Backtesting and Optimization

RunStrategyBacktest is the MCP entry point into NT8's actual Strategy Analyzer engine. It takes a compiled NinjaScript strategy class and runs it through the same Strategy.RunBacktest() path the NT8 UI uses, so the returned trades and metrics are bit-identical to what you would see in NT8 Strategy Analyzer for the same configuration.

Two modes are supported on the same tool, switched by the presence of an optimization argument:

Single backtest: one run with the supplied parameters. Returns the full SystemPerformance dump with every trade and every metric NT8 computes.
Parameter sweep: cartesian enumeration of caller-supplied parameter ranges. Returns a ranked list of per-iteration summaries, sorted best-first.

Quick Start

Single backtest

RunStrategyBacktest(
  strategy_class:  "SampleMACrossOver",
  instrument:      "MES 06-26",
  bars_period:     {period_type: "Minute", value: 5},
  from:            "2026-04-01T00:00:00Z",
  to:              "2026-04-30T00:00:00Z",
  parameters:      {Fast: 10, Slow: 25},
  fill:            {slippage_ticks: 0},
  account:         {initial_cash: 100000}
)
→ { job_id, poll_with: "GetMcpJob" }

Poll GetMcpJob(job_id) until status: "completed". The result block contains trades[], performance.{all,long,short,realtime}, equity_curve[], and actual_bars_range.

Parameter sweep

RunStrategyBacktest(
  strategy_class:  "SampleMACrossOver",
  instrument:      "MES 06-26",
  bars_period:     {period_type: "Minute", value: 5},
  from:            "2026-04-01T00:00:00Z",
  to:              "2026-04-30T00:00:00Z",
  optimization: {
    fitness:           "MaxNetProfit",
    parameters_sweep: [
      {name: "Fast", min: 5,  max: 15, step: 5},
      {name: "Slow", min: 20, max: 30, step: 5}
    ]
  }
)

Returns a results[] array (sorted best-first by NetProfit) plus a best callout at the top level.

Required Arguments

Field	Type	Meaning
`strategy_class`	string	Bare NinjaScript class name (e.g. `MyEma20Reversal`). Must be compiled into `NinjaTrader.Custom.dll`.
`instrument`	string	Full instrument name (e.g. `MES 06-26`). Resolved through NT8's InstrumentManager.
`bars_period`	object	`{period_type, value, value2?, market_data_type?}`. See Bars period table below.
`from` / `to`	string	ISO-8601 UTC. `from` must be earlier than `to`.

Optional Arguments

Field	Type	Default	Meaning
`parameters`	object	`{}`	Map of `{property_name: value}` for every `[NinjaScriptProperty]` on the strategy. Unknown property names return `bad_parameter` with the full list of writable property names.
`trading_hours`	string	instrument default	NT8 TradingHours template name.
`fill`	object	see below	Fill/commission/slippage settings.
`account`	object	`{initial_cash: 100000.0, denomination: "UsDollar"}`	Synthetic isolated backtest account. The Sim101 account is reused with `ResetSimulationAccount(true)` so each backtest starts with a clean ledger.
`optimization`	object	absent	When present, switches the run from single backtest to parameter sweep. See Optimization section.
`timeout_ms`	int	none	Maximum elapsed time before the job aborts.

Bars period

bars_period.period_type accepts every value from NT8's BarsPeriodType enum (PascalCase): Minute, Tick, Second, Day, Volume, Range, Renko, HeikenAshi, Kagi, PointAndFigure, LineBreak, Volumetric, Delta, PriceOnVolume, Week, Month, Year.

Casing differs from GetBars

GetBars (the historical-bars query tool) uses lowercase period types (minute, day, week, month, year) and supports a smaller subset. RunStrategyBacktest.bars_period.period_type uses NT8's native PascalCase enum and supports the full set.

Fill settings

Field	Type	Default	Meaning
`type`	string	`Standard`	NT8 `OrderFillResolution` enum name.
`slippage_ticks`	int	`0`	Slippage applied per side.
`commission_template`	string	empty	NT8 commission template name from Tools → Commissions. Default is no commission to match NT8 Strategy Analyzer's out-of-the-box behavior.

Single-Backtest Result Shape

{
  "engine": "nt8_strategy_analyzer",
  "nt8_version": "8.1.6.3",
  "fingerprint": "sha256:...",
  "strategy": "SampleMACrossOver",
  "strategy_full_name": "NinjaTrader.NinjaScript.Strategies.SampleMACrossOver",
  "instrument": "MES 06-26",
  "from": "2026-04-01T00:00:00Z",
  "to":   "2026-04-30T00:00:00Z",
  "account": "Sim101",
  "trades": [
    {
      "TradeNumber": 0,
      "Quantity": 1,
      "Commission": 0.0,
      "ProfitCurrency": -21.25,
      "ProfitPercent": -0.00061,
      "ProfitPoints": -4.25,
      "ProfitTicks": -17.0,
      "MaeCurrency": 28.75,
      "MfeCurrency": 3.75,
      "EntryEfficiency": 0.12,
      "ExitEfficiency": 0.23,
      "TotalEfficiency": -0.65,
      "entry": {"time": "2026-04-13T17:15:00Z", "price": 6922.25, "quantity": 1, "market_position": "Short", "order_action": "SellShort", "name": "Sell short"},
      "exit":  {"time": "2026-04-13T19:25:00Z", "price": 6926.50, "quantity": 1, "market_position": "Long",  "order_action": "BuyToCover", "name": "Close position"}
    }
  ],
  "performance": {
    "all":  { "NetProfit": 800.00, "ProfitFactor": 1.1425, "SharpeRatio": 0.80, "TradesCount": 264, "..." },
    "long": { "NetProfit": 1718.10, "..." },
    "short":{ "NetProfit": -1366.90, "..." },
    "realtime": { "..." }
  },
  "equity_curve": [{"t": "...", "equity": 0.0}, "..."],
  "actual_bars_range": {"bar_count": 5979, "first_bar_time": "...", "last_bar_time": "..."},
  "state": "Finalized",
  "denomination": "UsDollar"
}

Every field of NT8's TradesPerformance (Currency/Pips/Points/Ticks/Percent sub-objects, Sharpe/Sortino/RSquared/Probability/ProfitFactor, MaxConsecutiveWinner/Loser, MaxTimeToRecover, LongestFlatPeriod, etc.) is dumped verbatim. If NT8 adds a metric in a future build, it shows up automatically.

Optimization (Parameter Sweep)

When optimization is present, MCP enumerates every combination of the supplied parameters_sweep ranges and runs one backtest per combination. Each combination's parameter values are pushed into the strategy via reflection on the property setters before the backtest runs, so the strategy code sees the swept values exactly as it would in NT8 Strategy Analyzer.

Optimization argument

optimization: {
  optimizer:                         "DefaultOptimizer",          // or GeneticOptimizer / StrategyGenerator
  fitness:                           "MaxNetProfit",              // or array for multi-objective
  keep_best_results:                 10,
  instantiated_on_each_iteration:    true,
  parameters_sweep: [
    {name: "Fast", min: 5,  max: 15, step: 5},
    {name: "Slow", min: 20, max: 30, step: 5}
  ],
  extra: { /* forwarded verbatim to the optimizer instance */ }
}

Optimization result shape

{
  "engine": "nt8_strategy_analyzer",
  "mode": "optimization",
  "strategy": "SampleMACrossOver",
  "expected_iterations": 9,
  "iterations_completed": 9,
  "optimization_config": { "..." },
  "optimizer": "NinjaTrader.NinjaScript.Optimizers.DefaultOptimizer",
  "fitness":   "NinjaTrader.NinjaScript.OptimizationFitnesses.MaxNetProfit",
  "results": [
    {
      "iteration": 1,
      "parameter_values": [{"name": "Fast", "value": 5}, {"name": "Slow", "value": 25}],
      "performance_value": 1277.50,
      "performance_summary": {"NetProfit": 1277.5, "ProfitFactor": 1.22, "SharpeRatio": 0.81, "TradesCount": 332, "..."}
    },
    "..."
  ],
  "best": { "iteration": 1, "parameter_values": [...], "performance_value": 1277.50, "performance_summary": {...} }
}

Results are sorted best-first by performance_value. The best field is a top-level callout for easy access.

Known optimization caveats

These limitations apply to the current optimization implementation. Plan around them; they are documented because they are real.

Only int / long / double / float / decimal parameter types are sweepable. Booleans, enums, and string parameters are accepted as fixed inputs via parameters but cannot be put in parameters_sweep.
All sweeps are exhaustive cartesian enumeration. The optimizer field ("DefaultOptimizer" / "GeneticOptimizer" / "StrategyGenerator") is resolved to a real NT8 optimizer instance and recorded in the response, but the iteration loop is driven by MCP itself. Genetic-algorithm semantics (random initial population, crossover, mutation) and StrategyGenerator's evolutionary search are NOT honored. Every sweep enumerates every combination.
Ranking is always by NetProfit. fitness (single or multi-objective array) is resolved to a real NT8 OptimizationFitness instance and reported in the result, but the actual sort key is currently performance_summary.NetProfit. Fitness-specific scoring (Sharpe, Sortino, ProfitFactor, drawdown minimization, etc.) is wired into the result metadata but not into ranking.
keep_best_results and instantiated_on_each_iteration are forwarded to the optimizer instance but do not affect MCP's cartesian loop. All iterations are kept and returned.
Sweep results omit full per-trade detail. Each iteration entry includes performance_summary (NetProfit, ProfitFactor, Sharpe, Sortino, trade counts) and the winning parameter_values, but not the full trades[] array. To inspect every trade for the best parameters, call RunStrategyBacktest again with parameters set to best.parameter_values and no optimization arg.

Failure Modes

Every error path returns a structured {success: false, error, detail, ...} envelope so AI clients can self-correct without a human in the loop.

Code	Meaning	Extra fields
`nt8_build_unsupported`	The user's NT8 build is missing one of the reflection targets the engine needs.	`missing_symbols[]`, `nt8_version`, `fingerprint`
`strategy_class_not_found`	`strategy_class` does not match any compiled NinjaScript Strategy.	`compiled_strategies[]`
`instrument_unknown`	`instrument` not resolvable through NT8 InstrumentManager.	`requested`
`bad_arg`	Invalid ISO-8601 date, From ≥ To, invalid `bars_period.period_type`.	`detail` includes the valid enum list
`bad_parameter`	Unknown property in `parameters` or coercion failure.	`parameter_name`, `expected_type`, `valid_parameters[]`
`no_historical_bars`	BarsRequest returned an empty bars set for the range.	`instrument`, `from`, `to`, `bars_period`
`trading_hours_unknown`	`trading_hours` template not found.	`requested`
`optimization_empty_sweep`	`parameters_sweep` is empty.	(none)
`optimization_parameter_unknown`	Sweep entry references a non-existent or non-writable strategy property.	`requested`, `optimizable_parameters[]`
`optimization_parameter_unsupported_type`	Sweep parameter type is not int/long/double/float/decimal.	(none)
`optimizer_unknown` / `fitness_unknown`	Name does not resolve against the probe's supported list.	`requested`, `available[]`
`bad_sweep_entry`	Missing `min`/`max`, or coercion failure on min/max/step.	(none)
`strategy_runtime_failure` / `optimization_runtime_failure`	An iteration's `OnBarUpdate` threw.	`exception_type`, `stack_trace`
`backtest_aborted`	User cancel or `timeout_ms` exceeded.	(none)

Parity Verification

The single-backtest path was verified bit-identical to NT8 Strategy Analyzer on a known reference run:

Strategy: SampleMACrossOver (ships with NT8)
Instrument: MES 06-26 (Globex Micro E-mini S&P 500)
Bars: 5-minute, April 1–30, 2026
Inputs: Fast=10, Slow=25
Account: Sim101, $100,000 initial cash, no commission, 0 slippage

Metric	NT8 SA UI	`RunStrategyBacktest`
Net Profit	$800.00	$800.00
Profit Factor	1.14	1.1425
Total Trades	264	264 (counter) / 263 (trade list, see note)
Sharpe	0.80	0.80
Long / Short	132 / 132	132 / 132
Long NetProfit	$1,718.10	$1,718.10
Short NetProfit	-$1,366.90	-$1,366.90

One-trade off-by-one

performance.all.TradesCount (the counter NT8 increments live during the run) reports 264, but the trades[] array length is 263. This matches NT8 Strategy Analyzer UI behavior: the counter includes a partial position straddling the date boundary; the trade-list view only shows completed round-trips. Use performance.all.TradesCount for "how many fills happened" and trades.length for "how many round-trip P&L events."

Performance

Scenario	Typical elapsed
Single backtest, 1 month of 5-min bars, ~5000 bars	0.5 – 1.5 s
9-iteration cartesian sweep (3x3), same date range	0.4 – 1.0 s (bars loaded once, reused)
100+ iteration sweep	scales linearly per backtest; bars are loaded once

The job manager auto-expires job records 30 minutes after completion. For very long sweeps, set timeout_ms to bound the run and poll GetMcpJob for progress_pct and progress_note.

Recommended Workflow

GetMcpCapabilities                       // confirm backtest_engine.available
GetNinjaScriptHelp(topic: "strategy_template")
LookupNinjaScriptSymbol(name: "Strategy")
CompileNinjaScript(source: <C# source>, in_memory: true)   // verify syntax
WriteNinjaScriptFile(name: "MyStrat", kind: "strategy", source: ..., overwrite: true)
RunStrategyBacktest(strategy_class: "MyStrat", instrument: ..., ..., optimization: {...})
GetMcpJob(job_id: ...)                                       // poll until completed
// Inspect best.parameter_values, decide whether to deploy
RunStrategyBacktest(strategy_class: "MyStrat", parameters: best.parameter_values)   // full single-backtest detail
DeployStrategy(strategy_class: "MyStrat", account: "Sim101", instrument: ..., parameters: best.parameter_values)

For copy-ready prompts, see MCP Trading with AI.

Quick Start​

Single backtest​

Parameter sweep​

Required Arguments​

Optional Arguments​

Bars period​

Fill settings​

Single-Backtest Result Shape​

Optimization (Parameter Sweep)​

Optimization argument​

Optimization result shape​

Failure Modes​

Parity Verification​

Performance​

Recommended Workflow​