Class IcSignificance

java.lang.Object
com.quantfinlib.microstructure.IcSignificance
public final class IcSignificance extends Object
Sampling error of an exponentially weighted correlation, used to gate learned signals on significance rather than on the sign of a noisy IC.

An EWMA with weight α after k observations puts weight w_j = α(1−α)^j on the j-th most recent one. Its Kish (1965) effective sample size is (Σw)²/Σw²:

  n_eff(k) = (1 − (1−α)^k)² · (2 − α) / (α · (1 − (1−α)^{2k}))

which tends to (2 − α)/α (about 199 at α = 0.01). Under the null of zero correlation the sample correlation has standard error close to 1/√n_eff. A true-zero component therefore shows a positive IC half the time; gating on IC > 0 admits noise, gating on IC/SE > z (z = 2 by default) admits it about 2.3% of the time.

  • Field Summary

    Fields
    Modifier and Type
    Field
    Description
    static final double
    DEFAULT_Z
    Default one-sided gate, about the 97.7th percentile of N(0,1).

  • Method Summary

    Modifier and Type
    Method
    Description
    static double
    effectiveSamples(double alpha, long samples)
    Kish effective sample size of an EWMA with weight alpha after samples.
    static double
    shrunkWeight(double ic, double alpha, long samples, double z)
    The IC shrunk toward zero by z standard errors: max(0, IC − z·SE).
    static double
    standardError(double alpha, long samples)
    Null standard error of the EWMA correlation, 1/√n_eff; +∞ before any sample.

    Methods inherited from class Object

    clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

  • Field Details

    • DEFAULT_Z

      public static final double DEFAULT_Z
      Default one-sided gate, about the 97.7th percentile of N(0,1).
      See Also:

  • Method Details

    • effectiveSamples

      public static double effectiveSamples(double alpha, long samples)
      Kish effective sample size of an EWMA with weight alpha after samples.

    • standardError

      public static double standardError(double alpha, long samples)
      Null standard error of the EWMA correlation, 1/√n_eff; +∞ before any sample.

    • shrunkWeight

      public static double shrunkWeight(double ic, double alpha, long samples, double z)
      The IC shrunk toward zero by z standard errors: max(0, IC − z·SE).