HK IPO AI metric guide
How to read the metrics shown on HK IPO AI pages, including their limits. They are not return or allocation guarantees.
P90 relative error
P90 relative error means about 90% of the comparable errors in the backtest set were at or below this level.
Read explainerTime-weighted relative error
Time-weighted relative error measures proportional deviation between predicted and official allotted lots, with more influence from recent listings and key tiers.
Read explainerExpected lots
Expected lots is the model-estimated average allocation. It is not a guaranteed number of lots.
Read explainerReference rating
The reference rating is an informational rating with three labels: high attention, neutral and cautious.
Read explainerBacktest hit rate
Backtest hit rate is the share of historical samples where the model's direction call matched the actual result.
Read explainerTop2 coverage
Top2 coverage checks whether the actual Pool A and Pool B patterns land in the two listed candidate scenarios.
Read explainerSample weight
Sample weight is the relative influence a historical record has in up/down backtest statistics.
Read explainerGrey market anchor
Grey market anchor uses grey market return versus listing price as one input for the first-day range.
Read explainerData coverage
Data coverage is the share of applicable rating inputs that are already available at the current stage.
Read explainerMedian relative error
Median relative error is the midpoint of comparable relative errors and describes a typical case.
Read explainerTop-1 scenario hit rate
Top-1 scenario hit rate is the share of historical samples where the default Pool A/B pattern matched the official result.
Read explainerModel basis
Model basis describes whether demand, deal terms and historical comparison inputs are sufficient. It is shown as strong, moderate or limited.
Read explainerTrend confirmation
Trend confirmation is the backtest reading: it uses the intraday path, not only close versus offer price.
Read explainer