Notes from the workbench.
Methodology decisions, data problems, and market observations — written up as they are resolved.
Why event-driven beats calendar rebalancing
A fixed monthly rebalance samples the opportunity set at arbitrary points in time. Dividends are not uniformly distributed — they cluster around specific registration dates. Capturing each event at the moment it becomes actionable removes that sampling noise entirely.
2026-09-26
Registration date vs. disclosure date
Using the disclosure date to fetch price data introduces look-ahead bias: the market has not yet priced the event. Anchoring every calculation to the registration date, derived from the ex-rights date through the trading calendar, keeps the backtest honest.
2026-09-24
The ex-dividend gap, and why we accept it
On the ex-date the share price drops by roughly the dividend amount. The strategy assumes this gap recovers over the holding period. When it does not, realized return falls short of the signal. This is a known, accepted exposure — not a modeling error.
2026-09-22
What changes when the strategy goes to Hong Kong
The yield logic transfers. The data source, the ex-date convention, and the calibration thresholds do not. HK payouts are structurally higher-yielding, so a 3% threshold calibrated for A-shares would be far too loose.
2026-09-20