Reaction to nonscheduled news during financial crisis: Australian evidence

Abstract

News analytics software applies linguistic algorithms to newswire releases in order to assign a sentiment score; this allows users to comprehend the unstructured data flowing through newswires. I examine the market reaction of leading Australian stocks to stock-specific news flow during the financial crisis of 2007–2009. A high-frequency VAR model with GARCH effects modelled through a VECH(1,1) specification is utilized. I find a significant market impact induced by contemporaneous news items, a significant and positive relationship between volume and volatility, an increase in bid–ask spreads following periods of increased volatility, and evidence of volatility persistence.

Keywords:

• VAR-GARCH
• nonscheduled news
• stock market
• news analytics
• ASX

JEL Classification:

• G10
• G14
• G15
• G00

Notes

¹ This processed data is available to market participants (at a cost) almost instantaneously; academic researchers are able to access this information only at a later stage – usually several months afterwards.

² More extensive information on the nature of the RavenPack news analytics tool may be found at www.ravenpack.com

³ The day on which AIG issued a warning that credit defaults were spreading beyond the subprime sector, and coordinated intervention by major central banks is 9 August 2007. The period 1 September 2009 corresponds with the return of credit market indicators to pre-crisis levels. The defined crisis period also approximately correspond to dates of structural breaks identified in a wider 2000-2011 sample.

⁴ This score represents the news sentiment derived from the tone of a story using a combination of a traditional tagging methodology that uses an algorithm to map keywords and phrases to pre-defined sentiment values, and an expert consensus methodology that entails training classification algorithms on the results of financial experts manually tagging stories.

⁵ A three-factor CAPM is assumed when determining abnormal returns.

⁶ Akaike information criteria (AIC) and Schwarz information criteria (SIC) are utilized to obtain optimal lag length, and results for the first two lags are reported for the VAR equation.

⁷ A VECH(1,1) specification is adopted on the basis of AIC and SIC, as well as in consideration of computational issues.

⁸ The results are robust to the ordering of variables.