Essays on the role of information in shaping short-term price dynamics: evidence from a high-frequency setting

Abstract

This thesis asks how information enters prices at intraday horizons, and it approaches the question through three essays that move progressively from the reaction of prices to news toward the mechanism by which that reaction is produced. The first essay studies the intraday processing of firm-specific ESG news for Dow Jones Industrial Average constituents. News items are classified by pillar and sentiment using a large language model, and abnormal returns and trading volumes are measured by event study at the minute frequency. The results are heterogeneous across pillars, and the muted return responses admit two readings that a minute-level design cannot separate: rapid absorption, or limited statistical power. The second essay moves from returns to the limit order book itself. Using message-level data for six NASDAQ-listed stocks of differing relative tick size, I estimate a score-driven model in which instantaneous price impact is a time-varying parameter. Impact proves news-dependent — earnings releases generate the largest peaks, FOMC announcements sharper but more moderate ones — and it changes sign, which a static specification cannot produce at all. The third essay asks whether the association between order flow and returns is causal. For five stocks across sixteen FOMC announcements, I decompose risk-adjusted returns into an adverse selection component and an order flow imbalance component, and I instrument the latter. Adverse selection is negative and significant throughout; order flow imbalance, significant under OLS, loses significance once simultaneity is corrected. Cross-sectionally, the adverse selection effect is strongest in large-tick stocks. Taken together, the three essays follow a single question into finer and finer detail: how a piece of news becomes a price change. The first measures the size of the reaction, the second tracks how the impact of each trade shifts while the news is being absorbed, and the third asks how much of that impact carries information and how much is mechanical. Where the answer rests on an assumption I cannot test directly, I say so.