Fix the data, timestamps, fees, and formula. Use a small worked example to see how costs affect compound results, then extend the research with independent samples and simulation.
Fix the data, timestamps, fees, and formula. Use a small worked example to see how costs affect compound results, then extend the research with independent samples and simulation. Fictional returns A=[4%,-2%], B=[2%,1%]. At fee f per trade, net compound is product(1+return-f)-1. Calculate at 0.2%, then 0.4%.
Fix the data, timestamps, fees, and formula. Use a small worked example to see how costs affect compound results, then extend the research with independent samples and simulation.
Fictional returns A=[4%,-2%], B=[2%,1%]. At fee f per trade, net compound is product(1+return-f)-1. Calculate at 0.2%, then 0.4%. Here f means a single abstract total cost deducted once per complete toy trade; it is not Freqtrade’s --fee setting, which charges entry and exit separately.
Use identical trades and assumptions, retain calculations, then test a holdout and realistic fills. Two trades cannot establish an edge or support a live recommendation.
Compare the example input with the result and acceptance checks.
Compare two cost levels on the frozen return data below. Here f is one abstract total cost deducted per complete toy trade, not Freqtrade’s --fee setting; two observations cannot support a performance claim.
returns:
A: [0.04, -0.02]
B: [0.02, 0.01]
cost_per_complete_toy_trade f: [0.002, 0.004]
net_return = product(1 + return - f) - 1Calculate at both fees; return JSON with exactly these string fields: fee_0_2pct_A, fee_0_2pct_B, fee_0_4pct_A, fee_0_4pct_B, limitation. Two trades cannot support a performance claim.
{
"fee_0_2pct_A": "1.5164%",
"fee_0_2pct_B": "2.6144%",
"fee_0_4pct_A": "1.1136%",
"fee_0_4pct_B": "2.2096%",
"limitation": "two fictional trades; no performance inference"
}Why it is included: Build reproducible simulations that address look-ahead, survivorship, overfit, and trading costs.
When it does not fit: Two toy trades are too small to choose or validate a production backtesting framework.
Why it is included: Compare strategies under the same pairs, timeframe, fees, and wallet.
When it does not fit: Do not compare strategy results from different datasets or time windows.
Why it is included: Vary the backtest fee explicitly to measure its impact per round trip and across the strategy.
When it does not fit: Include slippage, market impact, and funding when modeling trading costs.
Why it is included: Separate development and later validation windows instead of repeatedly tuning and reporting on one period.
When it does not fit: Two trades cannot provide a meaningful holdout performance estimate.
Why it is included: Check whether a strategy uses future candle information unavailable at decision time.
When it does not fit: A hand calculation with no time-series features has no lookahead path to inspect.
Why it is included: Calculate volatility, VaR/CVaR, drawdown, and risk-adjusted returns from aligned observations.
When it does not fit: Two observations cannot estimate stable drawdown or tail-risk metrics.
Why it is included: Use a simulated wallet and dedicated database to observe signals and fills without live orders.
When it does not fit: A local arithmetic example does not need exchange credentials or a paper account.