Bot Track Record
Two simple, rule-based trading bots, running live on an Alpaca paper account (simulated money, real market prices). Live results are shown next to what the backtest predicted, so you can see whether the backtest holds up.
How the bots work
🪙 Crypto — time-series momentum (BTC, ETH)
Each coin gets half the budget. The bot holds a coin while its 30-, 60- and 90-day returns are positive (each counts as one vote) and its price is above the 200-day average. Position size shrinks when volatility rises (target 40% annualised volatility). Checks every 15 minutes, trades only when the target changes by 10% or more. Long only, no leverage.
📈 ETF — dual momentum (10 ETFs)
Universe: SPY, QQQ, IWM, EFA, EEM, VNQ, TLT, IEF, GLD, DBC — US, international and emerging-market stocks, real estate, bonds, gold and commodities. At each month end the bot ranks them by 9-month return and holds the top 4 at 25% each, but only ETFs with a positive return that trade above their 200-day average; empty slots stay in cash. Trades only during US market hours. Long only, no leverage.
How they were tested
Rules were chosen on older data and judged only on newer data they had not seen ("out-of-sample"), plus a yearly walk-forward test, with fees and slippage included. Crypto: tuned on 2021–2024, tested on 2025–2026. ETFs: tuned on 2016–2022, tested on 2023–2026. Neighbouring parameter values were checked to make sure the result isn't a lucky setting. Earlier ideas that failed this test (several trend and breakout rules, an LLM-driven trader, funding-rate arbitrage) were discarded.
Risk limits
Each bot has a fixed $1,000 budget, never buys with more than available cash, stops buying for the day after losing 3% of its budget, and has an independent kill switch.
Interested?
If you run a trading bot and would want its live results verified side by side with its backtest like this, open an issue and say so. This is an experiment; nothing is for sale.