In recent years, sophisticated institutional traders in financial markets have increasingly used new sources of information such as ‘sentiment’ indicators derived from news wire articles. Such news analytics are created by computer algorithms and can tell traders within milliseconds whether an article is positive or negative and contains relevant information about a firm’s value. In parallel with this, the growth of computerised trading has accelerated the process of accessing such information and increased the speed with which it is incorporated into stock prices. Access to such low-latency ‘meta information’ – ‘news about the news’ – can provide a competitive advantage to its users, which are mainly high frequency and algorithmic traders like hedge funds. However, inaccurate low-latency signals can lead to unintended consequences when algorithms automatically initiate trades based on inaccurate information. For example, in April 2013 an incorrect twitter feed about a White House explosion caused a mini flash crash in US markets. Some quickly blamed algorithmic trading for the reaction, while others argued that human traders were mainly responsible.1 In any case, news reading algorithms may be more likely to misinterpret news than human traders. Thus, understanding the magnitude of the price effects associated with such meta information is critical for policymakers concerned about financial market stability.
In July 2013, New York State Attorney General Eric Schneiderman rebuked Thomson Reuters for selling access to key economic survey data two seconds earlier to some high-frequency algorithmic traders than to other customers. Unlike the early release of such economic survey data, however, news analytics are based on publicly available news. Therefore, they constitute a ‘fairly earned’ advantage. However, since news analytics enable faster trade based on public information, they give a trader a similar advantage as would early access to private information. In either case, an important question is whether quick-triggered trading initiated by such low-latency information has an impact on the market that is distinct from the underlying informational content of the news. That is, are there potentially distortionary price effects induced by high frequency trading based on news analytics? It seems that only the existence of such distortions should justify regulatory intervention.
Media, institutional traders, and stock prices
A large academic literature has studied the effects of the information from traditional media sources on stock prices (e.g., see Tetlock 2007, 2011, Fang and Peress 2009, Dougal et al 2011, Engelberg and Parsons 2011, and Peress 2013). Others (e.g. Riordan et al 2013, Gross-Klugmann and Hautsch 2011, Sinha 2012, Zhang 2013) have investigated the market reaction to news analytics but are unable to demonstrate a separately identifiable influence of the meta information in news analytics that is distinct from the information in the underlying news on which it is based. In a new working paper, we are able to identify the causal effect of meta information (von Beschwitz et al 2015). We exploit differences in the high frequency signals derived from Dow Jones Newswire stories between older and newer versions of Ravenpack, one of the major providers of news analytics. We use the corrected and back-filled signals from the new version of RavenPack to proxy for the ‘true’ informational content of the news, while traders actually reacted to the original signal that was released in real time. The differences in stock market response between the ‘true’ and real-time versions of the news analytics enable us to study the causal impact of meta information. While such differences were rare in a relative sense (roughly 3% of our sample), their absolute number of 24,963 is high enough to allow for tests that have sufficient power.
The signals derived by RavenPack from the DJ Newswire measure (a) the importance of the article to the firm that is the subject of the story (‘relevance’), and (b) whether the story conveyed positive or negative information about the firm (‘sentiment’). Relying on differences in relevance scores, we define three article classifications: High-relevance articles originally Released as being High-relevance (we use the acronym HRH); Low-relevance articles originally Released as being High-relevance (LRH); and High-relevance articles originally Released as being Low-relevance (HRL). HRH and HRL articles have similar informational content, but only the HRH articles were released as relevant to the stock market and thus should affect stock prices more. This allows us to determine the causal effect of meta information on the stock market by comparing differences in stock price response between HRH and HRL articles. Overreaction in stock price in response to an LRH article, when no effect should occur, also indicates the causal effect of meta information.
We find that stock market participants react differently to Accurate and False Negative articles and identify three differential effects:
- Price Effect. The stock price reaction concentrated in the first 5 seconds after the release of news analytics, compared to the total reaction over 120 seconds, is significantly greater for HRH than for HRL articles. This difference in speed of the stock price response is 1.3 percentage points or 10% relative to the mean. We also find that the market overreacts to LRH articles in the short term and starts mean-reverting after 30 seconds. This finding is consistent with a causal effect of RavenPack that leads algorithmic traders to trigger an initial overreaction to the article that is then corrected by human traders (see Figure 1).
Figure 1. Difference in stock price response between HRH and LRH articles
Notes: This figure displays the cumulative return from t-30 to t+120 seconds around the articles. Returns are multiplied with the sentiment direction of the article. HRH refers to articles that have high relevance according to both versions of RavenPack. LRH refers to articles that are marked as having low relevance in the older RavenPack version (which was released to the market), but marked as having high relevance in the newer RavenPack version.
- Volume Effect. The share of trade volume for the stock concentrated in the first 5 seconds, compared to the two minute interval after release, is significantly greater for HRH than for HRL articles. This larger trade volume is consistent with existing theory that predicts investors with a speed advantage trade very aggressively on signals that they can exploit before other traders (e.g., Foucault et al 2013).
- Liquidity Effect. A stock’s liquidity measured during the five seconds after an article’s release is significantly lower for HRH than for HRL articles. Why? While trading on news analytics improves the informational efficiency of stock prices, the fact that only a subset of market participants has access to news analytics increases information asymmetry in the market. Under this scenario, our findings are consistent with market makers and other liquidity providers reducing liquidity provision after a relevant news release to avoid being picked off by informed order flow (e.g. Kim and Verrecchia 1994, Menkveld 2013, Hagströmer and Nordén 2013). An important question, which is outside the scope of our study, is whether the ensuing increase in information efficiency is high enough to more than offset the accompanying reduction in liquidity.
In this paper, we study how the information delivered by news analytics companies affect the stock market and, in particular, how this alternative delivery mechanism affects liquidity and market efficiency. This question is important, because news analytics are a major source of information for institutional investors, which are responsible for the majority of the trading volume in the market.
Our findings show that providers of media analytics have a significant impact on the market that is separate from the information contained in the news. These findings have normative implications in terms of the recent regulatory debate on high-speed delivery of information and the effects of algorithmic trading. They show that news analytics improve price efficiency, but at the cost of reducing liquidity and potentially distortionary price effects.
Carney, J (2013) “The Trading Robots Really Are Reading Twitter”, Yahoo Finance, April 24.
Dougal, C, J Engelberg, D Garcia and C Parsons (2012) “Journalists and the stock market”, Review of Financial Studies, 25: 639-679.
Engelberg, J and C A Parsons (2011) “The causal impact of media in financial markets”, Journal of Finance, 66(1): 67-97.
Fang, L H and J Peress (2009) “Media coverage and the cross-section of stock returns”, Journal of Finance, 64: 2023-2052.
Foucault, T, J Hombert and I Rosu (2013) “News trading and speed”, Working Paper.
Groß-Klußmann, A and N Hautsch (2011) “When machines read the news: Using automated text analytics to quantify high frequency news-implied market reactions”, Journal of Empirical Finance, 18, 321-340.
Hagströmer, B, and L Nordén (2013) “The diversity of high frequency traders”, Working Paper.
Kim, O and R Verrecchia (1994) “Market liquidity and volume around earnings announcements”, Journal of Accounting and Economics, 17(1-2), 41-67.
Menkveld, A (2013) “High frequency trading and the new market makers”, Journal of Financial Markets, 16, 712-740.
Peress, J (2011) “The Impact of the media in financial markets: Evidence from newspaper strikes”, Working Paper.
Riordan, R, A Storkenmaier, M Wagener and S S Zhang (2013) ”Public information arrival: Price discovery and liquidity in electronic limit order markets”, Journal of Banking and Finance, 37, 1148-1159.
Sinha, N R (2012) “Underreaction to news in the US stock market”, Working Paper.
Tetlock, P (2007a) “Giving content to investor sentiment: The role of media in the stock market”, Journal of Finance, 62, 1139–1168.
Tetlock, P (2007b) “All the news that's fit to reprint: Do investors react to stale information?”, Review of Financial Studies, 24, 1481-1512.
von Beschwitz, B, D B Keim and M Massa (2015) “First to 'Read' the News: News Analytics and Institutional Trading”, CEPR Discussion Paper 10534.
Zhang, S S (2013) “Need for speed: An empirical analysis of hard and soft information in a high frequency world”, Working Paper.
1 See, for example, Carney (2013).