Improving Momentum with Online Changepoint Detection

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:

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

  • Improving Time-Series Momentum Strategies with Online Changepoint Detection

  • Daily Crypto Market Prediction and Trading with Machine Learning Models

  • Machine Learning vs. Economic Restrictions: Evidence from Stock Return Predictability

• 💊 AI Essentials: The section on top AI & Quant Finance learning resources: Today, I'm thrilled to share an outstanding video from the Algorithmic Trading Club at the University of Washington Seattle. In this 10-minute session, "How AI Traders Will Dominate the Hedge Fund Industry," you'll explore the fascinating advancements in AI within the realm of financial trading.

• 🥐 Asset Pricing Insights: In this edition, I recommend the study "Factor Momentum and the Momentum Factor," which delves into the fascinating connection between individual stock momentum and factor momentum. This paper reveals a fresh perspective on momentum trading by demonstrating how these two types of momentum are closely linked.

Today’s Sponsor

Why This Gold Stock is Our Top Trade of the Month

Tiny Float: Just 19 million shares outstanding with strong insider ownership.
Stock Performance: Up 200% in recent years and primed for the next breakout.
Ideal Timing: Upcoming catalysts + gold pullback = prime opportunity.

Discover This Hidden Stock Pick

This is a sponsored advertisement on behalf of Four Nines Gold. Past performance does not guarantee future results. Investing involves risk. View the full disclaimer here: https://shorturl.at/73AF8

AI-Finance Insights

“Improving Time-Series Momentum Strategies with Online Changepoint Detection”

This paper introduces a novel approach to enhance momentum trading strategies by integrating an online changepoint detection (CPD) module into a Deep Momentum Network (DMN) using an LSTM architecture. This method improves the model's adaptability to rapid market changes, particularly during significant events like the 2020 market crash.

The main takeaways and results are:

👉 The addition of the CPD module leads to a 33% improvement in the Sharpe ratio over the period from 1995 to 2020, with a performance boost of approximately two-thirds from 2015 to 2020 when traditional momentum strategies underperformed. The model also achieves significant gains with very small Lookback Windows (LBWs).

👉 The CPD-enhanced DMN outperforms the standard DMN across all metrics, including higher profit-to-loss ratios, reduced volatility, and lower maximum drawdown (MDD). The Sortino ratio improves by 35%, and the Calmar ratio by 25%.

👉 The enhanced DMN shows a 130% improvement in Sharpe ratio compared to the best traditional time-series momentum (TSMOM) strategy, demonstrating superior risk-adjusted performance and robustness to market crashes and nonstationary conditions.

“Daily Crypto Market Prediction and Trading with Machine Learning Models”

This paper explores various machine learning models for predicting daily cryptocurrency market movements and trading. The study focuses on the 100 largest cryptocurrencies.

The main takeaways and results are:

👉 Predictive Accuracy: The models achieve prediction accuracy ranging from 52.9% to 54.1% for all cryptocurrencies, which increases to 57.5% to 59.5% for the top 10% of predictions with the highest model confidence.

👉 Trading Strategy: A long-short portfolio strategy based on LSTM and GRU ensemble models yields impressive annualized Sharpe ratios of 3.23 and 3.12, respectively, after accounting for transaction costs. In comparison, the buy-and-hold benchmark strategy yields a Sharpe ratio of only 1.33.

👉 Model Performance: The Random Forest (RF) model performs best overall with an accuracy of 54.2%, followed closely by the LSTM at 54.1%. For specific portfolio sizes, the GRU model outperforms others, achieving an accuracy of 59.5% for k=5 and 61.0% for k=2.

👉 Robustness Over Time: The model's predictive accuracy varies over different study periods, with all models consistently performing significantly better than random chance. The GRU model performs best in the first study period with an accuracy of 60.8%.

“Machine Learning vs. Economic Restrictions: Evidence from Stock Return Predictability”

This paper explores the economic significance of machine learning methods in predicting stock returns, comparing them with traditional economic restrictions.

👉 ML methods identify mispriced stocks in line with traditional anomaly-based trading strategies. Notably, stocks in long positions tend to be small, value, illiquid, and old with low prices and low beta. Despite the opaque nature of machine learning models, they generate economically interpretable trading strategies.

👉 ML-based strategies significantly outperform traditional strategies. The average long-short portfolio return is 1.81% per month across all stocks and 1.55% after adjusting for the FF6 model over the 1987–2017 sample period. Specifically, intra-industry strategies account for 84% of the unconditional raw returns and 93% of risk-adjusted returns.

👉 ML portfolios remain profitable, particularly during periods of high market volatility and low liquidity. The results show that intra-industry strategies yield higher returns than inter-industry strategies, confirming that machine learning signals are more effective for stock selection rather than industry rotation.

Sponsor 👇

You missed Amazon. You don't have to again.

Amazon, once a small online bookstore, grew into a global behemoth, transforming industries along the way. Now, imagine yourself at the forefront of the next revolution: AI. In The Motley Fool's latest report, uncover the parallels between Amazon's early trajectory and the current AI revolution. Experts predict one of these AI companies could surpass Amazon's success with market caps nine times larger. Yep, you read that right. Don't let history repeat itself without you. Sign up for Motley Fool Stock Advisor to access the exclusive report.

Sign up today

AI-Essentials

📹 Check out this engaging session from the Algorithmic Trading Club at the University of Washington Seattle. In this 10-minute video, you'll dive into “How AI Traders Will Dominate the Hedge Fund Industry?”. 👇

Asset Pricing Insights

“Factor Momentum and the Momentum Factor”

This paper uncovers a novel perspective on momentum trading by revealing how individual stock momentum is intertwined with factor momentum. The research presents compelling evidence that most factors exhibit positive autocorrelation, significantly influencing stock returns.

The main takeaways and results are:

👉 Factors show strong momentum, with the average factor earning 51 basis points (bps) per month following a year of gains, compared to just 6 bps after a year of losses. This difference is highly significant, with a t-value of 4.22.

👉 Mom is more pronounced in factors that explain a greater cross-section of returns. Strategies trading the top 10 high-eigenvalue principal components (PCs) achieve a five-factor model alpha with a t-value of 6.51, highlighting their robust performance.

👉 Factor mom largely subsumes various forms of individual stock momentum. When a momentum factor derived from high-eigenvalue PCs is included, traditional stock momentum strategies lose statistical significance, showing that momentum times other factors rather than acting as an independent risk factor.

👉 Residual mom strategies, which select stocks based on CAPM residuals, show higher profitability than those based on total past returns. However, as more factors are removed, the effectiveness of these strategies diminishes. No residual momentum strategy remains significant when accounting for factor momentum, emphasizing the importance of considering multiple factors in momentum trading.

These findings suggest that to maximize returns, investors should focus on factor mom, particularly high-eigenvalue PCs, to capture robust and consistent gains.

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! 🌟🤖💼