why backtest differs from live trading

Why Your Grid Backtest Beats Your Live Account: Data Frequency, Not Strategy

Run the same grid on monthly averages and on daily gold fixes and the return changes by up to 1.10 percentage points a year, while the drawdown deepens by up to 6.24 points. The gap is the sampling frequency of your price data.

By Auriatech Research Desk · · Last updated 2026-11-16 · 3 min read
backtestinggrid tradinggoldmethodologydata quality Contains our own backtest data

There is a standard explanation for the gap between a backtest and a live account. The tester fills better than the broker. The spread is wider than you modelled. The slippage is worse. The news moved against you. These are all real and all worth handling.

But there is one cause that gets almost no attention, because it is not a broker problem at all: the frequency of your price data. Change only the interval you feed the tester and the same strategy, on the same market, over the same 58 years, produces a different result. Sometimes a better one.

We measured this directly. The full method and every number are published on our grid backtest versus live price path dataset page; this article explains what it means and what to do about it.

Correction, 16 November 2026. The first version of this article stated that a cost scan from $0 to $20 per ounce found no crossover where the daily path falls below the monthly path, and that the daily path is always the higher-returning one. That was wrong, and it was wrong in the direction that flatters the finer data. The crossover scan was comparing the daily path's equity curve before it was collapsed to month-end, so a 58.6-year sample of daily bars was being treated as a 1,226-year sample and the daily return was collapsed to nearly zero. The scan has been fixed and recomputed; every configuration has a crossover, and they are listed below. Two drawdown figures in this article also quoted the control-versus-daily gap while the tables quoted the monthly-versus-daily gap, and one FAQ paired the 3% grid's monthly worst month with the 0.5% grid's daily worst month. Those are corrected in place, and the numbers here now match the dataset page and its published CSV.

What we compared

Four gold grid configurations, each with a spacing between levels and a number of levels, run through one identical simulator on three price paths:

  • The monthly path — the World Bank Pink Sheet monthly average of the gold price. This is what most published grid backtests use.
  • The daily path — the LBMA Gold Price PM benchmark fix, 14,716 daily bars from 1968-04-01 to 2026-10-30, re-sampled to month-end for equity reporting.
  • The control path — each month recomputed as the average of its own daily bars. Same data source as the daily path, monthly granularity.

The control path is the part that makes the result interpretable. Without it, a monthly-versus-daily comparison confounds two different things: how often you sampled, and where the numbers came from. The control holds the source fixed and changes only the frequency, so the gap between control and daily is the frequency effect on its own.

703 months of history, four configurations, a flat $0.30 per troy ounce round-turn cost on every fill, no spread, no slippage, no financing, no tax.

Why the frequency changes anything at all

A grid does not trade on price. It trades on price touching a level.

Open a monthly bar and the only price it contains is a single number: the month’s average. If that average sits below a grid level, the grid looks filled. If it sits above, the grid looks unfilled. There is no way to know from a monthly bar whether price dipped to the level on the third of the month and came back, or never came close.

So on monthly data the grid is not trading the market. It is trading a monthly average, which is a number no account ever transacted at. The fill happens at the average price, the exit happens at another average price, and the equity curve is a curve through a series of averages.

The same logic applies inside the tester, at every bar interval. MetaQuotes’ own documentation for the MetaTrader 5 strategy tester is explicit that the three modelling modes — 1-minute OHLC, “Every tick”, and “Every tick based on real ticks” — produce materially different results for the same strategy, and that in real-tick mode the tester downloads the broker’s actual tick history while in “Every tick” mode it generates the intrabar path from bars. The documented rule of thumb is the one that matters here: the faster the test, the lower the simulation accuracy.

The frequency of your data is not a setting. It is a modelling decision about whether your strategy can see the thing it trades on. A grid that fills on a touch needs a bar that can contain a touch.

What the numbers say

The return moves, and it moves up

Config Spacing Levels Monthly CAGR Daily CAGR Change
G1 1.0% 20 1.656% 2.373% +0.717 pp
G2 2.0% 10 1.276% 1.661% +0.385 pp
G3 3.0% 10 1.033% 1.506% +0.473 pp
G4 0.5% 40 1.866% 2.965% +1.100 pp

All four configurations return more on the daily path. The pattern is systematic, not random, and the size of the move tracks the spacing: the tightest grid (G4, 0.5% spacing) moves most at +1.100 percentage points, and the widest grid (G3, 3.0% spacing) moves least at +0.473 points. A tight grid has more levels, so it has more opportunities for the frequency to change whether a level was touched.

The drawdown gets worse every time

Config Monthly max drawdown Daily max drawdown Change
G1 -49.68% -55.92% -6.24 pp
G2 -36.26% -41.41% -5.15 pp
G3 -34.71% -38.23% -3.52 pp
G4 -57.49% -63.23% -5.74 pp

Every configuration draws down more deeply on the daily path. This is the part that should make you stop and read: the finer data path is not simply a better result. It returns more and it is riskier, and a backtest that showed you only the return column would have hidden the second half.

The worst single month on G4 goes from -14.19% on the monthly path to -21.97% on the daily path, and the longest underwater stretch goes from 304-307 months across the four configurations to 332 months on three of them (G3 reaches 316). Average exposure rises by 2.1 to 10.0 percentage points: on the daily path the grid is holding more positions for more of the time, because it finds more fills.

The source of the data barely matters

This is the part that isolates the cause. If the difference were about which data vendor you used, the Pink Sheet versus LBMA comparison would carry it.

It does not. Against the monthly series, the control path differs by 0.011 to 0.031 percentage points of annualised return — smaller than the difference between any two configurations on the page. And the move from the control path to the daily path accounts for 0.42 to 1.12 points.

Config Control vs monthly Control vs daily (the frequency effect)
G1 -0.011 pp +0.728 pp
G2 -0.031 pp +0.415 pp
G3 -0.017 pp +0.490 pp
G4 -0.023 pp +1.122 pp

Switch data vendors and nothing happens. Change the sampling frequency and everything happens.

The trade count changes by a factor of nearly five

Config Monthly fills Daily fills Ratio
G1 279 1,125 4.03×
G2 129 405 3.14×
G3 94 270 2.87×
G4 595 2,749 4.62×

A monthly bar offers exactly one fill opportunity per level per month. A daily bar offers 20.93 on average across this sample. So a monthly backtest of a tight grid is not a conservative estimate of a live strategy — it is a structurally different strategy that happens to be less active.

This has a direct cost consequence, and it is the part most likely to bite a real account. The break-even cost per ounce, where annualised return reaches zero:

Config Monthly path Daily path
G1 $6.40 $2.90
G2 $7.73 $4.12
G3 $9.32 $5.59
G4 $5.27 $2.31

For the tightest grid, a strategy that looks able to absorb $5.27 per ounce of round-turn cost on monthly data can only absorb $2.31 once you count the trades it would actually have made. A backtest that undercounts fills also undercounts the bill for those fills, and it undercounts it by up to 4.6 times.

We also scanned round-turn costs from $0 to $20 per ounce for all four configurations. At the modelled $0.30 per ounce the daily path is ahead in every one of them. But each configuration has a crossover — a cost level above which the monthly path’s lower fill count wins:

Config Spacing × levels Crossover cost per ounce
G1 1.00% × 20 $1.40
G2 2.00% × 10 $1.75
G3 3.00% × 10 $3.05
G4 0.50% × 40 $1.29

The mechanism is mechanical, not mysterious. The daily path finds more touches, so it fills more often, so it is charged the round-turn cost more often. At $1.40 per ounce on the 1% grid, the monthly path’s advantage in fill count is worth more than the daily path’s advantage in fill timing. On this historical path the daily path’s extra fills paid off at $0.30; the crossover figures say how much room there is before they stop paying off. That is a statement about this sample, not a rule.

What this is, and what it is not

Be careful about the label. This is not a live-versus-backtest comparison. There is no broker statement behind any of these numbers, no server-side execution, no requotes, no swaps, and no intrabar path inside a daily bar. What we measured is the price-path granularity effect, in isolation, on a single historical path.

Three limits are worth stating plainly:

  1. A daily bar is an auction price, not an executable bid/ask. The daily path understates intraday excursion, and no spread or slippage is modelled beyond the flat $0.30 grid cost.
  2. The monthly path is a monthly average, which is not a price any account traded.
  3. This is one path through one market over 58 years. It is not a distribution, and one historical path does not tell you the frequency of the outcomes you saw in it.

What we can say is narrower and more useful: for a strategy whose logic depends on price touching a level, the granularity of the price data is a first-order modelling decision, and it is one that a lot of published grid backtests make implicitly and without measuring.

What to do about it

1. Check whether your strategy needs intrabar information at all

Not every strategy does. If your rules only look at a bar’s close and enter on the next bar, a monthly or daily bar is a fair representation of what you could have known. If your rules need to know whether a level was touched — which is every grid, every stop, every trailing stop, every partial close, every scale-in — then the bar interval is part of the strategy and you cannot treat it as a free parameter.

The diagnostic is cheap: run the identical strategy on two bar intervals and compare. If the results barely move, the strategy does not depend on granularity and you can stop worrying. If they move, you have found a real modelling assumption that was never stated.

2. Read the journal, not just the report

MetaTrader 5 reports a “History Quality” percentage, and it is easy to read that as a data quality score for the whole run. It is not. The documentation is clear that where a symbol’s history has a minute bar with no tick data, the tester generates ticks for that minute instead — so a run can report high History Quality while a substantial share of the period executed on generated ticks. The per-run, minute-level breakdown of which minutes used broker ticks and which used generated ticks appears in the tester journal, not in the report.

If your broker’s tick history does not extend back over your test period, part of your backtest is a simulation whether the report says so or not.

3. Treat the drawdown column as the headline, not the return column

Our own dataset page puts the return and the drawdown side by side for this reason. A daily path that returns more and draws down more is not a better strategy, it is a more active strategy. If you are choosing between two backtests of the same logic, the one with the deeper drawdown is telling you something true about the risk you would have carried. Do not let a higher return figure in the summary table stand in for that.

4. Keep the reproduction path open

Every number above comes from a script and a second, independent script that re-derives them. The full method, the source files, the month-by-month outputs and the verification harness are published on the dataset page. The value of a methodological claim is not that it sounds right. It is that you can run it and check whether it still says the same thing.

Frequently asked questions

What is the most common reason a backtest does not match live trading?

The price data is sampled too coarsely for the strategy's own logic. A grid strategy needs to know when price touched a level. On a monthly bar the touch is only visible at the month close, so the level either fills at the close price or never fills. On a daily bar the same touch is found earlier in the path, at a different price, and the whole equity sequence changes.

How much can the data frequency alone change a backtest result?

For the four gold grid configurations measured over 703 months from 1968-04 to 2026-10, moving from a monthly-average price path to a daily path changed the annualised return by 0.38 to 1.10 percentage points and deepened the maximum drawdown by 3.52 to 6.24 points. The tightest 0.5% grid moved the most: 1.87% to 2.97% a year, with the drawdown going from -57.49% to -63.23%.

Is it the price source or the sampling frequency that causes the difference?

The sampling frequency. A control path that recomputes each month as the average of that month's own daily bars sits within 0.031 percentage points of annualised return of the original monthly series, while the move from that control to the daily path accounts for 0.42 to 1.12 points. The source of the data is essentially neutral; the frequency is what moves the numbers.

Why does a finer data path make the drawdown worse even though it makes the return better?

Because the daily path finds more fills, and each fill is an additional position held through a period the monthly path never saw. Average exposure rises by 2.1 to 10.0 percentage points, the worst month goes from -8.47% to -13.54% on the 3% grid and from -14.19% to -21.97% on the 0.5% grid, and the longest underwater stretch goes from 304-307 months to 332 months on three of the four configurations.

Does a backtest that uses monthly data understate trading costs?

It understates them, and the size of the understatement scales with how often the strategy trades. The tightest grid makes 2,749 fills on the daily path against 595 on the monthly path, a ratio of 4.62 times. The cost per ounce at which annualised return reaches zero falls from $5.27 on the monthly path to $2.31 on the daily path. A backtest that undercounts fills also undercounts the bill for those fills.

What should I check before trusting a backtest?

Three things. First, what the bar interval is and whether the strategy's own logic needs intrabar detail: a grid needs the touch, not just the close. Second, read the tester's journal to confirm which ticks were real broker ticks and which were generated, because the report's History Quality figure does not tell you the share of the period that ran on generated data. Third, check whether the number of trades per year is plausible for the strategy, because a monthly bar can only produce one fill opportunity per level per month no matter how often the strategy would actually have traded.

Sources

  1. Auriatech Gold Desk — Gold grid backtest versus live price path, 1968-2026 (dataset and full method) — accessed 2026-11-13
  2. MetaQuotes — Testing trading strategies on real ticks — accessed 2026-11-13
  3. MetaTrader 5 Help — Real and generated ticks — accessed 2026-11-13
  4. MetaTrader 5 Help — How the tester downloads historical data — accessed 2026-11-13
  5. World Bank — Commodity Price Data (The Pink Sheet), Monthly Prices — accessed 2026-11-13
  6. LBMA — LBMA Gold Price PM benchmark data — accessed 2026-11-13

Use the free tools

Open-source calculators — no signup required.

Use the free tools
Risk disclosure — read before acting on anything above

All content on this site is provided for educational and informational purposes only. Nothing published here constitutes investment advice, a recommendation, or an offer to buy or sell any financial instrument. Trading foreign exchange, CFDs, and precious metals involves a high level of risk and may result in the loss of your entire capital. Past performance of any strategy, robot, or expert advisor (EA) — whether simulated, backtested, or live — is not indicative of future results. You are solely responsible for your trading decisions. If you are unsure whether a strategy is suitable for you, seek independent advice from a licensed financial professional in your jurisdiction.

Educational content only. Not investment advice. Trading forex and CFDs carries substantial risk of loss.