Crypto backtesting has unique challenges that general guides miss. Here is how to handle them and avoid the most common errors specific to crypto.
Crypto backtesting shares the same core methodology as backtesting any trading strategy, but introduces several challenges that general guides do not address. A 24/7 market with no natural session close, pronounced multi-year bull and bear cycles, exchange-specific data fragmentation, funding rates on perpetual futures, and very short market history all affect how crypto backtests should be designed, run, and interpreted.
Three properties of crypto markets make backtesting harder than backtesting traditional instruments.
Very short market history. Bitcoin has approximately 15 years of tradeable data. Most altcoins have 3 to 7 years. A strategy requiring 3 years of historical data to validate is consuming 20 to 50% of the available history. There is no good solution, only the constraint to acknowledge. More frequent live validation via shadow data tracking compensates for the limited historical baseline.
Pronounced macro cycles. Crypto has clearly distinct bull and bear phases: 2017 bull, 2018 bear, 2020-2021 bull, 2022 bear, 2023-2024 recovery. A backtest covering only 2020-2021 covers one of the strongest bull markets in asset history. Any valid crypto backtest must span at least one complete cycle, including a meaningful bear phase, to produce results applicable to the full range of market conditions.
No natural session boundaries. Crypto trades 24/7. Daily bars close at midnight UTC by convention, not by market structure. For daily strategies, results can vary depending on which timezone's midnight is used. For 4-hour and shorter timeframes, this matters less. But the absence of overnight gaps, opening auctions, and institutional flows that structure traditional market sessions means some traditional technical patterns apply differently in crypto.
Exchange-specific data. A backtest on Binance data reflects Binance prices and Binance liquidity. A live trading account on a different exchange faces different fills. Price differences across exchanges create basis risk that single-exchange backtests cannot capture. For liquid assets (BTC, ETH) the difference is small. For less liquid altcoins, it can be significant.
Missing data and gap events. Most exchanges experienced outages during high-volatility events: the LUNA collapse, the FTX collapse, multiple flash crashes. Historical data from these periods may show prices that were not tradeable at the time. Strategies that appear to handle these events in backtesting may have been executing on prices unavailable in live conditions. Review the backtest period's major events and verify the data quality around those dates.
Survivorship bias. Backtesting on current top-20 assets selects for assets that survived and appreciated. Many assets in the top 20 by market cap in 2018 or 2021 have since lost 90%+ of value or ceased to trade. A portfolio manager in 2019 would have included those assets. A backtest on today's top 20 going back to 2019 is a pre-filtered set of survivors. The survivorship bias inflates returns for any diversified crypto strategy tested on current rankings.
Volume data reliability. Crypto exchange-reported volume includes wash trading, fee rebate incentives, and inflated reporting across many smaller exchanges. Volume-based indicators that work reliably on equity exchanges are less reliable in crypto. ATR-based measures of market activity are more robust than volume-based measures in crypto backtesting.
Crypto strategy performance depends on two regime levels: the tactical regime (trending vs ranging within a cycle, measured by ADX) and the macro regime (where the market sits in its multi-year cycle). A backtest that only segments by tactical regime misses the macro context that changes strategy performance fundamentally.
A mean-reversion strategy that works well during a ranging accumulation phase produces losses during the directional leg of a bull market. A trend-following strategy producing strong returns during 2020-2021 may produce nothing in the 2022 bear market if it only handles long-side signals.
The four macro phases to test across: the bull market trending phase (high ADX, directional momentum, trend-following performs well), the bull market topping phase (high volatility, regime uncertainty, SAR-based strategies may whipsaw), the bear market (requires short-side capability or flat market exposure), and the ranging/recovery phase (low ADX, mean-reversion opportunities, trend-following largely inactive).
A strategy that performs similarly across all four phases is genuinely robust. A strategy that outperforms in one phase is phase-specific, useful in that environment and unreliable outside it. Most crypto strategies are implicitly phase-specific, optimized for the macro environment during which most of their backtesting data was collected.
For the tactical regime classification framework, see What Is a Market Regime? and Ranging vs Trending Crypto.
Funding rates on perpetual futures. Perpetual contracts settle funding every 8 hours. When the market is bullish and most traders are long, funding rates are positive: longs pay shorts. During 2020-2021, funding rates on BTC perpetuals often ran at 0.01% to 0.03% every 8 hours, which compounds to 2% to 9% per month in holding costs for long positions. A backtest that ignores funding rates materially overstates the profitability of long futures strategies during exactly the periods when those strategies produce their best-looking results.
Liquidation risk at leverage. Leveraged futures positions can be liquidated if the position moves against the account. This risk exists in live trading but most backtesting frameworks do not model it. A backtest showing strong overall performance may include drawdown periods where a leveraged version would have been liquidated. Position sizing should ensure drawdowns never approach liquidation thresholds, and this constraint must be validated in the backtest.
Spread and slippage during volatility events. Spreads on crypto assets widen significantly during high-volatility events. Backtests using mid-price fills overstate profitability. Using conservative spread estimates, particularly around the major event dates in the backtest period, produces more realistic results.
For the general backtesting methodology including cost handling, see How to Backtest a Trading Strategy.
The shadow data system records every signal's conditions, entry price, and exit outcome in real time, a continuous out-of-sample backtest that runs alongside the live system. Historical shadow data is analyzed by macro regime phase as well as by tactical ADX regime.
One finding that emerged specifically from multi-phase analysis: the expectancy of RANGING long signals in a ranging/recovery phase showed materially higher positive expectancy than RANGING long signals produced during the high-volatility period following the prior bull run's peak. Same tactical ADX conditions, different macro context, different outcomes. The rolling live backtest revealed this pattern only after sufficient data accumulated across both macro contexts. A single historical backtest over one phase would not have shown the difference.
For the futures component, funding rate costs are estimated based on each position's expected duration and recent funding rate averages. Signals with expected holding periods exceeding 12 hours in strongly positive funding conditions receive an adjusted expected value incorporating the estimated funding cost. This is not standard in generic backtesting frameworks but is necessary for accurate futures expectancy calculation in crypto's high-funding environments.
The shadow data also tracks regime distribution over time, what percentage of bars were RANGING, TRENDING_BULLISH, and TRENDING_BEARISH. Shifts in this distribution between the backtest period and the live period are flagged as potential evidence that the market's macro character has changed, warranting a review of the strategy's phase-specific calibration.
Backtesting on bull market data only. 2020-2021 was an extreme bull market. Any long-only strategy with trend-following logic produced strong results. This does not indicate sustainable edge. It indicates the strategy captured the macro tailwind of one of the strongest bull runs in market history. Include at least 2018 or 2022 in the test period to evaluate bear market behavior.
Ignoring funding rates for futures. A perpetual futures strategy holding consistent long positions during a bullish period pays significant funding costs that never appear in the spot price. A strategy showing 15% annual return without funding costs may show 8% after realistic funding estimates. Always include funding rate assumptions in futures backtests, especially for any strategy with long average holding periods in trending markets.
Asset selection from current market cap rankings. Backtesting on today's top 10 or top 20 selects for survivors. The backtest shows how these specific assets performed, not how a real portfolio would have performed. Include assets that were historically prominent but have since declined or failed, or at minimum acknowledge that the results reflect a survivor-biased sample.
Insufficient out-of-sample periods. Given crypto's limited history, a standard 70/30 split on 3 years of data produces 10 months of out-of-sample testing. This is marginal. Compensate with continuous live tracking using shadow data from day one of deployment. The live data is the only true out-of-sample test for crypto strategies where the historical baseline is inherently limited.
Crypto backtesting applies a set of trading rules to historical cryptocurrency price data to measure how those rules would have performed in the past. It follows the same core methodology as backtesting any trading strategy, but crypto-specific considerations include limited historical data, exchange-specific data quality, pronounced bull and bear cycles that require multi-phase testing, and funding rate costs for futures strategies.
Define all rules before looking at results. Obtain historical OHLC data from the exchange being traded. Split the data chronologically: develop on the first 70% and validate on the remaining 30%. Apply the strategy rules to the in-sample period and measure performance. Apply unchanged to the out-of-sample period. Segment results by regime state (ADX trending vs ranging) and by macro market phase (bull vs bear vs recovery). Include realistic cost estimates including spread, slippage, and funding rates for futures positions.
The four most common mistakes specific to crypto: testing only on bull market data and assuming the results represent full-cycle performance; ignoring funding rates for perpetual futures positions; selecting assets from current market cap rankings, which introduces survivorship bias by excluding failed or collapsed assets; and not accounting for data quality issues around major market events where historical prices may not reflect actual tradeable fills.
Estimate the funding cost for each position based on expected holding duration and the average funding rate during the test period. For each 8-hour period a long position is open, subtract the estimated funding rate from the position's unrealized PnL. During bullish trending periods, funding rates can run at 0.01% to 0.03% per 8 hours, compounding to 2% to 9% per month in holding costs. This cost materially affects long-hold trend-following strategies and must be included for accurate futures expectancy calculation.
Use OHLC data from the exchange where live trading will occur, not a different exchange's data. This ensures the backtest reflects the actual prices and liquidity available. For Bitcoin strategies, at minimum use data spanning a complete bull and bear cycle. For altcoins, use the full available history from the asset's inception. Verify data quality around major market events by checking for gaps, zero-volume bars, or price anomalies that might indicate exchange outages or data errors.
Enough to span at least one complete bull and bear cycle, and to produce at least 100 trades in the strategy's intended conditions. For Bitcoin, this means going back to at least 2018-2019 to include a bear market phase alongside bull market data. For altcoins with shorter histories, use the full available history and compensate for the limited baseline with aggressive live shadow data tracking from the first day of deployment.
Yes, in four significant ways: crypto trades 24/7 with no session-based closing prices; crypto has much shorter market history (15 years maximum vs 50-100 years for major stock indices); crypto futures use perpetual contracts with funding rate costs that have no equivalent in equity futures; and crypto data quality is more variable due to exchange fragmentation, outages, and wash trading. The core methodology is the same, but the execution details differ significantly.