Skip to main content

Search a Backtest Parameter Grid

A parameter search compares predefined strategy variants under the same market data and execution assumptions. TradeJS builds the Cartesian product of the arrays in a saved backtest grid and evaluates each combination for each symbol.

Estimate the Search Size​

If MA_FAST has three values and MA_SLOW has two, the grid has six parameter combinations. Across 20 symbols, the full suite contains 120 tests before the --tests limit.

Keep the grid small enough to explain. A larger search increases compute cost and the chance of selecting noise. Parameters and ranges should follow from the strategy hypothesis rather than from repeated inspection of the same period.

Example Grid​

{
  "INTERVAL": ["15"],
  "MAX_LOSS_VALUE": [10],
  "MA_FAST": [14, 21, 34],
  "MA_SLOW": [55, 89],
  "LONG": [
    { "enable": true, "direction": "LONG", "TP": 2, "SL": 1 }
  ],
  "SHORT": [
    { "enable": true, "direction": "SHORT", "TP": 2, "SL": 1 }
  ]
}

Every value must be an array. See Define a backtest parameter grid for storage and naming.

npx @tradejs/cli backtest \
  --user root \
  --config MaStrategy:grid-v1 \
  --connector bybit \
  --timeframe 15 \
  --tickers BTCUSDT,ETHUSDT \
  --tests 12 \
  --parallel 4 \
  --cacheOnly

Useful controls:

  • --tickers selects explicit symbols;
  • --tickersLimit caps a connector-derived symbol set;
  • --exclude removes symbols;
  • --tests caps the complete symbol × parameter suite;
  • --parallel controls worker count;
  • --skip and --continue help resume planned work.

Use --cacheOnly when comparing configurations on a fixed historical dataset. If you refresh history, record that change as a new experiment input.

Select and Validate​

Inspect the saved results and symbol coverage:

npx @tradejs/cli results \
  --strategy MaStrategy \
  --coverage \
  --user root

When selection uses the same period as evaluation, the reported best result is in-sample. Reserve an independent period, test nearby parameter values, and stress costs before considering a live evaluation.

results --merge can keep a local list of the highest recorded result for each symbol. It does not update tradejs.config.ts or create a portfolio decision.

Common Mistakes​

  • using a scalar instead of an array;
  • leaving an empty array, which creates zero combinations;
  • omitting the strategy prefix from StrategyName:label;
  • changing the grid after seeing results without starting a new experiment;
  • searching too many combinations without an independent validation set;
  • ranking by profit alone.