dcegm.backward_induction

Interface for the DC-EGM algorithm.

Functions

backward_induction(→ Tuple[jax.numpy.ndarray, ...)

Do backward induction and solve for optimal policy and value function.

Module Contents

dcegm.backward_induction.backward_induction(params: Dict[str, float], income_shock_draws_unscaled: jax.numpy.ndarray, income_shock_weights: jax.numpy.ndarray, model_config: Dict[str, Any], model_funcs: Dict[str, Callable], model_structure: Dict[str, Any], batch_info: Dict[str, Any]) → Tuple[jax.numpy.ndarray, jax.numpy.ndarray, jax.numpy.ndarray]

Do backward induction and solve for optimal policy and value function.

Parameters:

  • params (dict) – Dictionary containing the model parameters.

  • income_shock_draws_unscaled (np.ndarray) – 1d array of shape (n_quad_points,) containing the Hermite quadrature points unscaled.

  • income_shock_weights (np.ndarrray) – 1d array of shape (n_stochastic_quad_points) with weights for each stoachstic shock draw.

  • model_config (dict) – Dictionary containing the model configuration.

  • model_funcs (dict) – Dictionary containing model functions.

  • model_structure (dict) – Dictionary containing model structure.

  • batch_info (dict) – Dictionary containing batch information.

Returns:

Tuple containing the period-specific endog_grid, policy, and value

from the backward induction.

Return type:

Tuple