumulative log-returns (see image), showcasing that LASSO portfolios not only consistently outperform the market portfolio over a 14-year test period but also command the highest cumulative returns.

If you're enjoying our newsletter and want to support us, please recommend it to anyone you know who's interested in AI and Finance. Your referrals are the biggest compliment and help us grow! 🌟🤖💼