How to backtest a grid strategy before deploying capital

A backtest replays your grid configuration against real historical price data. It doesn't tell you what will happen next — no backtest does — but it tells you how the strategy would have performed in a specific past period. Run it across several different periods and regimes and you get a much more honest picture than any Monte Carlo simulation based on synthetic price paths.

What you need before starting

You need an OHLCV CSV file covering the period you want to test. OHLCV stands for Open, High, Low, Close, Volume — the standard candle format. The simulator accepts any CSV with those five columns plus a timestamp, in any reasonable format. The file never leaves your browser.

Candle size matters significantly for backtest accuracy. Large candles — 4-hour or daily — can sweep through multiple grid levels in a single step, making fill order ambiguous. The simulator handles this gracefully, but smaller candles produce more accurate results. 15-minute to 1-hour candles are the practical sweet spot for most backtest periods.

Where to source OHLCV data

Most exchanges and data aggregators offer CSV exports of historical OHLCV data. Look for a "Data" or "Market Data" section, or an export button on the chart view. Export in CSV format and make sure the file includes a timestamp column alongside the OHLCV fields.

Aim for at least 90 days of data if you want the result to be meaningful. A 30-day backtest on a quiet month tells you very little. Test across periods that include at least one trend move and at least one period of consolidation.

Step-by-step walkthrough

1. Configure your grid parameters first. Set direction, entry price, range, grid count, leverage, fees, and funding rate in the left panel before touching the backtest tab. The backtest runs the exact configuration you have set — it is not a separate input form.
2. Open the Backtest tab. In the simulator, switch from the Monte Carlo tab to the Backtest tab. You will see a file drop zone.
3. Upload your CSV. Drag the file onto the drop zone or click to browse. The simulator parses the file and displays a summary: row count, detected interval, date range, and price range. Check these — a detected interval that doesn't match your expected candle size means the file has gaps or formatting issues.
4. Review the data quality panel. The simulator shows a fill accuracy rating and any warnings — gaps in data, timestamps out of order, inconsistent intervals. Address errors before running. Warnings are informational; errors will affect results.
5. Run the backtest. Click Run. The engine replays price candle by candle against your grid configuration, fills orders when price crosses a level, and tracks P&L, funding, and position through the entire period.
6. Read the results. The backtest produces a single deterministic path — one set of results for the exact historical period. You will see total P&L, number of round trips completed, funding paid, and whether the bot would have been liquidated.

What to look for in the results

A clean backtest result shows a P&L curve that grows steadily when price oscillates within the range, flattens when price is trending, and drops when price breaks out. This is the expected pattern. If the curve grows monotonically without any flat or negative periods, the backtest period was unusually favourable — treat that result with scepticism.

Pay attention to the oscillation count alongside the P&L. If the bot completed 400 round trips in 30 days on a specific historical period, consider whether that oscillation rate is realistic for future deployment or whether it reflects an unusually volatile or range-bound past period.

Funding paid is shown separately. Cross-reference it against total gross income to understand what proportion of earnings was consumed by carry costs.

Combining backtest and Monte Carlo

The two tools answer different questions. The backtest tells you what happened on specific real price data. The Monte Carlo tells you the distribution of outcomes across many plausible synthetic paths calibrated to a volatility assumption. Neither is more correct. A strategy that looks good on both — strong median outcome in Monte Carlo, reasonable results across multiple backtest periods — is more convincing than one that only looks good on one.