The day in conversation for 2026-11-03
discussion
Transcript
The discussion: Iris: Those deep learning architectures, ModernTCN and PatchTST leading in that 918 run—that’s impressive for forecasting across multiple horizons, but what did they find regarding model stability when moving from asset classes with highly correlated returns to those with lower correlation? Warren: I'm less focused on the architecture comparison itself right now, and more on the optimization side. The idea of admissible portfolio optimization linking causality to risk through those orders is interesting, but how robust are those causal identifiers when the underlying market dynamics shift rapidly, perhaps during volatility spikes that we saw last quarter? Ruth: That shift in dynamics is exactly where I think the forecasting models need to adapt. If PatchTST performs well because it handles long-range dependencies efficiently, does that imply its inherent structure gives it a better ability to filter out transient noise compared to the TCN variant when looking at cross-asset dependencies, or are we just seeing good parameter tuning mitigating structural weaknesses? Iris: The difference lies in how the constraints are defined. In my view, if the admissible constraints themselves rely on historical covariance matrices derived from stable regimes, those frameworks might over-constrain a portfolio during regime shifts; could you quantify the expected tracking error penalty when switching from a fixed constraint set to one that dynamically adjusts based on short-term causality signals? Warren: That brings us back
More episodes
- The day in conversation for 2026-11-19
- Daily Summary for 2026-11-19
- The day in conversation for 2026-11-18
- Daily Summary for 2026-11-18
- 2601.05274-On the use of case estimate and transactional payment data in neural networks for individual loss reserving
- 2009.00868-Loss-Given-Default Modeling by Post-Last Passage Time Process
- 2606.30070-Financial Resilience Evaluation: From Conditional Expectations to Dynamic Convex Risk Measures
- 2212.12687-Measuring price impact and information content of trades in a time-varying setting
- Daily Summary for 2026-11-17
- 2401.08323-Dynamic portfolio selection under generalized disappointment aversion