Peter Markoski Selected work

Selected work / 04

Prediction Strategy

What changes when the market gets new information?

Role
Research & engineering
Status
Ongoing research
A printed abstract composition for Prediction Strategy, independent of market data.
Conceptual illustration · independent of market data

An automated loop for testing trading ideas

I built a research pipeline that takes a trading hypothesis through backtesting, comparison and reporting in one automated run. Python runs each candidate and its baseline against the same market history. Both call the same Rust engine, so I can trace a difference in results back to the strategy change that produced it.

For market sentiment research, I integrated Jev, a fast AI model that screens headlines and summaries as they arrive over RSS. It flags developments such as earnings surprises and guidance changes, giving me a timestamped record of what the strategy could have known as trading unfolded.

How far back should a strategy look after news breaks?

A surprise earnings announcement raises a practical question. Do the last few minutes of trading tell us more than the hour before the news? The RSS collector records when it first sees the headline. Jev classifies the event and its relevance to equities. I’m exploring whether those signals can help select a shorter or longer lookback window for live trading. A fixed window gives me a baseline for testing whether adapting to the news adds value.

The project source is private.

Ask about a project walkthrough on LinkedIn