Readonlyaic−2·ln L + 2k, as arch reports it.
ReadonlybackcastThe pre-sample variance the recursion started from — its log for EGARCH.
Readonlybic−2·ln L + k·ln n, as arch reports it.
Readonlyconditionals_t for every observation, in the returns' units squared.
ReadonlyconvergedReadonlyextendedReturns appended by extendFit since the parameters were estimated: 0 for
a fit from fitGarch. Above 0, the parameters, inference and modes
are the estimate's; the variance path, lastResidual, the
log-likelihood and the criteria cover the whole series.
ReadonlyinferenceReadonlylastGARCH-X: x_T, the last exogenous value — what s_{T+1} takes. null without one.
Readonlylaste_T, the last residual.
ReadonlylogReadonlymeanReadonlymodelReadonlymodesEvery distinct optimum the starts reached, best first — modes[0] is the
fit. More than one means the likelihood is multimodal on this series.
ReadonlynuFor t innovations: the ν the likelihood wanted lies beyond a bound, and the fit is the constrained optimum with ν held there. null when ν is interior, and for normal innovations.
ReadonlyparameterThe free parameters the criteria count: μ (constant mean), ω (unless targeted), α, β, γ (gjr, egarch), ν (t, unless held on a bound), λ (skewed t).
ReadonlyparamsReadonlypersistenceReadonlyvarianceUnder variance targeting, the unconditional variance ω was tied to: the returns' variance about their sample mean (about 0 for a zero mean). null when ω was estimated.
A fitted model: its parameters, likelihood and criteria, variance path, the optima reached, and inference when asked.