Covariance estimation: shrinkage vs. factor models
For portfolio optimization, you need a reliable covariance estimate. Two main approaches:
Shrinkage (Ledoit-Wolf)
- Pro: Simple, no model assumptions
- Con: Still N*(N-1)/2 parameters to estimate
- Pro: Dimensionality reduction, more stable
- Con: Need to choose the right factors
6 Replies
Great analysis! I've been using a similar approach with rolling z-scores and it's been working well for mean reversion signals.
The key insight is that most alpha signals decay rapidly. You need to focus on signals with a half-life of at least 5-10 days.
I ran a quick backtest on this idea and got a Sharpe of about 1.2 before costs. Not bad for a simple strategy.
The competition scoring docs could definitely be clearer. I spent 2 hours debugging what turned out to be a normalization issue.
Has anyone tried using attention mechanisms for this? The temporal attention weights could tell you which historical periods are most relevant.
The documentation for the API is at /docs -- it's OpenAPI/Swagger format. Very helpful for understanding submission formats.
Be careful with the Kelly criterion for position sizing. Full Kelly is way too aggressive. I use quarter-Kelly in practice.