Managing the Market Portfolio

Hi! Here's Iván from Noax Capital 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:

• 📈 Noax Capital Updates : 📢 New Section! As our quantitative investment firm grows, I've decided to start sharing interesting updates through this newsletter. Today, I am sharing with you our reatail fund's monthly performance report (Oct 2024).

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

  • Managing the Market Portfolio

  • TradExpert: A New Approach to Stock Trading Using AI

• 💊 AI Essentials: The section on top AI & Quant Finance learning resources: Today, I'm sharing an insightful video lecture on "Deep Learning and Macro-Finance Models" by Gopalakrishna from École Polytechnique Fédérale de Lausanne. This comprehensive overview bridges deep learning theory with financial applications, making complex concepts accessible. If you're interested in understanding how neural networks are reshaping quantitative finance, this is an excellent resource to start!

• 🥐 Asset Pricing Insights: In this edition, I introduce "Deep Partial Least Squares for Empirical Asset Pricing," a groundbreaking study that combines deep learning with statistical techniques to predict stock returns more effectively while maintaining model interpretability.

NOAX: October 2024 Results

We're launching a new section to discuss Noax Capital updates. In this case, I want to share the fund's performance results for retail investors during October. The fund is only available for European retail investors, but we are open to engaging with institutional investors globally (SMAs & Sub-advising).

For institutional mandates, please contact directly: [email protected]

“Managing the Market Portfolio”

👉 The authors show that timing the market using predictor variables significantly enhances investment opportunities beyond static factor strategies. A combination-forecast-managed market portfolio generates significant alphas of >5% annually relative to all major factor models.

👉 Empirical Results:

🔓 $1 invested in 1972 grows to $85.84 by 2019 using their managed strategy vs only $15.03 for buy-and-hold.

🔓Strategy achieves Sharpe ratio gains of 0.193 vs CAPM (from 0.443 to 0.636).

🔓Performs exceptionally well in recessions with alphas >19%, while maintaining significant alphas during expansions

👉 Previous factor returns can effectively manage the market portfolio, with investment, profitability, and momentum factors proving particularly successful. The factor-combination-forecast strategy demonstrates notable performance, generating annual alphas exceeding 5% compared to multifactor models.

👉 First moment (return) predictability proves more important than second moment (volatility) predictability. Interest rates and economic ratios emerge as the strongest predictors, while the strategy maintains its profitability even after accounting for transaction costs. The performance demonstrates robustness across different time periods and market conditions.

👉 Sum up: The paper, comprehensively analyzes the connection between market predictability and cross-sectional pricing, showing how timing strategies can significantly improve upon static factor investing approaches.

“TradExpert: A New Approach to Stock Trading Using AI”

👉 This paper introduces TradExpert, a trading system that combines multiple AI models for stock market analysis. The system uses four specialized AI models, each focusing on a different aspect: news analysis, price movements, trading patterns, and company financials. A fifth AI model then combines these insights to make final decisions.

👉 The main advantage of TradExpert is its ability to process multiple data types simultaneously - from written news to numerical data to company reports. The system can perform two main functions: predicting stock price movements and ranking stocks for investment purposes.

👉 Testing shows solid performance metrics. TradExpert achieved 64% accuracy in predicting S&P500 stock movements, performing better than existing systems. In trading simulations, it generated 49.79% annual returns with a 5.01 Sharpe ratio ( ⚠ I know, maybe too good to be true), while maintaining 9.95% volatility - numbers that indicate strong performance with managed risk levels.

👉 The researchers have released a comprehensive market dataset alongside their findings. The system's design mirrors traditional investment firm structures, where different specialists analyze various aspects of potential investments before making collective decisions.

👉 This research demonstrates how AI can be applied to financial markets by combining multiple specialized models, potentially offering a new approach to systematic trading strategies.

AI-Essentials

In this edition, we're featuring a comprehensive mini-lecture on Deep Learning and Macro-Finance Models by Gopalakrishna (École Polytechnique Fédérale de Lausanne, Swiss Finance Institute). From neural network basics to practical implementations in Google Colab, this lecture covers essential concepts for finance professionals and AI enthusiasts.

Asset Pricing Insights

“Deep Partial Least Squares for Empirical Asset Pricing”

👉 This paper introduces DPLS as a novel methodology that combines Partial Least Squares (PLS) with deep learning to model and predict stock returns. The key innovation is using deep learning to capture non-linear relationships between firm characteristics and returns while still maintaining the interpretability and efficiency of traditional factor models.

👉 DPLS works by projecting firm characteristics onto a smaller set of dynamic risk factors, maximizing the covariance between characteristics and returns. Unlike other deep learning approaches, DPLS corresponds to a non-linear stochastic discount factor, making it theoretically well-grounded in asset pricing theory.

👉 Testing on Russell 1000 stocks (1989-2018) shows impressive performance metrics. DPLS achieved 6.9% out-of-sample R², significantly outperforming both LASSO (5.9%) and standard neural networks (0.1%). In portfolio simulations, it generated a 0.925 information ratio, beating neural networks at 0.783 (⚠️ before transaction costs). The model also proved computationally efficient, training 2-20x faster than comparable neural networks while using 10x fewer parameters.

👉 The model excelled at identifying market outliers and factor interactions, which is particularly valuable during periods of market stress. During the 2008 financial crisis, for example, DPLS identified key non-linear relationships that traditional linear models missed. The framework also provides clear attribution of portfolio risk to interpretable factors, making it practical for real-world investment applications.

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! 🌟🤖💼