mykola-quant/funding-rate-carry-falsification

This project is a research study that tests a popular crypto trading idea and shows that it does not actually work. The idea being tested is a strategy called delta-neutral funding rate carry, where a trader holds offsetting long and short positions in a cryptocurrency's spot and futures markets, hoping to collect a profit from the funding rate payments that futures traders pay each other, while staying protected from price swings. Before collecting any data, the author locked in the exact rules of the test in a document called the pre-registration, including the trading signal, the assets to test (Bitcoin, Ethereum, and Solana), the cost assumptions, and the date that separates the data used to build the strategy from the data used to check it honestly. This is done specifically to prevent tweaking the rules after seeing the results, which is a common way research like this becomes misleading. The result is that the strategy loses money in every single combination of asset and variant tested, once realistic trading costs are included. For Bitcoin and Ethereum, the problem is that trading costs are simply too high relative to the funding income available, no version of the timing rule gets close to breaking even. For Solana, the story is different and more useful: the funding income earned was actually higher than the costs, but the strategy still lost money because the long and short legs of the trade did not track each other perfectly, a problem the author calls basis risk. The project also finds that the funding income opportunity shrank over time, consistent with more sophisticated trading firms competing that opportunity away. The repository includes the locked pre-registration document, Python scripts for collecting the funding rate and price data from Binance, running the core analysis, and breaking down exactly why the strategy failed component by component, plus the resulting data files. It is one entry in a public series of similar studies by the same author testing other retail trading signals. The code is released under the MIT license.