GJR-GARCH
Gives negative and positive return shocks different weights in conditional variance.
How it works
GJR-GARCH adds a negative-shock indicator to a GARCH variance equation. The effect of a shock can therefore differ by sign. A parameter multiplying the squared shock only for negative innovations controls the extra response, while lagged variance carries past volatility forward.
Worked example
Suppose the squared-shock coefficient is 0.05 and the extra negative-shock coefficient is 0.10. Equal shocks of +2% and −2% contribute respectively 0.05 × 0.0004 = 0.00002 and 0.15 × 0.0004 = 0.00006. The remaining terms are common to the two forecasts.
Validation and limitations
Check parameter restrictions for positive and negative shocks and the persistence formula under the assumed innovation distribution. Compare likelihood and out-of-time forecasts against plain GARCH and assess whether the asymmetric term improves portfolio risk forecasts.