To observe the intraday market reaction to news, we measure market activity using the price volatility, the number of transactions, and the transaction volume every minute for each stock. Price volatility is defined by the absolute value of the stock price log-return over one minute:
$$ V'(d,t)=\big|\log{P(d,t+1)}-\log{P(d,t)}\big|, $$
(1)
where d and t represent the date and the time of day (by the minute) (e.g., \(d=\text{5/18/2015}\), \(t=\text{9:30 a.m.}\)), respectively. Note that the three market activity indicators considered here are highly correlated with each other [46].
Market activity has seasonal and daytime variations. We remove these variations from typical market cycles to correctly estimate the market impact on market activity for a day by introducing the normalized volatility, the normalized number of transactions, and the normalized volume as follows:
$$\begin{aligned}& V(d,t)=\frac{\frac{V'(d,t)}{\langle V'(d,t) \rangle_{d}}}{\langle\frac {V'(d,t)}{\langle V'(d,t) \rangle_{d}} \rangle_{t}}, \end{aligned}$$
(2)
$$\begin{aligned}& N(d,t)=\frac{\frac{N'(d,t)}{\langle N'(d,t) \rangle_{d}}}{\langle\frac {N'(d,t)}{\langle N'(d,t) \rangle_{d}} \rangle_{t}}, \end{aligned}$$
(3)
$$\begin{aligned}& \mathit{Vol}(d,t)=\frac{\frac{\mathit{Vol}'(d,t)}{\langle \mathit{Vol}'(d,t) \rangle_{d}}}{\langle \frac{\mathit{Vol}'(d,t)}{\langle \mathit{Vol}'(d,t) \rangle_{d}} \rangle_{t}}, \end{aligned}$$
(4)
where \(N'(d,t)\) and \(\mathit{Vol}'(d,t)\) are the number of transactions and their volume at time t on date d. Since \(\langle\cdots\rangle _{d}\) expresses the mean on date d, daily seasonality is removed from the level of market activity by the first term in each of the equations. \(\langle \cdots\rangle_{t}\) also expresses the mean at time t in all sample periods. The second term removes the intraday cycles in market activity.
Next, we investigate the intraday market reaction to news displayed on the RTRS electronic trading platform. For illustration, we focus on GM stocks traded on the NYSE. We observe the three indicators of market activity in GM stocks on the NYSE at time Δt (i.e., \(V(\Delta t)\), \(N(\Delta t)\), \(\mathit{Vol}(\Delta t)\)), knowing that there was an ALERT or a HEADLINE with ‘GM.N’ at time \(\Delta t=0\). Figure 1 shows the mean of each of the market activity measures, i.e., \(\langle V(\Delta t) \rangle\), \(\langle N(\Delta t) \rangle\), and \(\langle \mathit{Vol}(\Delta t) \rangle \). In the ALERT case, the mean jumped by about 60% at time \(\Delta t=0\) and slowly decayed in an exponential function (\(=0.45 \exp(-0.073 \Delta t)+1\)). On the other hand, when a HEADLINE was displayed the mean hardly moved.
Figure 2 shows the intraday market reaction to the news of 64 NYSE stocks and 14 NASDAQ stocks. The correspondence between the stock numbers in Figure 2 and ticker symbols is provided in Table 1. Note that the stocks we analyze here are reststricted to those satisfying the following requirements: (1) the number of ALERTs is above 500, (2) the number of HEADLINEs is above 500, and (3) the sum of the two is above 3000. Specifically, each of the panels in Figure 2 presents the conditional mean of one of our measures of market activity in the three minutes after a particular news item was displayed, i.e., \(\langle V(\Delta t)|0 \le\Delta t < 3 \rangle\), \(\langle N(\Delta t)|0 \le\Delta t < 3 \rangle\), \(\langle \mathit{Vol}(\Delta t)|0 \le\Delta t < 3 \rangle\). In the case of ALERTs, we observe a jump in market activity in almost all stocks. The mean of these jumps is 36.5%. On the other hand, none of the stocks responded much to HEADLINE news.
Table 1
Ticker that corresponds to stock number in Figure
2
To check the statistical significance of these results, we randomize the timing of the arrival of a news item and repeat the same exercises as in Figures 1 and 2. We then use the randomized data to calculate the mean and the standard deviation of the market activity measures. The error bars shown in the two figures are for ALERTs. As seen in Figures 1 and 2, the reaction of the market activity indicators observed in the actual data is much greater than that calculated using the randomized data, implying that market reactions to news are statistically significant.
The results presented in Figures 1 and 2 suggest that we need to distinguish news that attract a lot of attention from market participants and news that receive little attention, and focus on news attracting a lot of attention in assessing the market response to such news. In the following sections, we examine the statistical laws regarding linguistic similarity among news articles, and propose measures for the novelty of a news article and for the topicality of an article.