I switched off the bot: five strategies, 77,000 candles and zero net edge over fees | Xiliux Blog

In a trading experiment, a bot was put on hold due to a measurement rather than a loss streak. Five strategies were tested using 77,000 one-minute BTC and ETH candles, with fees subtracted from the results. The best strategy returned merely 0.0135% per trade, with a statistically insignificant t-statistic of 0.16. This means the result could have easily occurred by chance.

The five main strategies analyzed were momentum, reversion, order-book imbalance, grid, and funding carry. Each strategy's performance was measured by subtracting the real fees paid, not an estimated fee or zero. The fee impact was crucial, as a strategy that seemed profitable without considering fees could have been a money drain.

The t-statistic is a measure of whether the result is significant or just random noise. In this case, a t-statistic of 0.16 showed that the result was not distinguishable from random chance. The experiment highlighted the importance of factoring in fixed costs, such as server expenses, as they can erode any potential edge. The conclusion was that switching off the bot and moving the capital to a savings account was the right decision, as a system requiring maintenance would consume valuable attention better spent elsewhere.

The takeaway was to always measure strategies with fees from the very first backtest, demand a statistical significance, and not overlook the fixed costs associated with maintaining a system.