Is your EA broken, or is it a normal bad run? The backtest as a range of futures (Monte Carlo cone)

Your EA has been down six weeks in a row. The five-year backtest had a beautiful curve and now the live account is sliding. Do you switch it off? Wait? For how long?

It's the hardest question for any EA trader, and it's usually answered with your gut: you switch off just before the system recovers, or you keep running one that stopped working months ago. This guide explains a more objective way to look at it: stop seeing the backtest as a single curve and start seeing it as a range of possible futures.

The problem: one curve can't tell you what's normal

A backtest report gives you one equity curve. But that curve is just one possible order of the trades the EA made. Had the same trades come in a different order, the streaks would have landed at other times, the max drawdown would be different, and the curve would have another shape.

When you compare your live account against that single curve, you're comparing against a reference with no margin: any week below it "looks" like a failure. What you need to know isn't whether you're below the curve, but whether you're inside what the backtest made possible.

The idea: resample the backtest thousands of times

Monte Carlo applied to a backtest works, in essence, like this:

1. Take the backtest's trades.
2. Build one possible future by drawing trades at random from that history, in a different order, until you cover as many weeks as the EA has been live.
3. Repeat step 2 thousands of times.
4. For each week, look at where most of those simulated futures ended up.

The result is a cone: a band that widens over time and says "with this backtest, normal was somewhere between here and here". On top of it you draw the EA's live curve.

Two method details make a big difference:

• Resample in blocks, not trade by trade. Drawing single trades breaks up streaks: in reality, losses tend to cluster when the market doesn't suit the system. Drawing blocks of several weeks keeps those streaks and gives a more honest band (wider on the downside).
• Use the live pace. If the EA trades less live than in the backtest, the cone should be built with the live number of trades, not the backtest's.

How to read the cone

• Inside the band: what's happening is consistent with the backtest. A bad run inside the cone is exactly that: something the backtest itself contained. Switching off here is usually reacting to noise.
• Below the band: few simulated futures ended up this badly. It doesn't prove the EA is broken, but something is outside what was expected and deserves a look: has the market changed? Is the broker filling worse? Are costs higher than in the backtest? Does the EA have the same settings?
• The percentile sums up where the live curve sits: at the 8th percentile, 8% of simulations ended worse than the live EA. The lower it is, the rarer the result according to the backtest.

A common band is the 90% band: the space between the worst 5% and the best 5% of futures. By construction, even a healthy EA should drop below it around 5% of the time. One week outside isn't a verdict; staying outside, or falling far outside, is a signal.

Hypothetical example: an EA has been live for 14 weeks. The last six were negative, but the curve is still inside the band, at the 30th percentile. According to its own backtest, that run is normal. Another EA with a similar run falls below the band in week 9 and never gets back in: that's the one to review first.

What the cone can't do

The cone is only as good as the backtest behind it. Be clear about its limits:

• An over-optimised backtest gives a false cone. If the backtest is curve-fitted to the past, the cone will be too optimistic and the live EA will fall out of it quickly. In that case the cone isn't wrong: it's telling you the backtest wasn't reliable.
• Costs and execution. If the backtest doesn't include realistic commissions, swaps and slippage, the comparison starts with the backtest at an advantage.
• Short samples. With few live trades, any conclusion is fragile. Better to wait for a reasonable minimum before giving a firm verdict.
• The past doesn't repeat. The cone describes what was normal over the backtest period. A regime change in the market can leave it behind.

And one practical recommendation: decide in advance what you'll do. For example: "if it leaves the band and stays out for four weeks, I halve the risk and review". A rule written with a cool head beats any decision made in the middle of a losing run.

How TrendPax does it

In TRACKING, each EA can have its backtest: you upload the MetaTrader report and the EA's page shows its Monte Carlo cone with the live curve on top, the percentile and a verdict ("On track" or "Review"). The method resamples 4-week blocks of the backtest at the EA's live trading pace, and the screen explains how it's calibrated. The verdict is provisional until the EA has 30 live trades, and the backtest has to cover at least 3 years and 100 trades for the comparison to make sense.

You can see it with made-up data in the demo, no sign-up needed: one EA that leaves the cone (EA 1002) and one still inside it (EA 1003). If it's useful to you, connecting your accounts is free.

Risk warning. This article is educational and is not investment advice. Monte Carlo analysis is based on past data and statistical assumptions: it does not predict future results or guarantee that an EA will recover. Trading leveraged products carries a high risk of loss.