DeepVol: Volatility Forecasting with Dilated Causal Convolutions

Financial data through dilated causal convolutions to predict next-day volatility

In partnership with

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 two must-read academic papers that mix cutting-edge ML/DL with Asset Pricing & Quant Finance:

    • DeepVol: Volatility Forecasting from High-Frequency Data with Dilated Causal Convolutions

    • Supervised Autoencoders with Fractionally Differentiated Features and Triple Barrier Labelling

  • 🥐 Asset Pricing Insights: Today, I am preseting the paper: “Leveraging the Low-Volatility Effect in Modern Portfolio Management“

Today’s Sponsor

Solving a Silent Threat to Your Health

This is a paid advertisement for Med-X’s Regulation CF Offering. Please read the offering circular at https://invest.medx-rx.com

Chemical pesticides are being used in our food, homes, and offices. But Med-X has developed a 100% chemical-free pesticide that outperforms traditional solutions. It’s one of the fastest-growing products in professional pest control, with operators using it coast to coast across the USA. E-commerce partnerships with brands like Amazon, Walmart, and Kroger are now helping Med-X reach 41 new global markets in the next five years. For a limited time, you can become a shareholder as they make these moves.

“DeepVol: Volatility Forecasting from High-Frequency Data with Dilated Causal Convolutions”

👉 DeepVol processes raw high-frequency financial data through dilated causal convolutions to predict next-day volatility, eliminating the need for pre-computed realized measures.

👉 The architecture operates at 5-minute intervals using a one-day window, demonstrating that longer historical periods don't improve accuracy. Direct processing of raw data outperforms traditional approaches using handcrafted features.

👉 Testing on NASDAQ-100 data showed a 14.5% MAE reduction versus HEAVY model and 24.7% versus martingale benchmark. Out-of-sample MAE reached 7.288, with particularly strong performance during COVID-19 volatility.

👉 Model maintained accuracy across different market regimes and quickly adapted to volatility spikes. Adding realized measures as supplementary inputs provided no improvement.

👉 Implementation requires minimal data preprocessing and computational resources, making it practical for real-time trading applications.

👉 Extensive testing across multiple stocks confirms robust generalization capabilities, suggesting strong potential for portfolio-wide deployment.

“Supervised Autoencoders with Fractionally Differentiated Features and Triple Barrier Labelling”

👉 Novel approach combines supervised autoencoders with fractional differentiation and triple barrier labeling to enhance cryptocurrency trading performance. Study examines Bitcoin, Litecoin, and Ethereum from 2016-2022.

👉 Implementation leverages noisy data through balanced augmentation and optimized bottleneck size. Fractional differentiation maintains data memory while ensuring stationarity.

👉 30-minute strategy achieved highest information ratio (2.03) and starred information ratio (6.22) for Bitcoin, with lowest maximum drawdown (14.02%) versus buy-and-hold's 53.30%.

👉 Portfolio of SAE-MLP strategies outperformed buy-and-hold benchmarks despite high return correlations. 20-minute strategy delivered highest total return (619.80%) and annualized return (79.41%).

👉 Walk-forward validation demonstrates robust performance across different market regimes, including COVID-19 volatility period. Model shows strong adaptation to market changes while maintaining low drawdowns.

Today’s Sponsor

This Smart Home Company Has Already Hit $10 Million in Revenue—and It’s Just Getting Started

What if your window shades could do more than just block sunlight?

Meet RYSE, the company transforming everyday window shades into smart, automated devices. With over $10 million in total revenue, 10 granted patents, and products already available in 127 Best Buy locations, RYSE is growing at an incredible pace—200% over the last month, to be exact.

And they’re only getting started. With plans to expand internationally and partnerships with major retailers like Home Depot and Lowe’s, RYSE is set to become a household name in the rapidly growing smart home market.

Now, you have the chance to invest for just $1.75 per share and join a company poised for explosive growth.

Asset Pricing Insights

“Leveraging the Low-Volatility Effect in Modern Portfolio Management”

A new paper from Robeco researchers reveals innovative ways to capitalize on one of the most persistent market anomalies - the low-volatility effect. Keep reading! 👇

This research presents five practical strategies to leverage low-volatility investing, addressing common constraints that prevent investors from fully utilizing this powerful factor.

➡️ By combining low volatility with momentum and value factors, this strategy delivers impressive 12.5% annual returns compared to the market's 10.4%.

➡️The approach achieves a superior Sharpe ratio of 0.73 versus 0.51 for the market, while simultaneously reducing relative risk from 11.4% to 8.4%. This enhancement demonstrates how integrating multiple factors can amplify the low-volatility effect.

➡️ Moving beyond the traditional 60/40 portfolio, replacing it with a 70% Low-Vol+ and 30% bonds allocation achieves 10.6% returns while maintaining similar volatility levels.

➡️This approach boosts the Sharpe ratio to an impressive 0.82, providing significantly better risk-adjusted returns than conventional portfolios. The strategy particularly shines in preserving capital during market downturns.

➡️ Through careful application of 140% exposure to low-vol stocks, this approach delivers exceptional 15.7% annual returns compared to the market's 10.4%.

➡️This outperformance is achieved through modest leverage while maintaining strong risk control. An alternative implementation using futures requires only 30% leverage, offering a more capital-efficient solution.

➡️ This market-neutral strategy employs short positions to generate 9.6% returns with zero correlation to market movements. The approach has demonstrated remarkable consistency, producing positive returns in 83% of 12-month periods. This makes it an excellent portfolio diversifier, providing a unique, uncorrelated return stream.

➡️ Offering comparable downside protection at a fraction of the cost of traditional put options, this strategy reduces portfolio volatility from 15.4% to 9.2%. The approach proves particularly effective during moderate market declines, addressing a key weakness of put option strategies while avoiding their significant carry costs.

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! 🌟🤖💼