Model families and native outputs
Latent factors

PCAModel, RPPCAModel, IPCAModel, CAEModel

PCA and RP-PCA use PersistentPanelBatch. IPCA and CAE use CrossSectionBatch.

Native outputs are loadings or conditional betas, factor history, and structural state.

Stochastic discount factor

StochasticDiscountFactorModel

Input: CrossSectionBatch.

Outputs: asset_weights, sdf_values, and phase-aware checkpoints.

Direct asset prediction

SAEModel

Input: CrossSectionBatch.

Output: cross-sectional signals in AssetSignalResult.

Portfolio learning

LinearFeaturePortfolioModel, LSTMPortfolioModel, DeepPortfolioModel

Input: PortfolioSequenceBatch.

Output: target weights in PortfolioWeightsResult.

Neural learners can include trading costs in their training objective.

Latent-factor return forecasts require three stages
  1. Structural estimation produces LatentFactorState.
  2. A separate factor-premium forecaster produces FactorForecastResult.
  3. An asset mapper combines state and forecast into AssetForecastResult.
Downstream handoff

PredictionsFrame carries forecasts or direct signals. WeightsFrame and ContextFrame support backtests and portfolio diagnostics.