Definitions

Please refer to the main definitions docs to understand the basic functions referenced below.

Targets

• target_raw_return_20/60

  • timeline: 20D2L / 60D2L

  • bins=5, uniformity=10%, 40%, 50%

  • neutralizers: none

• target_factor_neutral_20/60

  • timeline: 20D2L / 60D2L

  • bins=5, uniformity=10%, 40%, 50%

  • neutralizers: Standard Factors

• target_factor_feat_neutral_20/60

  • timeline: 20D2L / 60D2L

  • bins=5, uniformity=10%, 40%, 50%

  • neutralizers: Standard Factors and other features not listed

• target_chili_60

  • timeline: 60D2L

  • bins=5, uniformity=10%, 40%, 50%

  • neutralizers: everything and the proprietary Numerai risk model

  • other adjustments: ADV-weighted and per-level exposure constraints

Meta Models

• Signals submissions are cleaned first by:

  • tie-kept-rank each submission

  • fill nans in each submission with 0.5

  • tie-kept-rank each submission

  • gaussianize each submission

• Signals Stake-Weighted Meta Model w/ minimum stake (SSWMM)

  • Stake-weighted average of cleaned Signals submissions

• Signals Naive-Weighted Meta Model w/ minimum stake (SNWMM)

  • Average of cleaned Signals submissions

• Signals Naive-Weighted Meta Model w/ minimum 10 NMR stake (SNWMMmin10)

  • Average of cleaned Signals submissions with at least 10 NMR stake

Meta Portfolio

• neutral weights

  • applying neutralizers matrix and sample weights vector to a submission

• Stake-Weighted Portfolio (SWP)

  • Stake-weighted average of all neutral weights in Signals

Scores

• submissions are cleaned before used in scoring:

  • drop invalid tickers

  • tie-kept rank the submission

  • fill nans with 0.5

• ALPHA

  • dot product neutral weights with target_chili_60

  • timeline: 60 score days w/ 2 days returns lag (+ 2 days data lag)

• MPC - Meta Portfolio Contribution

  • Gradient of the SWP multiplied by the target_chili_60 with respect to stakes

  • timeline: 60 score days w/ 2 days returns lag (+ 2 days data lag)

• MMC - Meta Model Contribution

  • correlation contribution of a submission, SNWMMmin10, and target_factor_neutral_20

  • timeline: 20 score days w/ 2 days returns lag (+ 2 days data lag)

• CORRV4 - Correlation v4

  • numerai corr of a submission and target_factor_feat_neutral_20

  • timeline: 20 score days w/ 2 days returns lag (+ 2 days data lag)

• ICV2 - Information Coefficient V2

  • numerai corr between binned returns and submission

  • timeline: 20 score days w/ 2 days returns lag (+ 2 days data lag)

• RIC - Residual Information Coefficient

  • numerai corr between target_factor_neutral_20 and submission

  • timeline: 20 score days w/ 2 days returns lag (+ 2 days data lag)

• FNCV4 - Feature Neutral Correlation v4

  • submission is tie-kept-ranked then gaussianized then neutralized

  • tie-broken-rank correlation between target_factor_feat_neutral_20 and submission

  • neutralizers: Standard Factors and V4 Medium Safe Features

  • timeline: 20 score days w/ 2 days returns lag (+ 2 days data lag)

• CWSNMM - Corr w/ Signals Naive Meta Model

  • s` = tie-kept rank, then gaussianize, then pow 1.5 a submission s

  • calculate pearson correlation between s` and SNWMM

  • timeline: 4 days data lag / not dependant on returns

• MCWSM - Max Corr w/ Signals Models

  • Maximum pearson correlation of a submission with any other Signals submission

  • only compared to other submissions made in the same round

  • timeline: 4 days data lag / not dependant on returns

• APCWSM - Average Pairwise Corr w/ Signals Models

  • Average pearson correlation of a submission with each other Signals submission

  • only compared to other submissions made in the same round

  • timeline: 4 days data lag / not dependant on returns