A strong-looking backtest number is easy to read and easy to misuse. Sharpe, Sortino and profit factor use different denominators and often different ledgers. Before comparing them, state the return series, interval, fees, annualization convention and rule for missing or open trades. The formulas below use invented data to show the arithmetic; they do not describe a crypto market result.
Sharpe and Sortino answer different questions
For periodic returns, the Sharpe ratio is mean excess return divided by the standard deviation of excess returns over the same interval. This guide uses the sample standard deviation. Its result changes if you use trade returns rather than evenly spaced account returns, omit financing costs or choose another risk-free input.
The Sortino ratio replaces total volatility with downside deviation below a stated target. One explicit convention is mean(return − target) ÷ sqrt(mean(min(0, return − target)²)), where the downside squares are averaged across all observations, including zero shortfalls. Other tools can use a different target or denominator; record the one your report uses. CFA Institute’s risk-adjusted performance overview describes the Sortino ratio as a complement to other downside-risk measures.
Consider four fictional monthly returns, already net of an assumed model cost: +1%, −1%, +1%, +1%. Assume a zero monthly risk-free return and a zero Sortino target. The mean is 0.5%. The sample standard deviation is 1.0%, so the unannualized monthly Sharpe is 0.50. Under the downside convention above, only one month falls below target: downside deviation is sqrt((1%²) ÷ 4) = 0.5%, making the monthly Sortino 1.00. These precise-looking ratios come from just four observations.
If monthly returns were independent and identically distributed, the familiar annualized Sharpe shortcut would be 0.50 × sqrt(12) ≈ 1.73. That dependence assumption can fail: Andrew Lo’s Sharpe-ratio analysis shows that serial correlation changes annualization. Four observations cannot establish whether this sample is independent or estimate that dependence reliably.
The same four returns turn a fictional $1,000 starting balance into about $1,020: $1,000 → $1,010 → $999.90 → $1,009.90 → $1,020. The largest month-end peak-to-trough decline is 1.0%, from $1,010 to $999.90. That path adds information the final return and ratios do not show.
Here is a second invented sequence with a similar rounded ending balance but a much larger recorded decline. Both start at $1,000, use four monthly observations, and round equity to cents:
| Month-end | Series A: +1%, −1%, +1%, +1% | Series B: +10%, −10%, +3.03%, 0% |
|---|---|---|
| Start | $1,000.00 | $1,000.00 |
| 1 | $1,010.00 | $1,100.00 |
| 2 | $999.90 | $990.00 |
| 3 | $1,009.90 | $1,020.00 |
| 4 | $1,020.00 | $1,020.00 |
| Month-end maximum drawdown | 1.0% | 10.0% |
The unrounded month-four balances differ by less than one cent; each displays as $1,020.00. Series B first falls 10% from its $1,100 month-one peak to the $990 month-two trough, while Series A’s largest decline is 1% from $1,010 to $999.90. Monthly sampling can miss a deeper move between observations, so these are month-end drawdowns, not claims about each path’s intramonth risk.
Profit factor belongs to the trade ledger
Profit factor is gross profit from winning trades divided by the absolute gross loss from losing trades. In a separate invented ledger, winners of $20 and $10 against one $15 loss give ($20 + $10) ÷ $15 = 2.00 before costs. If each of those three completed trades is charged a hypothetical $1 round-trip cost, net winners total $28 and the net loss is $16, so profit factor falls to 1.75. The cost is illustrative, not a venue quote.
Profit factor hides when trades occurred, how much capital was tied up and how many opportunities were rejected or unfilled. Report the number of trades, holding periods, exposure and cost treatment beside it. If the sample contains no losing trades, the denominator is zero: report the ratio as undefined or explain the platform’s treatment. Do not present it as proof of safety.
Read every metric with its path and process
Maximum drawdown depends on how often equity is valued and whether open positions are marked. Trade-based returns, hourly returns and daily returns can produce different volatility and Sharpe estimates. Exposure, turnover, losing sequences, baseline choice and the full equity curve help explain the result. State whether returns include fees, funding and open-position marks; compare strategies on the same sampling and cost convention.
If many assets, rules, parameters or intervals were tested, the best score has been selected from that search. Bailey and López de Prado’s Deflated Sharpe Ratio paper discusses adjustment for multiple trials and non-normal returns. The adjustment still depends on assumptions and cannot repair look-ahead bias, inaccurate data or unrealistic fills. Keep the trial count and discarded variants in the research record. The performance-metric audit worksheet separates the periodic-return series, trade ledger and equity marks before you compare ratios.
SuperQuant’s bundled histories are synthetic and its execution rules simplified. Its metrics describe the specified simulation, not a verified live return or standardized comparison with other platforms. Check the engine’s calculation, return interval and cost settings before interpreting a displayed value. Continue with the backtesting workflow, cost analysis and overfitting guide. This is educational research guidance, not financial advice or a performance promise.