How do you backtest crypto trading strategies?

Define the exact trading rules, gather point-in-time market data, model fees and slippage, run the strategy across realistic historical windows, and judge the result with drawdowns, turnover, stability, and out-of-sample behavior rather than headline return alone.

Write down the universe, signal formula, rebalance timing, position sizing, exposure caps, and exit rules before looking at the result. Rules that change after seeing the chart are part of the research process, not clean validation.

Crypto backtests often fail because they use clean candle data but ignore funding, exchange outages, changing liquidity, delistings, spread, borrow constraints, and timestamp alignment. The data model should match what the strategy would have known at decision time.

Review maximum drawdown, volatility, turnover, Sharpe or Sortino, hit rate, capacity, tail periods, parameter sensitivity, and whether the edge survives a separate evaluation window. A tradeable strategy needs a durable process, not one lucky equity curve.

Every backtest run should leave enough evidence for another researcher to understand what changed. That means logging the data snapshot, code version, parameters, cost model, evaluation window, benchmark, and reason the run was created.

A backtest loop should stop when new changes mostly explain the past instead of improving the process. If each iteration adds another exception, filter, or parameter to fix one historical drawdown, the research is probably overfitting.

Alphora's public docs and catalogue are organized around explicit signals, strategy construction, validation notes, and implementation caveats so traders can keep the research workflow inspectable.