The New Momentum: AI, Transformers, and the Future of Trend-Following

Day 1: The Foundations of Momentum & The Shift to AI (1.5 Hours)

*Part 1: The Anatomy of Momentum & Classical Trend-Following (45 mins)*

* The Philosophy of Momentum: Deconstructing the "buy high, sell higher" heuristic and exploring why it works, diving into efficient market anomalies, behavioral finance quirks, and institutional herding.Classical Time-Series Momentum (TSMOM): How traditional CTAs construct trend-following portfolios using moving average crossovers and MACD indicators.

* The Quant Meltdown Survival: Analyzing historical case studies where momentum strategies acted as a portfolio stabilizer during systemic shocks.

* Limitations of Heuristics: The breakdown of classical momentum during fast-reverting markets, setting the stage for AI-driven position sizing.

*Part 2: Transitioning to Deep Learning (45 mins)*

* From Heuristics to AI: Moving beyond rigid rules to Deep Momentum Networks (DMNs) that learn trend sizing dynamically.

* Optimizing for the Right Metric: Rethinking model training by designing loss functions that directly maximize risk-adjusted performance (e.g., Sharpe, Sortino) rather than simply minimizing prediction error (MSE).

* Day 1 Q&A and Wrap-up

Day 2: Transformers, Modern Architectures, and Practical Realities (1.5 Hours)

*Part 1: Inside the Momentum Transformer (45 mins)*

* Architecture of the Momentum Transformer: Discussing the Trans4mom architecture. How the self-attention mechanism relates past timesteps to construct a representation of the current market regime.

* Interpretability in AI Trading: Demystifying the "black box." We will look at how to use attention weights to see exactly how the model identifies momentum turning points and blends classical strategies based on historical significance.

*Part 2: New AI Trends, Data Bottlenecks, & Implementation (45 mins)*

* Alternative Transformer Architectures: A high-level overview of newer time-series models (e.g., Temporal Fusion Transformers, PatchTST, or Informer architectures) and how they handle long-term dependencies versus localized fluctuations.

* The Data Reality Check: Why model architecture is only half the battle. Sourcing good data matters. Discussing the crucial impact of sourcing institutional-grade data, transaction costs, and asset volatility on being able to backtest real-world Sharpe ratios. Discussing why training may stall without this.

* Implementation & Backtesting: Best practices for backtesting AI transformer models, mitigating look-ahead bias, and surviving structural market breaks.

Day 2 Q&A and Final Wrap-up

Speaker: Dr. Nick Firoozye

Dr. Nick Firoozye has over 20 years of experience in the quantitative finance industry, spanning both buy-side and sell-side firms.

He began his career at Lehman Brothers in MBS/ABS modeling and later held senior roles in Quant Research and Strategy at Goldman Sachs and Deutsche Bank. He has also worked at leading asset managers and hedge funds including Sanford Bernstein, Citadel, and Exodus Point, focusing on quantitative strategies and trading.