Monte Carlo for discretionary zone and event traders

Your trade made sense because price returned to a zone you marked before the session. Or because a scheduled announcement changed the market. Move that entry to another date and the setup disappears.

That is a reasonable objection to “randomizing your trades.” The useful distinction is what gets randomized: the recorded outcome, rather than the market conditions that produced the entry.

A Monte Carlo trade-resampling exercise asks how a collection of outcomes behaves under a chosen sampling model. For a discretionary trader, that can expose losing-run pressure hidden by an attractive historical return. Its usefulness depends on whether the trade sample and risk assumptions describe the way you actually trade.

The trade outcome is the observation

Here, a trade means a completed entry-and-exit outcome: for example, +$350 or −$120. It is not an individual transaction printed by an exchange. If your journal splits one idea into several partial exits, keep track of that relationship; those rows do not automatically represent independent bets.

Suppose you have a consistently recorded set of zone-retest trades. The chronological journal answers questions about context: which zones existed, when you recognized them, and whether the event or session conditions matched your plan.

Resampling asks a different question: if outcomes resembling this sample arrived in another sequence, how uncomfortable could the account path become? A March outcome placed later in a synthetic sequence does not imply that March’s prices or news could have occurred in November. Keep the original dates for investigating the setup; read the resampled path as a conditional risk exercise.

The same profit can involve a different drawdown

Consider a deliberately simple teaching example, with $10,000 starting equity, three +$1,500 outcomes and three −$1,000 outcomes. The amounts are synthetic, fixed account-currency P&L, with no open-position fluctuations, deposits, or execution constraints. They are not product results or suggested position sizes.

Same profit, different account pressure

Alternating

W L W L W L

8.70%Maximum drawdown

Losses first

L L L W W W

30.00%Maximum drawdown

Both finish at $11,500. The account path changes.

Every path makes the same $1,500. In the losses-first sequence, equity reaches $7,000 before recovering. In the wins-first sequence, it reaches $14,500 before falling to $11,500. Maximum drawdown measures the largest decline from a preceding equity peak, including starting capital.

The decision is whether that path would cross a capital limit or make the process impractical to follow. Final profit alone does not answer it. This tiny example illustrates arithmetic; six trades cannot establish the distribution of a real strategy’s future outcomes.

Choose the randomization that matches the question

Shuffle changes order without replacement. Every recorded outcome appears exactly once. For the fixed dollar model above, total P&L stays constant while drawdowns and losing streaks can change.

Independent bootstrap draws outcomes with replacement. A loss can appear several times; a winner can disappear. Both the sequence and the total can change. Sampling with replacement is the standard bootstrap mechanism described in SciPy’s documentation.

Block bootstrap samples groups of neighboring trades. It retains order within a selected group, which can preserve some short-run dependence. New group boundaries still change the sequence. Different block algorithms handle boundaries differently, as the arch time-series bootstrap documentation explains.

Parameter perturbation changes inputs and reruns the strategy through chronological market data. Varying a zone-width threshold tests the rules’ sensitivity to that definition. It does not answer the same question as rearranging already completed outcomes.

Random sampling explores the selected model. It does not guarantee that the simulation finds the worst possible ordering.

Record what your sizing model actually does

Fixed dollar outcomes, percentage account returns, and R-multiples describe different experiments. An R-multiple expresses the outcome relative to the trade’s defined initial risk; compare trades only after checking that risk definition and treatment of costs.

A wide zone can require a smaller quantity to maintain the same dollar risk. Rearranging realized dollar P&L does not recalculate that quantity against each new synthetic balance. To investigate a different sizing policy, the model must apply that policy explicitly.

With a constant 1% of current equity lost on each stopped trade, ten consecutive full-risk losses reduce the balance by about 9.56%. At 2%, the reduction is about 18.29%. The calculation is 1 − (1 − risk fraction)^10, assuming each loss equals the intended risk and there are no extra costs or gaps.

That percentage decline is the same whether the isolated streak starts early or after profits. Under unconstrained fixed-fraction compounding, rearranging the same percentage-return basket also preserves final equity. Changing the basket, or introducing capital limits, minimum sizes, overlapping exposure, or liquidation rules, changes the experiment.

Stratifyre’s trade-resampling mode adds sampled, recorded net account P&L to starting capital. It does not resize every trade or rerun margin checks along each synthetic path. Its entry point is a saved strategy’s baseline backtest; the journal discussion here does not describe an external trade-log import feature.

Zones and events make dependence important

Five losing trades around one announcement may be five expressions of one event. Trades across correlated instruments can share the same exposure. A trend-following setup may lose repeatedly during chop and win during a sustained trend.

Independent sampling removes those chronological relationships. That can make the sample a poor description of the risk you care about. Blocks preserve some local order, but a fixed number of consecutive trades is not necessarily one session, one event, or one market regime.

Investigate event or session groups, overlapping exposure, and chronological performance alongside resampling. Treat grouping and block length as assumptions to examine, rather than a setting that automatically fixes dependence.

A good journal matters more than another thousand simulations

Keep a consistent definition of which executed trades qualify, and include their losses. Log skipped opportunities separately to examine selection decisions; do not add unexecuted hypothetical P&L to the observed closed-trade sample. Marking a zone before the event and selecting an attractive zone after seeing the outcome create different evidence. If discretion changes substantially, a combined history may mix several trading processes.

A small sample can reveal how sensitive its arithmetic is. It cannot supply an unobserved crisis or rescue an overfit setup. Running 10,000 paths from 30 trades creates more synthetic paths, not more observed market experience.

Likewise, a 95th-percentile drawdown is an upper-tail result under the chosen model, not a ceiling on future losses. A drawdown-threshold breach rate describes the sampled paths; zero breaches does not establish zero real-world risk. Neither a favorable tail nor stable parameters proves a repeatable edge.

Turn the exercise into a trading decision

Start with one question and define the boundary before looking at the results: which drawdown would require reducing exposure, stopping, or collecting more evidence? Record the trade sample, observation unit, starting capital, method, and sizing assumptions beside it.

Then compare what the sample says with your actual constraints. A concerning path can justify testing less exposure or reviewing clustered event risk. An acceptable path is one piece of evidence alongside chronological out-of-sample and forward evaluation.

For the supported controls and interpretation, use the Monte Carlo simulation guide; read its simulated results in the context of the assumptions above. The separate trade-order and drawdown walkthrough shows a recorded trade-list worksheet and distinguishes it from native product results.