Tiny HFT bot, huge returns claimed

A social post this week claimed a tiny high-frequency trading bot turned $3,000 into $356,000 in two weeks by betting on price moves as small as one-hundredth of a cent. The post says the bot used Markov-chain state transitions above 87% and links to a one-hour lecture, but the claim is still just a social-media case study unless someone reproduces it with full trade data. (x.com) High-frequency trading is just trading on the market’s smallest clock ticks. Instead of asking whether Bitcoin or a stock goes up this month, it asks whether the next tiny move over the next few milliseconds is more likely to be up or down. (sec.gov) The raw material for that kind of trading is the limit order book. That book is the live queue of buyers and sellers sitting at each price, and it changes every time a new order appears, gets canceled, or gets filled. (arxiv.org) A microprice is a weighted guess of where the next trade will happen inside that queue. Recent research describes it as a short-horizon price estimator built from the spread and the imbalance between bids and offers, then refined with higher-order order-book information. (arxiv.org) Order-book imbalance is the simple idea underneath the pitch. If there are far more shares or contracts trying to buy than trying to sell at the best prices, the next move is often more likely to be upward, although academic work says that relationship is real but nuanced rather than magical. (arxiv.org) A Markov chain is one way to turn that stream into a model. It treats the market like a board game where the next square depends mostly on the current square, so a bot can map states like “heavy bid, thin ask, one-tick spread” to the next likely state. (ijml.org) That makes the 87% figure sound more impressive than it may be. In microstructure trading, a model can be right about the next state often and still lose money after fees, slippage, queue position, and adverse selection, which is when faster traders hit your quote just before the price moves against you. (quantstrategy.io) The hardest part is not prediction but execution. A bot chasing moves worth 0.01 to 0.1 cent needs data feeds, exchange access, and software fast enough that delays measured in microseconds can erase the whole edge before the order lands. (sec.gov) That is why most credible high-frequency trading results are shown with tick-by-tick backtests, exchange fee schedules, fill assumptions, and ideally broker or exchange statements. A screenshot of returns without those details is closer to a hypothesis than an audited performance record. (arxiv.org)