@krzysztofsztompka/garch
    Preparing search index...

    Type Alias VolatilityForecast

    volatilityForecast's answer: the as-of fit and forecast, and the model graded walk-forward against its baseline.

    type VolatilityForecast = {
        aic: number;
        bic: number;
        conditionalVariance: number[];
        converged: boolean;
        forecast: number[] | null;
        gammaPValue: number | null;
        logLikelihood: number;
        mean: MeanModel;
        model: GarchModel;
        nuBound: "lower" | "upper" | null;
        params: GarchParams;
        persistence: number;
        rivals: readonly RivalMode[];
        validation: Validation;
        varianceTarget: number | null;
    }
    Index
    aic: number

    See GarchFit: −2·ln L + 2k and −2·ln L + k·ln n.

    bic: number
    conditionalVariance: number[]

    One per return, in the returns' units squared.

    converged: boolean
    forecast: number[] | null

    s_{T+1} … s_{T+h}; null when the fit did not converge.

    gammaPValue: number | null

    p-value of γ (robust); null for the symmetric model.

    logLikelihood: number
    mean: MeanModel
    model: GarchModel
    nuBound: "lower" | "upper" | null

    t innovations only: ν held on a bound (see GarchFit.nuBound).

    params: GarchParams
    persistence: number
    rivals: readonly RivalMode[]

    The other optima whose log-likelihood is within rivalsWithin of the fit's — the modes the data cannot tell from it. Empty unless asked.

    validation: Validation

    Against the next-simpler model: naive trailing variance for garch, symmetric garch for gjr.

    varianceTarget: number | null

    See GarchFit.varianceTarget: null unless ω was targeted.