Define the return series before calculating
The historical Sharpe ratio is the average of period-by-period excess returns divided by their standard deviation. For each period, subtract the risk-free or specified benchmark return from the strategy return. A constant benchmark reduces this to (average return − benchmark return) / standard deviation of returns. Use matching units and periods; do not subtract an annual rate directly from monthly returns.
In the example above, 0.40 describes monthly excess return relative to monthly variability. A conventional square-root-of-12 conversion gives about 1.39, but depends on assumptions about aggregation and serial correlation. Zero standard deviation makes the mathematical ratio undefined; a software zero is a placeholder, not a measured risk-adjusted return. William Sharpe’s definition and time-dependence discussion.
What values of 0, 1 and 2 actually establish
With a positive denominator, a negative Sharpe means the sample mean underperformed its benchmark; zero means the mean excess return was zero. A value of 1 or 2 means one or two units of average excess return per unit of standard deviation on that basis. None establishes a probability of profit, a maximum loss or a safe position size.
A higher positive Sharpe is a higher historical excess-return-to-variability ratio only when the measurement conventions are comparable. A short backtest selected from many trials is different evidence from an untouched test period. There is no supported fixed percentage haircut, deduction per optimization pass, or trade-count threshold that turns an optimized score into a live forecast.
TradingView and Pineify use different inputs
TradingView’s support page describes its strategy Sharpe using monthly returns and a default risk-free rate of 2% annually, configurable through the strategy declaration. It is not defined there as the mean of exported trade returns. TradingView: Sharpe ratio.
Pineify Backtest Deep Report currently uses the imported trade P&L percentages. It subtracts the configured risk-free value from the mean trade percentage, divides by the population standard deviation, and multiplies by the square root of trades per year. That frequency is estimated from the earliest entry to the latest exit. The rolling chart uses a 20-trade window without the annualization multiplier.
This trade-based statistic is not a reconstruction of a calendar-time account-equity Sharpe. Irregular holding periods, overlapping trades, idle capital and position sizing can make the two measures materially different. A difference is not evidence that TradingView inflated its number or that one report found the “real” Sharpe.
The report currently subtracts its Risk-Free Rate (%) input directly from each trade-return mean; it does not convert an annual rate to a per-trade rate. A nonzero annual input therefore does not produce a period-matched excess-return calculation. The default zero avoids that conversion mismatch but describes a zero-benchmark trade statistic. Treat the result as an explicitly defined diagnostic, not a like-for-like replacement.
Check costs, dependence and selection
Use the same instrument, dates, capital basis, position sizing and cost assumptions when comparing results. Commission and slippage belong in the source simulation; an exported-trade analyzer cannot restore omitted fills or expenses. TradingView: simulating trading costs.
A large trade count does not guarantee many independent observations. Several overlapping entries can express one market exposure, and a long calm period may contain little evidence about a stressed market. Record the observation frequency, calendar span, number of parameter trials and untouched test interval alongside the score.
Square-root annualization is a convention with assumptions, not a correction for dependence. Serial correlation and compounding can invalidate a simple scaling. Repeatedly selecting the highest score from historical variants also changes the meaning of the selected result. Neither problem is repaired by adding more random simulations to the same selected trade sample.
Read drawdown and tail losses separately
Standard deviation summarizes dispersion around the mean; it does not describe the timing of losses or the full return distribution. Read the equity path, worst observations, drawdown depth and recovery duration alongside Sharpe. A high score can coexist with a large loss outside the observed sample.
Sortino replaces total variability with downside deviation relative to a specified target. Its normalization and target must be stated. The ordering of Sharpe and Sortino does not by itself diagnose symmetry, positive-outlier dependence or safety. Pineify’s current downside denominator averages squared negative trade returns over the negative trades only; compare only with implementations using the same convention. See TradingView’s Sortino definition for its separate methodology.
Calmar relates annualized return to maximum drawdown, while recovery factor relates net profit to drawdown in matching money units. Neither measures time to recovery. VaR is a loss quantile for a specified horizon and model, not a maximum possible loss; expected shortfall examines the tail beyond that threshold.
Use an untouched test period, then track forward results
Choose the training and test periods before selecting parameters. Evaluate the fixed strategy on the later test period without tuning to that result. If you repeatedly inspect and revise against it, it is part of model selection. Forward testing adds new observations but still cannot establish a future guarantee. TradingView: overfitting.
Pineify’s Monte Carlo feature samples imported money P&Ls with replacement. It explores outcomes conditional on that empirical sample; it does not detect every source of overfitting or create unseen market shocks. Simply reordering a fixed list of money P&Ls cannot change its total, although it can change drawdown. No simulated loss-rate cutoff proves a strategy is ready for capital.
CSV parsing and numerical calculations run in the browser. If you request AI Analysis, the app sends metrics and summarized monthly, rolling and available simulation results to its API. AI commentary is an interpretation to verify, not evidence that the backtest predicts live returns. For the broader review, use the backtest profitability checklist.
Frequently asked questions
Is a Sharpe ratio above 1 good?
It means one unit of average excess return per unit of standard deviation on the stated basis. Whether that is useful depends on the sample, costs, benchmark, return frequency and selection process; it is not a universal trading threshold.
Is a Sharpe ratio above 3 an error?
Not necessarily. Check input units, return frequency, annualization, cost assumptions and independent test results. The value alone proves neither an error nor an investable strategy.
How many trades make Sharpe reliable?
No fixed count guarantees reliability. Calendar coverage, independent observations, tail losses and the number of strategies tried all matter. Trade count alone is especially incomplete for a statistic computed from monthly equity returns.
Why does Pineify differ from TradingView?
TradingView documents monthly returns for its strategy Sharpe. Pineify currently uses imported trade-return percentages and a trades-per-year multiplier. The datasets and conventions differ, so the numbers should not be treated as interchangeable.
Can I enter an annual risk-free rate in Pineify?
The current report subtracts the input directly from mean trade-return percentages without converting annual to per-trade units. A nonzero annual value therefore does not produce a matched-period calculation. The default zero describes a zero-benchmark trade statistic.
Does Monte Carlo validate a high Sharpe?
No. Resampling the selected trades explores a model based on that sample. It does not supply independent market observations or remove the bias from selecting the best historical strategy.
Is the report free and entirely offline?
Sign-in is required, and uploading your own CSV requires Advanced or a higher plan. CSV parsing and numerical calculations run in the browser. Requesting AI Analysis sends report summaries to the API.