Metrics

What is Monte Carlo Simulation?

Reviewed by Karthikeyan Anbarasan (Founder, TradeLyser) · Updated 19 Oct 2026 · Editorial policy

DefinitionA Monte Carlo simulation re-runs your trade results thousands of times, in random orders or as random draws, to show the range of drawdowns and outcomes the same edge could produce. 120 Nifty option trades whose worst drawdown was ₹40,000 in the order they happened can show a 1-in-20 drawdown of ₹80,000 once reshuffled (illustrative).
Illustrative simulation · 45% wins · wins 1.5× losses · 500 tradesRuin here = equity falls to half of starting capital<0.1%risk 1% a trade69 losses in a row0.5%risk 2% a trade35 losses in a row3.8%risk 3% a trade23 losses in a row18%risk 5% a trade14 losses in a row57%risk 10% a trade7 losses in a rowBottom row: losses in a row that halve the account · full Kelly for this edge ≈ 8.3%
Figure 1. Illustrative Monte Carlo, 2,00,000 runs per bar: 45% win rate, wins 1.5 times the amount risked, a fixed percentage of current equity risked per trade, 500 trades. The chance of the account falling to half its starting value at some point was under 0.1% at 1% risk, 0.5% at 2%, 3.8% at 3%, 18% at 5% and 57% at 10%.

How a Monte Carlo simulation works

  1. Collect the net result of every closed trade for one strategy, in R, after charges and slippage. Use a single strategy tag; mixing setups blurs the result.
  2. Choose the method: reshuffle the same trades into a new order, or draw trades at random with replacement. The cards below show what each one tests.
  3. Run it many times, typically 10,000 or more, and for each run record the maximum drawdown, the longest losing run and the final result.
  4. Sort each measure and read the median and the 95th percentile: the value that 95% of runs stayed within.
  5. Size so the 95th-percentile drawdown is one you could sit through without abandoning the plan.

Reshuffle

Same trades, new order. Total P&L never changes; drawdowns and losing runs do.
Tests: sequence risk

Resample (bootstrap)

Draw trades at random, with replacement, to build new histories. The mix of wins and losses changes too.
Tests: sequence risk and luck in the result

Worked example: 120 Nifty option trades

Illustrative: one setup with 120 closed trades, 48 winners of +2R and 72 losers of −1R, risking ₹5,000 a trade. The net result is +24R, or ₹1,20,000, an expectancy of 0.2R per trade. In the order the trades happened, the worst drawdown was 8R (₹40,000) and the longest losing run was 7 trades. The figures below come from 10,000 runs of each method.

Measure (reshuffled)As it happenedSimulated medianSimulated 95th percentile
Maximum drawdown8R, ₹40,00010R, ₹50,00016R, ₹80,000
Longest losing run7 trades8 trades12 trades
Median result, resampled
+₹1,20,000
5th-percentile result
−₹15,000
Runs ending below zero
About 5%
95th-percentile drawdown, resampled
₹1,00,000

Two things stand out. The real drawdown of ₹40,000 was milder than the median reshuffled one, so planning around it would leave you short: half of all orderings were worse. On a ₹5,00,000 account, the 95th-percentile drawdown of ₹80,000 is a 16% fall, and that is the number to check against your nerves and your limits. And about one in twenty resampled histories finished 120 trades below zero even though the edge is positive, which is a reminder of how much luck sits in a sample of this size. A losing quarter on its own does not prove the edge has gone.

Using the result

  • Set risk per trade from the drawdown you can tolerate. If ₹50,000 is your limit and the 95th-percentile drawdown is 16R, risk at most ₹3,125 a trade.
  • Expect losing runs up to the simulated 95th percentile, here 12 in a row, and write that into the plan so the run does not trigger rule changes.
  • Draw a "something is wrong" line just beyond the simulated range. A live drawdown past it is a reason to pause and review, not to add size.
  • Rerun the simulation each month with the new trades from your journal, using the same strategy tag. If the percentiles shift a lot from one month to the next, the sample is still too small to lean on.

What it cannot tell you

  • It assumes the future looks like the trades you fed in. A change in volatility or market structure is outside its view.
  • Shuffling treats trades as independent. If your losses cluster, for example on expiry days, a shuffle spreads them out and can understate drawdowns.
  • It cannot add risks that never appeared in the sample, such as a large gap or a liquidity freeze.
  • Biases in a backtest, such as look-ahead or missing charges, flow straight into the simulation.
For F&O traders in India
Older trades may come from different contract rules: NSE moved expiries from Thursday to Tuesday in September 2025, Bank Nifty weekly options ended in November 2024, and lot sizes are revised periodically. Re-net older results at current charges too, including STT of 0.15% of the sell-side premium on options from 1 April 2026, before feeding them in.

Reference guide

ContextValueReading
Size set so the 95th-percentile simulated drawdown is bearablePreparedA bad but normal run will not force you to abandon the strategy
Size set from the backtest drawdown as it happenedUnderpreparedThat is one ordering out of many; worse orderings are likely over time
Simulation on gross results or under 30 tradesFalse comfortUse net results after charges and slippage, and 50 to 100 trades or more

Questions

What is a Monte Carlo simulation in trading?

A way to see how different your results could have been by re-running your trade outcomes thousands of times in random orders or random draws. It shows the range of drawdowns, losing runs and final results the same edge could produce.

How many simulations should I run?

10,000 runs is a common starting point. The number is enough when two runs with different random seeds give nearly the same percentiles.

How many trades do I need for a Monte Carlo simulation?

At least 50, and 100 or more is better. With fewer trades, the simulation mostly reshuffles noise, and the percentiles will move a lot as new trades are added.

Is Monte Carlo simulation better than backtesting?

They do different jobs. A backtest produces one history of trades from your rules; Monte Carlo takes those trades and shows how much worse or better the same edge could plausibly have looked.

Can I run a Monte Carlo simulation in Excel?

Yes. Put your trade results in a column, draw random rows with RAND and INDEX to build new sequences, compute the running drawdown for each, and repeat with a data table. Dedicated tools or a short script are faster for 10,000 runs.

Related terms