In exploring the role of empirical investor networks in information diffusion, the authors find strong support for the existence of stable and relatively diffuse structures on the Istanbul Stock Exchange. Within these structures, more central members trade earlier and more successfully. The finding is consistent with an information-sharing network.
What’s Inside?
Empirical investor networks (EINs) lie somewhere between public media and private information sharing, with links forged informally via conversations, social networking, or blogs. The authors estimate EIN structures from observed trading patterns and characterize them in terms of their connectivity and degree of centralization. They compare trading returns and trade timing for central figures in the network with those for more peripheral figures on the assumption that price-sensitive information reaches more central points earlier. Most analysis is conducted in-sample, but the main conclusions are unchanged when an out-of-sample period is included.
How Is This Research Useful to Practitioners?
The authors establish EIN stability by splitting their one-year sample into two parts and finding 6.1–7.6 times the connection overlap than is implied by chance alone. They regress investor returns against centrality in the network and numerous control variables. A one standard deviation increase in centrality is associated with a 0.7%–1.8% monthly return enhancement, a statistically significant result. Degrees of connectivity and centrality are highly correlated, but it is centrality that appears to drive higher returns. Focusing on 11 stock-specific events, the authors show that centrality in the network is related to earlier trading. A one standard deviation increase in centrality corresponds to trading 2.55 days earlier, or 2.8% extra return per event. That centrally placed investors benefit from timely information is confirmed by the observation that delaying the most central half of investors’ trades by one day erodes returns by 0.21% but aids other investors by 0.26%. The economic significance of centrality drops 0.15% when all trades are delayed.
Out-of-sample testing using the first eight months as the network-defining in-sample period reveals that central investors earn a reduced, but still significant, 0.2%–2.1% premium over their peripheral peers, and the results of the event studies remain valid as well. Further evidence that central investors benefit from an information advantage is found when the authors observe that in the four months with the most earnings announcements, the coefficient on centrality in the return regression is twice as large as in other months.
How Did the Authors Conduct This Research?
The dataset includes all trades on the Istanbul Stock Exchange during 2005, encompassing 580,142 active accounts, of which just 489 were institutional, and 200,000 trades per day. The authors establish their EIN model validity by deciphering the layout of an artificially generated network of known structure. Such a process is found to be robust to excluding many of the agents in the network.
The authors consider two investors to be linked when they trade the same stock in the same direction within 30 minutes of each other on at least three occasions, and they measure returns using a one-month horizon. Absolute and market-relative returns account for the diffusion of both stock-specific and overall market-sensitive information. With a window of 30 minutes, the median number of links per investor is 159, consistent with typical social network sizes. In the 2005 period, the authors find a network composed of 1,109 communities—that is, sets of investors more linked to each other than to others—each with an average of 523 members. The situation is suggestive of a decentralized network, not very reliant on mainstream media.
The results are robust to varying the link threshold between 1 and 10, excluding institutional investors, using actual realized returns on the subsample of trades closed within a month, extending the return window to three months, and excluding network neighbors who share the same brokerage house, which may provide common information to both investors.
Abstractor’s Viewpoint
The authors find very strong statistical support for their conclusions concerning the existence of EINs, the importance of EINs in the information diffusion process, and the advantage of being more centrally placed in such networks. It would be interesting to update the analysis to the current year because the opportunities for online networking have continued to grow since the 2005 study period. The authors make reference to the difficulties of working with their large dataset. Nevertheless, a duplicate study on one of the major global stock exchanges could establish the generality of their conclusions.