event-driven

To convert news articles into algorithmic trading signals, you use an LLM scoring step to assign a -1.0 to 1.0 impact score to the text, then apply exponential time decay to model diminishing market impact over a lookback window.

Time decay for event signals works by applying exponential decay to historical news scores across an event lookback window, modeling diminishing market impact to prevent look-ahead bias when combining them with technical indicators for backtesting.

The required CSV schema for storing historical event data includes five specific columns: date, event_type, score, source, and summary, ensuring reproducible backtests and standardized record-keeping across equities, FX, and crypto.

Yes, you can combine LLM-scored news signals with technical indicators for live trading by using weighted aggregation to merge the time-decayed event scores with your existing technical signals into a final tradable momentum signal.

Yes, this event signal aggregation approach works for crypto and FX markets, alongside equities. It standardizes unstructured macro news and announcements into time-decayed signals applicable across multiple asset classes for algorithmic trading.

You should use a constrained -1.0 to 1.0 scale for LLM news scoring because it standardizes unstructured article text into a quantifiable metric, allowing consistent weighted aggregation with technical indicators to produce a final tradable signal.