Analyze TradingView CSV: Reconcile the Trades Before Reading the Metrics

Read each metric together with its input series and formula. This report uses trade records, which do not contain a continuous account equity history. Overlapping positions, variable holding periods and missing excursions can make an apparently familiar label measure something different from another report.

Risk-adjusted returns: Sharpe, Sortino, Calmar, SQN

The current Sharpe calculation uses exported trade percentages, their population standard deviation and an annualization factor derived from trade frequency. Sortino uses the root mean square of negative trade percentages, with only negative observations in that denominator. SQN also uses trade percentages, not risk-normalized R multiples. These conventions do not establish skill or match every platform. The risk-free input is not frequency-normalized before subtraction; verify its units rather than inserting an annual rate unquestioningly.

The Calmar ratio calculator and the SQN calculator give you a standalone deep dive on each metric if you want to understand the math behind them.

Drawdown and recovery: Ulcer Index, Recovery Factor, UPI

The cumulative curve adds net trade P&L in parsed trade order. The parser sorts by entry time, so overlapping exits can produce a sequence different from actual realized cash timing. It also cannot reconstruct floating losses inside a trade. Ulcer Index averages squared percentage drawdowns at these observation points, without calendar-duration weighting. Recovery Factor divides net profit by the maximum cash drawdown; Calmar compares annualized return with percentage drawdown. Zero denominators and short samples require manual review rather than a favorable grade.

For a detailed breakdown of each, the Ulcer Index calculator and the Recovery Factor calculator walk through the formulas and interpretation.

Tail risk: Value at Risk and CVaR

The report estimates historical VaR95 from the lower tail of exported trade percentages and CVaR95 from all observations at or below that threshold, including ties. Neither is a guaranteed future loss ceiling or automatically an account-level daily risk measure. In a synthetic 20-trade sample with -10%, -4% and eighteen +1% values, the current index convention selects -4% and averages the two lower values to -7%. That tail contains 10% of this small sample, not exactly 5%. Compare definitions in the CVaR guide and VaR guide.

Distribution: Skewness and Kurtosis

Skewness measures asymmetry of the trade-percentage distribution; it does not determine the number of winning days. Excess kurtosis compares the fourth standardized moment with the normal-distribution reference of zero. Both estimates can be dominated by a few observations. Inspect the actual extreme trades and their data quality instead of inferring luck, fraud or a stable future tail probability from one number.

Rolling analysis uses a 20-trade window and recalculates trade-level Sharpe, Sortino and win rate. Each neighboring window shares most of its observations, and a window can span very different calendar durations. The rolling ratios are not annualized in the same way as the full-sample ratios.

A change in rolling results is a reason to inspect the underlying trades, costs and market conditions. It does not establish that a regime changed or predict when future performance will weaken. Record any rule change as a new research decision and evaluate it on data that did not prompt the change.

Monthly, weekly and daily views group closed results by exit dates; the time-of-day view groups by entry time. Browser-local date interpretation can affect the buckets. Check timestamps and timezone before comparing a heatmap with the chart or a broker report. A profitable month in one sample is not evidence of a recurring seasonal edge.

The histogram and normal reference are descriptive comparisons of imported trade returns. Sparse tails, dependent trades and a limited date range make extrapolation uncertain. Inspect each large gain and loss, and preserve excluded or missing records rather than cleaning away inconvenient outcomes.

Optional AI analysis sends derived metrics and monthly, rolling and Monte Carlo summaries to the analysis service. The request excludes the full raw trade and equity arrays. It still transmits information about the backtest, so the local numerical calculation should not be described as a guarantee that all analysis stays on the device.

Treat the generated interpretation as a set of claims to check. It cannot inspect source-code lookahead or reconstruct missing fills from a compressed summary. A numerical score is not a calibrated probability of profit, and suggestions require an independently specified test before adoption.

Use the following sequence to separate import correctness, statistical description and an eventual forward test.

1. Save the original TradingView result. Keep the script version, symbol, timeframe, dates, costs and execution settings with the report.
2. Export compatible trade records. Use the trade-list CSV where offered; retain the original workbook if converting XLSX. Do not rename XLSX as CSV.
3. Reconcile the import. Check IDs, paired entries and exits, timestamps, currency, trade count and total net P&L before ratios.
4. Declare report assumptions. Set capital, direction and open-trade handling. Record the return units and risk-free convention.
5. Inspect the observed path. Compare the trade-order curve with actual exit timing and investigate overlapping positions and missing fields.
6. Run conditional scenarios. Interpret bootstrap results as resampling of this sample; test dependence, costs and unobserved losses separately.
7. Save a reproducible report. Reconcile workbook sheets with the selected filters and keep all original files and settings.
8. Design the next test. Freeze a hypothesis before later data or paper trading. Report failures and uncertainty alongside gains.

The point of the report is to make assumptions and discrepancies visible. Preserve failed checks and alternative settings alongside successful results. There is no elapsed-time promise or threshold score that turns this review into evidence of future returns.