Class Egarch11

java.lang.Object
com.quantfinlib.volatility.Egarch11
public final class Egarch11 extends Object
EGARCH(1,1) — Nelson's exponential GARCH, the LOG-variance dynamics the plain family cannot express: 
  ln h_t = ω + β·ln h_{t-1} + α(|z_{t-1}| − √(2/π)) + γ·z_{t-1}
with standardized shocks z = r/√h. Two things the log form buys: no positivity constraints AT ALL (any parameter signs give a valid variance — the exp does the work Garch11's ω,α,β ≥ 0 constraints do), and leverage as a SIGN — γ < 0 means a down move raises tomorrow's volatility more than an equal up move, and the magnitude reads directly. Stationarity is |β| < 1, nothing else.

Estimation mirrors the family: Gaussian MLE over a coarse-to-fine grid spanning the EMPIRICALLY PLAUSIBLE box — α ∈ [0, 0.9], γ ∈ [−0.9, 0.9], β ∈ [0, 0.995]. Negative α or β, while formally admissible in the log form, are not searched (stated, not hidden: they describe oscillating log-variance no asset-return series exhibits). ω is targeted to the sample's log variance (ω = (1−β)·ln σ̄² — an approximation, since E[ln h] ≤ ln E[h] by Jensen; stated, not hidden). One-step-ahead nextVariance(double[], com.quantfinlib.volatility.Egarch11.Params) is exact; multi- step forecasts are deliberately NOT offered — iterating the log recursion forecasts the MEDIAN variance, not the mean, and quietly returning it as "the forecast" is the kind of lie this library refuses (Garch11/GjrGarch11 forecast multi-step honestly; use them when you need horizons). Research lane.

  • Method Details

    • fit

      public static Egarch11.Params fit(double[] returns)
      Fits EGARCH(1,1) to (demeaned) returns by grid MLE.

    • conditionalVariances

      public static double[] conditionalVariances(double[] returns, Egarch11.Params p)
      Conditional variance series under the fitted parameters.

    • nextVariance

      public static double nextVariance(double[] returns, Egarch11.Params p)
      One-step-ahead variance — EXACT (tomorrow's ln h is deterministic today).