Predict the next-minute stock return (ratio of time T+1 price to time T price) using market microstructure features such as volume, trade imbalance, spreads, and order book statistics.
Aggregated per minute per stock:
num_trades,o,h,l,c,total_volume,total_buy_cap,total_sell_cap,weighted_pricetrade_imbalance_ratio,volume_imbalance_ratioavg_spread,max_spread,min_spreadweighted_avg_bid_price,weighted_avg_ask_price
Target variable:
- Return = (Next minute's close price) / (Current minute's close price)
- Aggregated trade and quote data to 1-minute intervals
- Engineered predictive features
- Modeled using Linear Regression
- Evaluated with:
- Minute-wise correlation plot (actual vs predicted returns)
- Correlation matrix heatmap across all companies
Measures how well predictions match real returns over each minute.
Stock-wise return correlations across companies on the selected date.