How Do You Compare a Live Trading Account to Its Backtest?

You compare a live trading account to its backtest on the things the backtest actually predicts, not on the equity curve alone: how deep a bad stretch goes, how often the system trades in each kind of market, and how hard the guards are working. Every gap you find is a question with a test attached, not a verdict. Six weeks of live trades can raise a question. They rarely answer one.

MasterVP · Systematic Trading Journey

Written by the founder of KenKem, a software engineer of twenty years who traded gold by hand before building the research stack behind MasterVP. This is knowledge sharing from my own build log, not a recommendation. Backtest figures come from my release 1.26 write up of 15 September 2026, on real tick data with all costs charged. Live figures come from my public myfxbook account and are dated where they appear.

Why does the first bad stretch live feel like proof the backtest lied?

Because a backtest is a number you read once, and a live drawdown is something you feel every day.

My two gold systems have run on real money since early September 2026. The backtest from January 2024 to May 2026 had a worst drop of 11.2 percent. On 14 October the live account showed 218 trades and a maximum drawdown of 10.45 percent. On 15 October it showed 225 trades and 13.12 percent, about 13 percent below its 28 September peak. So the live drawdown is now deeper than anything that backtest printed.

When I traded by hand, a moment like this is where I would start changing things. Tighten the stop, skip the next session, add size to win it back. Every one of those feels like taking control. Every one of them is a new strategy nobody tested, launched at the exact moment I am least able to judge it.

What does a backtest's worst drawdown actually tell you?

It tells you the worst stretch on one historical path, not a ceiling for the future.

A backtest replays one sequence of trades. Its maximum drawdown is the deepest hole in that single sequence. A different order of the same kind of trades would produce a different worst hole, sometimes deeper. So passing the backtest's worst is a reason to look closely, not proof the system is broken.

The honest reading has two halves, and 225 trades cannot separate them yet. Either a normal bad stretch arrived early, or live is running hotter than the test. I publish both readings because picking the comfortable one is how people talk themselves out of a real problem.

What do I check instead of guessing?

The parts of the system that should behave the same way live as they did in the test.

The first is the daily loss guard. It stops trading for the rest of the day once the day's losses hit a 5 percent limit. In the 29-month backtest it paused trading on 3 days. Live, the worst single day so far used about a quarter of that limit. So the drawdown is a slow accumulation of ordinary losing trades, not one day where something broke. That is a different problem from a blown-up day, with a different set of suspects.

The second is per-trade risk, which is fixed by the settings, not by how I feel that week. If losses were arriving at the planned size, the drawdown is a sequence problem. If they were bigger, it is an execution or sizing problem. Those two lead to completely different fixes, so I check which one it is before touching anything.

Why does my live account trade more often than the backtest?

Because trade count follows the market, and the market right now is busier than the average of the test.

The backtest took 2,676 trades in 29 months, roughly 4 a trading day. Live has been closer to 7 a day. My first thought was a bug. Then I checked what the backtest itself says about frequency. On quiet gold from 2021 to 2023, the five-minute system took 89 trades in three years. On busier gold from 2024 to 2026, it took 1,643 in two and a half.

That follows from how the system is built. Across 5.4 years, the average result per trade was about minus 0.19 R on days when gold moved less than $15, and about plus 0.21 R on days over $60, with break-even around a $20 to $22 range. R is the amount risked, so position size cant flatter it. The slope holds in every year from 2021 to 2026. A breakout method has a habitat, and when gold moves more, it finds more trades. More of them in a busy market is expected. More of them on quiet days would be the real red flag, because it would mean the stand-aside rules dont fire live the way they did in the test. That is the next check I am running, day by day.

Why not take more trades when a signal is good?

Because a thin edge copied five times multiplies the costs, not the edge.

Wanting to win back a drawdown faster usually ends in this idea. I had already tested it, on gold M5 with real ticks and every cost charged: ten stacking variants, from stacks of 5 and 20 to pyramiding and a martingale, each against one position at a time. The pass mark was written first: beat the single position on result and on recovery, or the idea is dead.

None cleared even one of those bars. The best kept about a quarter of the single position's result, and the martingale finished below zero. The stacks drew down less only because each position was smaller, and thats something you get for free by risking less.

How do I judge a fix before I ship it?

On the trades the fix actually changes, not on the whole book.

One of my own ideas taught me this. On a second or third breakout attempt, I wanted to anchor the stop behind a fair value gap on the 10-minute chart instead of a fixed volatility distance, so good trades could breathe. Pass mark first, then a paired test: same entries, same window, only the stop changed. Judged on the trades the new rule actually moved, it lost in every variant. The version closest to my idea gave up about 4.9 R across the 68 trades it changed. One whole-book total looked mildly positive, but that was noise. A different stop changes which later trades the system is free to take, and those knock-on trades have nothing to do with the idea.

What makes a change worth keeping?

It barely matters in good markets, matters a lot in bad ones, and survives a test designed to embarrass it.

The volatility floor on my ten-minute system is the example. In MetaTrader 5 on quiet 2021 to 2023 gold, its worst drop was 24.9 percent without the floor and 14.1 percent with it. On 2024 to 2026 the floor changed essentially nothing. But I picked the floor while studying 2021 to 2023, so I dont count that improvement as evidence. What counts: against random removals of the same number of trades, p = 0.001. Nearby settings all work, a plateau rather than a spike. And a blind re-tuner, re-fitting every six months on past data only, chose a non-zero floor in all nine periods.

That is the bar a change has to clear before it goes near a live account. Six weeks of drawdown does not lower it.

Frequently asked questions

Is it normal for a live account to differ from its backtest? Yes, some difference is expected. Real spreads, fills and the order of trades all vary. The useful question is whether the difference shows up in the things the backtest predicts, such as trade frequency in each market condition and the size of each loss.

What should I compare between a backtest and a live account? Compare drawdown depth, trade frequency by market condition, loss size per trade, and how often each guard fires. On my systems the backtest's daily loss guard paused trading on 3 days in 29 months, which gives the live account a concrete number to be measured against.

Does a live drawdown deeper than the backtest mean the strategy is broken? Not on its own. A backtest's worst drawdown comes from one historical sequence, and a different sequence can go deeper. My live account passed its 11.2 percent backtest worst at 13.12 percent on 15 October 2026, after 225 trades, which is too few to separate bad luck from a real gap.

How many live trades do you need before comparing to a backtest? More than most people want to wait for. A few hundred trades can flag a question about frequency or execution, but drawing conclusions about the edge needs months. My live account had 225 trades after about six weeks.

Why does a strategy trade more in some months than others? Because many systems only act when the market gives them room. My five-minute gold system took 89 trades on quiet gold from 2021 to 2023 and 1,643 on busier gold from 2024 to 2026, so frequency rising with gold's daily range is the designed behaviour.

Does adding more positions on a good signal improve results? In my testing, no. Ten stacking variants, including pyramiding and a martingale, all lost to one position at a time on both result and recovery. The best kept about a quarter of the single position's result.

How do you test whether a rule change actually helps? Run a paired test on the trades the change touches: same entries, same window, only the rule changed. My fair value gap stop looked mildly positive on the whole book but lost about 4.9 R across the 68 trades it actually changed.

If you want to see where this work ended up, MasterVP is my own intraday method built on public volume profile theory. It ships as the Master Volume Profiler indicator on TradingView and MetaTrader 5, which shows the read and leaves the order with you, and as Master Volume Sniper on MQL5 market, an Expert Advisor that executes the rules and risk settings you configure. The live account stays public through the bad weeks, this one included. Which check would you run first?

---

This article was composed from the Reject Pile and Backtest vs Live posts of the MasterVP build log series. Educational purpose only. Not financial advice. Backtest figures are research results on real tick data, live figures are from a public account over a short period, and past performance does not guarantee future results. Every trader remains responsible for their own risk settings and decisions.