Appearance

Personalize your experience

Mode
Accent Color
Direction
Step 7 — Strategy Testing

Three testing modes

The testing step offers three modes, each answering different about behavior. Use all three in sequence: backtest to validate logic, forward test to check robustness, and paper trading to real-time execution.

Backtest — historical data

  • ·Runs the strategy over past price data
  • ·Shows what would have happened if you traded this in history
  • ·Fast — processes months or years in seconds
  • ·Best for: validating strategy logic and parameter tuning

Forward test — walk-forward analysis

  • ·Splits data into training and testing windows
  • ·Trains on in-sample data, tests on out-of-sample data
  • ·Repeats over multiple windows to detect overfitting
  • ·Best for: assessing robustness and generalization

Paper trading — real-time simulation

  • ·Connects to live market data feeds
  • ·Places simulated orders at real market prices
  • ·Tests order execution, latency, and slippage
  • ·Best for: validating real-time behavior before going live

Backtest — historical validation

Backtest runs the against historical . For every bar, it evaluates entry and exit conditions, places orders at the close price (or next open, depending on configuration), tracks positions, and calculates profit and loss. The result is a , equity curve, and .

Backtest configuration

FieldTypeDefaultDescription
startDatestring1 year agoStart of the backtest . Earlier dates give more but may include irrelevant market regimes.
endDatestringtodayEnd of the backtest . Leave as "today" to test up to the most recent .
initialCapitalnumber100000Starting account balance. Position sizing and scale to this number.
commissionnumber0.1Commission per trade as a . Add slippage separately.
slippagenumber0.05Slippage per trade as a . Models the gap between signal and fill price.
symbol*string—The to backtest. Must match a symbol in the feed.
timeframe*string— interval: 1m, 5m, 15m, 1h, 4h, 1d. Shorter intervals produce more .
broker*string— source. Different brokers have different historical coverage and instruments.

Reading backtest results

After a backtest completes, the shows key , an equity curve, and a trade . Focus on these metrics first:

MetricDescription
Total TradesNumber of completed . At least 30–50 trades are needed for significance. Fewer trades mean are luck, not edge.
Win Rate of profitable . A good can have 40–60% if winners are larger than losers.
Profit FactorGross profit divided by gross loss. Values above 1.5 are solid; below 1.2 is marginal. Below 1.0 means the loses money.
Max DrawdownLargest peak-to-trough decline in equity. This is what you experience during bad stretches. Can you tolerate this loss psychologically?
Sharpe RatioRisk-adjusted return. Above 1.0 is good, above 2.0 is excellent. Negative means returns do not justify the risk.
Average TradeMean profit per trade. Should be positive and larger than commission + slippage.
!
Caution:A backtest with 10 and 80% proves nothing. matters. Run the backtest over a long enough to generate 50+ trades, ideally 100+. Otherwise, the are noise.

Equity curve and drawdown

The equity curve shows account balance over time. A upward curve is ideal; a jagged one with deep dips signals high volatility. The overlay shows how far equity falls from its peak — this is the pain you endure during losing streaks.

ℹ
Note:Look at the equity curve , not just the final number. A that 100k into 150k with three 40% drawdowns is far riskier than one that reaches 130k with a 15% . The second is more tradeable.

Forward test — walk-forward analysis

Forward testing splits the into multiple windows: train on data, test on out-of-sample data, then the forward and repeat. This detects overfitting—if the performs well in-sample but fails out-of-sample, it memorized the training data and will not generalize to trading.

How walk-forward works

Walk-forward window structure
Window 1:  Train 2020-2021  →  Test 2022-H1
Window 2:  Train 2021-2022  →  Test 2022-H2
Window 3:  Train 2022-H1-H2 →  Test 2023-H1
Window 4:  Train 2022-H2-2023-H1 →  Test 2023-H2

Each window trains on the most recent N months, tests on the next M months.
If test performance degrades significantly, the strategy overfit the training data.

Interpreting walk-forward results

OutcomeDescription
Train +30%, Test +28%Robust — performance out-of-sample. The generalizes well.
Train +45%, Test −10%Overfitted — memorized noise. Simplify the or add regularization.
Train +15%, Test +22%Robust and fortunate — out-of-sample was kinder than . Still valid.
Train −5%, Test −8%Consistently unprofitable. The has no edge. Redesign from scratch.
★
Pro tip:If walk-forward shows overfitting, simplify the . the least conditions, reduce the number of parameters, or widen indicator periods. Simpler strategies generalize better.

Paper trading — real-time simulation

Paper trading connects to market and places simulated orders. The evaluates conditions on every new bar, generates signals in real time, and submits orders to a simulated broker. execute at market prices, but no real money moves.

What paper trading tests that backtest does not

  • Signal — how long between and order submission?
  • feed lag — is data delayed compared to what the backtest assumed?
  • Order rejection — do brokers reject orders for margin, size, or price limits?
  • Slippage — are worse than backtest estimates?
  • boundaries — do positions close correctly at market close for intraday strategies?
  • Gap behavior — does the handle overnight or weekend gaps as expected?
  • Re-entry execution — do layer in without errors?
!
Caution:Paper trading is mandatory before . A backtest assumes perfect at ; live markets do not behave that way. Paper catches execution issues before real capital is at risk.

How long to paper trade

Run for at least 20–50 signals. For intraday strategies, a week usually suffices. For positional strategies, two to four weeks covers different market conditions. Stop earlier only if a critical bug appears.

ℹ
Note:Monitor paper trading actively for the first few . Check that signals fire when expected, orders execute at reasonable prices, and triggers correctly. Most issues early.

Common backtest pitfalls

PitfallDescription
Look-ahead biasUsing from bar N to make a that affects bar N. Example: entering at the low of a bar, which is only known after the bar closes. Always use or next bar open.
Survivorship biasTesting only on instruments that still exist. Delisted or bankrupt stocks disappear from the , inflating . Use survivorship-free datasets.
OverfittingTuning parameters until the backtest looks perfect. The memorizes noise instead of signal. Use walk-forward to detect this.
Ignoring costsRunning backtests with zero commission and slippage. Real trading has friction. Set commission to 0.1–0.2% and slippage to 0.05–0.1% as a .
Small sample sizeConcluding from 10 that the works. noise dominates with samples. Aim for 50+ trades, ideally 100+.
Cherry-picking datesTesting only on bull markets or favorable periods. The will fail when conditions change. Test over at least one full market cycle.

Before moving on

  • Backtest run over at least one year of , generating 50+
  • , , and Sharpe ratio are acceptable
  • is tolerable — could you stomach that loss streak?
  • Walk-forward test shows consistent out-of-sample performance
  • No significant performance degradation between train and test windows
  • Paper trading planned: and for moving to
  • Commission and slippage set realistically in backtest
  • reviewed for anomalies: unrealistic , gaps, or errors

Related articles

Related to Step 7 — Strategy Testing

  • Step 0 — Particulars

    Automation / Strategy Builder / Wizard Steps

    Identity and metadata: the name, description, tags and visibility that make a strategy findable a year from now. Also where the immutable Strategy ID is minted.

  • Step 1 — Settings

    Automation / Strategy Builder / Wizard Steps

    Global behaviour that decides how entries and exits interact: whether the strategy builds a position in stages, which multi-entry model it follows, and how…

  • Step 2 — Tagging

    Automation / Strategy Builder / Wizard Steps

    Optional. Client order ID templates, signal-level metadata and order linking — the machinery that makes live orders reconcilable against broker statements.

  • Step 3 — Symbol

    Automation / Strategy Builder / Wizard Steps

    The trading context: broker, segment, instrument, timeframe and session mode. These choices decide what data exists, what order types are available, and…

  • Step 4 — Quantity & Capital

    Automation / Strategy Builder / Wizard Steps

    How much to trade. Four sizing modes, the capital they draw on, the leverage applied, and how capital is divided across entry sets.

  • Step 5 — Entry Signals

    Automation / Strategy Builder / Wizard Steps

    Where the strategy is actually written. Conditions, operators, nesting, order configuration, multiple entry sets, and the re-entry engine behind Grid, DCA…

  • Step 6 — Exit & Risk

    Automation / Strategy Builder / Wizard Steps

    Exit conditions, stop loss, take profit, staged partial exits and portfolio-level limits. The step that decides whether a losing trade costs you a percent…

  • Step 8 — Review & Deploy

    Automation / Strategy Builder / Wizard Steps

    Review is the final step before deployment. It presents the complete strategy configuration in one place, runs validation checks, highlights errors and…

  • Execution Modelling

    Backtesting / Execution Modelling

    Commission, slippage, spread and fill assumptions — the difference between a backtest that means something and one that flatters. Costs are where most…

Browse all knowledge base articles →