numeraire.core.engine.ForecastOutput#

class numeraire.core.engine.ForecastOutput(forecasts: DataFrame, realized: DataFrame, benchmark: DataFrame, method: str, config_hash: str, data_vintage: str, run_id: str, capability: str = 'to_forecast', meta: dict[str, ~typing.Any]=<factory>, horizon: int = 1)[source]#

Bases: object

OOS output for a to_forecast method: per-origin forecast, realized return, benchmark.

All three are (origin x asset) indexed by the forecast origin t: forecasts.loc[t] predicts the return over (t, t+h], realized.loc[t] is that realized return, and benchmark.loc[t] is the prevailing/window historical-mean forecast the engine computes for free at each origin (the Goyal-Welch OOS R^2 reference).

__init__(forecasts: DataFrame, realized: DataFrame, benchmark: DataFrame, method: str, config_hash: str, data_vintage: str, run_id: str, capability: str = 'to_forecast', meta: dict[str, ~typing.Any]=<factory>, horizon: int = 1) → None#

Methods

__init__(forecasts, realized, benchmark, ...)

Attributes

`capability`
`horizon`	Effective forecast horizon `h` (steps of the decision calendar); drivers set it explicitly.
`universe`	Compact universe label (`n=<#assets>` for panels, the name for a single asset).
`forecasts`
`realized`
`benchmark`
`method`
`config_hash`
`data_vintage`
`run_id`
`meta`

horizon: int = 1#

Effective forecast horizon h (steps of the decision calendar); drivers set it explicitly.

forecasts.loc[t] predicts and realized.loc[t] records the return over (t, t+h]; the benchmark carried alongside is the historical mean compounded to the same h-period target.

property universe: str#: Compact universe label (n=<#assets> for panels, the name for a single asset).