The measurement
- Schema
- canli.alphac-prereg-parameters.v1
Committed parameters
- Hedge beta window sessions
- 252
- Hold sessions
- 63
- Minimum aligned observations
- 252
- Minimum stressed observations
- 63
- Momentum lookback session index
- -252
- Momentum lookback sessions
- 252
- Momentum skip session index
- -21
- Momentum skip sessions
- 21
Note. Stated in the document as '12-1 momentum ... adjusted close at session -21 divided by adjusted close at session -252', a trailing 252-session SPY beta hedge, and void conditions below 252 aligned or 63 stressed observations.
Document
- Path
- docs/design/PREREG_EARNINGS_NARRATIVE_CHANGE.md
- Public path
- /research/prereg-earnings-narrative-change
- Sha256
- sha256:60486da41f1e2c88e62b6e56d436d780f645420ffaeb13364540d0c4040cc3b9
Marginal test weights
- Candidate weight
- 0.100000
- Current sleeves
- 4
- Sleeve weight after
- 0.225000
- Sleeve weight after pct
- 22.500000
- Sleeve weight before
- 0.250000
Derivation. 4 equal sleeves at 25% each; the candidate is assigned 10% funded PRO RATA, so every existing sleeve keeps 90% of its weight: 0.25 x 0.90 = 0.225. This is arithmetic, not a measurement, and it is published with its inputs so it can be checked rather than accepted.
Measured book effect
- Base sharpe
- 0.956304
- Candidate 10pct sharpe
- 1.064387
- Source
- artifacts/probe/earnings_narrative_change/result.json
Note. Included so the weight the spec commits to and the weight the run used are visible on one page.
Parameters found verbatim in the document
- -21
- -252
- 21
- 252
- 63
Why this exists. A pre-registration's function is that the spec is fixed before measurement so a reader can check the run against it. These parameters existed only as English sentences, so nothing could compare the executed run to the committed spec.
Check it yourself
Every figure above is read from /glassbox/prereg_earnings_narrative_parameters.json, the
artifact the engine wrote.