Abstract

When a company releases earnings results or makes announcements, a sectoral wide lead-lag effect from the stock on the entire system may occur. To improve the estimation of a system experiencing system-wide lead-lag effects from a single asset in the presence of short time series, we introduce a model for Large-scale Influencer Structures in Vector AutoRegressions (LISAR). We study the asymptotic properties of the estimator and validate its performance in extensive synthetic data experiments. We study the performance of the LISAR model on high-frequency data for the constituents of the S&P100, separated by sectors. We find the LISAR model to significantly outperform on up to 14.7% of the days in terms of forecasting accuracy. Trading strategies with signals derived from the LISAR model achieved up to 60% excess return compared to other strategies. We then investigate the sources for the effects with a particular focus upon news occuring around the emergence of influential assets. We study the news in regards to their thematic and sentiment content to investigate which type of news are relevant for influencer behaviour. We validate empirically if news of certain types of strong sentiment act as a predictor for influencer emergence and if this spills over into better trading signals.