basantsingh-gif/marketmind ? reverse-engineered prompt

Reverse engineered prompt

Build me a Python project that tests whether daily news sentiment can predict the next day’s DJIA return, using FinBERT on the top 10 Reddit WorldNews headlines per day.

I want it to load the Kaggle news and DJIA CSV files, score each headline with a positive minus negative sentiment value, average that into a daily signal, and line it up so today’s sentiment is compared only with the next trading day’s return. Please include a leak free backtest with a chronological train and hold out split, a simple long and short strategy based on the train median sentiment, and evaluation with Spearman rank correlation, a permutation test, Sharpe ratio, cumulative returns, and a few robustness checks like different horizons, yearly breakdowns, and transaction cost sensitivity. It would be great if it also saves the main results to files and produces a chart of cumulative performance. If needed, look up current docs online for the FinBERT model and any Python package details.

Are you gonna build this?

make sure you review the code using coderabbit

Try freeSponsored — opens CodeRabbit in a new tab