Swarm Analyst Predictor is a research project that models how sell-side analysts interpret earnings releases and how their collective reactions influence short-term price movements. Instead of focusing only on financial results, it captures how analysts revise price targets, adjust sentiment, and respond to guidance. The system builds a set of analyst agents trained on earnings call transcripts, filings, and historical revision data. These agents simulate a consensus reaction to new events, which is then compared against real analyst behaviour and market outcomes. The results are used to test whether shifts in analyst sentiment can predict post-earnings returns, supporting event-driven trading strategies and improving understanding of how market narratives affect price movement.
This project studies whether short-term intraday “flag” patterns on 5-minute SPY data show a measurable momentum continuation effect. We define these patterns mechanically and analyze what happens next under different market conditions to understand when and why they work.
Traditional pairs trading assumes stable mean reversion, but this often breaks during regime shifts (e.g. macro shocks, earnings, volatility bursts). We aim to identify when mean reversion signals are reliable, using a compact model to recognize favourable market conditions. We will use ARIMA to forecast the spread’s short-term reversion direction, then use GARCH to forecast volatility for risk-aware sizing. We will feed these signals + technical & event features into a tiny transformer that outputs a regime score, and trade only when ARIMA reversion and Transformer regime agree, with smaller size when volatility is high.