s-leo-de/following-the-jets

Following the Jets is a research project that investigates a well known idea in alternative data investing: that tracking a company's private jet flights can reveal an upcoming merger before it is announced. The author rebuilt this trade from scratch using over 1.2 million flights and more than 200 acquiring companies, then tried as hard as possible to break the result rather than confirm it. The final conclusion is that there is no reliable signal that predicts mergers before they are announced, and no way to trade on it profitably. What does hold up is a smaller and less exciting fact: acquiring companies tend to fly slightly more often to the cities where their eventual targets are based, but this happens consistently over time rather than picking up right before a deal. A backward looking version of the analysis produces an impressive looking result, but that number only appears because the deals are already known and the flights are matched to them afterward. Once the question is flipped to ask how often a jet visits without a deal following, the apparent edge mostly disappears. Several rounds of more careful controls, including matching by calendar timing and by how active a company generally is, shrink the effect toward nothing. A test that only allows one vote per company, rather than counting individual flights, finds no effect in any version tried. A comparison group built to have no real signal also comes back null, pointing to ordinary business connections rather than early knowledge of a coming deal. The timing of flights actually increases after a merger is announced, not before, which argues against the predictive idea entirely. The author frames the real value of the project as the method itself, a demonstration of how to test a promising looking data signal using matched controls, careful statistics, and adversarial review, until only what is actually true remains. The project draws on large scale flight tracking data, aircraft registration records, and financial market data, combined with statistical techniques from event studies and quantitative finance. The underlying licensed financial data is not included in the repository, but the code and analysis scripts are provided so the work can be reproduced by someone with the right data access. This project is aimed at researchers, quantitative analysts, and anyone curious about how to rigorously test a data driven trading idea.