Intraday return spillovers and its variations across trading sessions

Among others, Baginski (1987) provides early evidence of return transmissions among stocks in the same industry through information transfer of management forecast of firm earnings. Meanwhile, more evidence of return spillover across international markets is documented within the literature, with such interdependence being potentially caused by the behavior of market participants ignoring fundamental economic information, and simply focusing instead on price movements in other countries. Related studies include, amongst others, Eun and Shim (1989), Becker et al. (1990), Hamao et al. (1990), King and Wadhwani (1990), Arshanapalli and Doukas (1993), King et al. (1994), Forbes and Rigobon (2002) and Yang and Lim (2004). Studies found either little or weak evidence of any relationship between macroeconomic innovations and stock variations (e.g., Schwert 1989; and King et al. 1994). Lin et al. (1994) argued that information revealed during the trading hours of one market may have impacts on returns in other global markets. Several studies also indicate that stock returns tend to exhibit stronger comovements during a period of financial market instability (e.g., see Forbes and Rigobon 2002; Yang and Lim 2004).

The observed asymmetric spillover effects are similar to those found in studies on international equity markets between small and large economies (e.g., see Hamao et al. 1990; Koch and Koch 1991; Lin et al. 1994; Lau and McInish 1993; and Karolyi 1995). Similar observation is also identified among individual stocks across markets, including return spillovers between OBM and its OEM/ODM companies (Yu and Hsu 2002), between stocks listed on the same exchange (e.g., see Rahman et al. 2002) and between stocks in the same market (Lee and Rui 2000; and Harris and Pisedtasalasai 2006).

See Bollerslev et al. (1992) for a comprehensive review of ARCH-type modeling. Rahman et al. (2002) presents a comprehensive analysis of the distributional and time-series properties of intraday returns of NASDAQ-listed stocks and find that the GARCH(1,1) model best describes the volatility of intraday returns.

The futures prices are collected based on the nearest-term contract and we rollover to a new contract on the fifth day of the delivery month of the futures contract.

We also carried out tests using the average of the bid and ask quoted prices; the results generally remain the same.

In order to ensure the accuracy of our sample data, we deleted all trades and quotes that were out of time sequence. We also omitted quotes that met the following three conditions: (1) either the bid or the ask price was equal to or less than zero; (2) either the bid or the ask depth was equal to or less than zero; and (3) either the price or volume was equal to or less than zero.

See Roll (1984) for a discussion on serial correlation in transaction data as a result of bid-ask bounce.

This pattern of reversion, possibly due to the bid-ask bounce in the data, has already been well documented for high-frequency equity returns. Although this lagged correlation is weak for large firms, it is somewhat exacerbated for smaller firms.

Taking the example of the Group_INTC stocks in Panel B, a 1.0 per cent unanticipated firm-specific return from INTC is expected to drive up the return of a G1a stock by an average of 0.142%, whilst that from a G2 stock would drive up the return of a G1a stock by an average of only 0.044%. The average results are representative for those of the individual stocks; thus, we report only averages for the purpose of saving space.

The Appendix presents the empirical results for the basic model, which does not include the common market factor. Thus, the associated transmission coefficients (β _i,j ) between an innovation stock and a responding stock reflect the ‘gross’ transmission effect that incorporates the common co-movement between these stocks. One may compare the results between the basic model in Table 6??? and the extended model in Table 3 by taking the example of Group_IBM stocks, the respective average gross transmission coefficient (β _i,j ) from IBM to G1a, G1b, G2 and G3 stocks is 0.236,0.102,0.054 and 0.068, while the average ‘net’ spillover coefficient \( \left( {\beta_{i,j}^{c} } \right) \) drops to 0.084,0.029,0.010 and 0.040 but still remain statistically significant. The difference is attributable to these stocks’ common responses to market-wide factors. Moreover, the transmission effects observed from large stocks to stocks of same or smaller sizes are driven both by positive reactions towards the common market factor and by positive spillover effects, while the return transmission effect from one stock to stocks of larger size is primarily attributable to responses towards the common market factor. For example, the average ‘gross’ transmission coefficient from G1a stocks to G2 (β _G1a,G2 ) is 0.045, while the average ‘net’ spillover effect \( \beta_{G1a,G2}^{c} \) is reduced but still significant at 0.026. In contrast, the average ‘gross’ transmission coefficient from G2 stocks to G1a (β _G2,G1a ) is 0.019 and significant, while the average net spillover effect \( \beta_{2,G1a}^{c} \) becomes only 0.005 and insignificant.

Another concern is that the spillover effects are attributable to non-synchronous trading. However, note that: (1) we omit those trades with zero trading volumes; (2) we observe strong spillover effects between G1 and G2, which are large-cap stocks and should be unlikely suffer from the non-synchronous trading problem; and (3) the spillover effects measured with larger sampling intervals are even stronger, as shown later in this study.

In their study of the real time market efficiency of ‘information’ from the Morning Call and Midday Call reports on CNBC-TV, Busse and Green (2002) found that the information was fully incorporated within one minute through a significant increase in the number of trades after these reports.

In the case of the ‘open-to-close’ period, we dropped the dummy variables in the variance equation so as to account for the opening and closing sessions.

An interesting comparison may be made to the study by Datar et al. (2008), which observed intraday spillovers between two international ETFs while the spillover effects no longer exist across trading days.

In the majority of the prior related studies, the relationship between trading volume and the arrival of information has been tested within an event study framework (for example, Bamber 1986; Bamber et al. 1999; Greene and Smart 1999). However, the results are inevitably event-dependent; thus, any generalization of such results is fraught with difficulties.