Data and methods.
The framework combines insider disclosures with market data and financial-news sentiment. It includes event studies, tabular neural models, LightGBM comparisons and multi-objective portfolio optimization.
When the information became available.
A signal cannot act on information before it arrives. The implemented signal construction activates on the next trading day, making the disclosure-to-decision boundary explicit.
Results after costs.
The project report describes a measurable post-disclosure event signal, but the portfolio optimization did not improve net returns after costs. The statistical finding and the economic outcome are different claims.
Gaps in the news data.
The available FinBERT news coverage reached only 9 of the 396 tickers in the reported universe. That is too sparse to treat the sentiment interaction as validated. Sector-label coverage is another limitation in the audit.
Tickers with available news coverage.
Too few to validate the sentiment result.
The event signal did not produce better net returns in this study.
Would broader, time-aligned news coverage change the interaction result?
Implementation and research materials. No live trading or investment recommendation.