Engle-Ng (1993) Measuring and Testing the Impact of News on Volatility

Summary

This paper defines the news impact curve — a graphical and analytical device for showing how a volatility model translates the latest return shock into next period's conditional variance — and uses it to compare symmetric and asymmetric ARCH models. Motivated by the leverage effect (bad news raises volatility more than good news of the same size), Engle and Ng introduce new diagnostic tests for asymmetry, propose a partially nonparametric (PNP) model that lets the data trace out the news-volatility relation, and estimate the competing models on daily Japanese (TOPIX) stock returns. They conclude that the Glosten-Jagannathan-Runkle (GJR) model is the best parametric specification, while EGARCH captures the asymmetry but implies an excessively variable conditional variance.

Key Claims

• The news impact curve (NIC). Holding all information dated $t-2$ and earlier fixed at the unconditional variance, the NIC plots the conditional variance $h_t$ as a function of the single shock $\varepsilon_{t-1}$. A symmetric GARCH gives a symmetric parabola centered at zero ($h_t=A+\alpha\varepsilon_{t-1}^2$); asymmetric models shift or tilt it so negative shocks raise variance more.
• Asymmetry / leverage. Good and bad news predict volatility differently. The NIC of EGARCH (Nelson 1991) and GJR-GARCH is centered away from zero and steeper for negative shocks, encoding the empirically documented leverage effect (Black 1976; Christie 1982).
• Three misspecification tests. The sign bias, negative size bias, and positive size bias tests regress squared standardized residuals on indicators/interactions of the sign and size of lagged shocks; significance signals that the fitted volatility model has missed asymmetry, without needing to nest the alternative.
• Partially nonparametric model. A PNP specification places the news impact on a flexible piecewise-linear grid in $\varepsilon_{t-1}$, letting the data reveal the shape of the news-volatility relation rather than imposing it.
• Empirical ranking. On daily TOPIX returns, GJR is the best parametric model; EGARCH captures most of the asymmetry but its implied conditional variance is too variable — a caution about EGARCH's log specification.

Concepts Introduced or Extended

• GARCH and BEKK-GARCH

Entities Mentioned

• Robert Engle
• Victor K. Ng

Quotes

"This paper defines the news impact curve which measures how new information is incorporated into volatility estimates."

"Our results suggest that the model by Glosten, Jagannathan, and Runkle is the best parametric model. The EGARCH also can capture most of the asymmetry; however, there is evidence that the variability of the conditional variance implied by the EGARCH is too high."

My Take

The news impact curve is one of those small conceptual inventions that outlasts its paper: it turns an abstract comparison of volatility recursions into a single picture — how does this model react to this shock? — and immediately makes the leverage asymmetry visible as a curve that is not symmetric about zero. The sign- and size-bias tests are the practical companion: they let you diagnose that a symmetric GARCH is missing something without committing to which asymmetric model to use. For the wiki this is the diagnostic backbone of the GARCH page's asymmetric-model material (EGARCH, GJR, power-GARCH), and it pairs naturally with the volatility-forecast-evaluation and news-driven stochastic-volatility threads. The verdict favoring GJR over EGARCH has aged well, though the choice is data-dependent and later work finds the ranking sensitive to the return series and sample.