Research · Dataset

Your grid backtest is a backtest of a monthly average

A grid backtest run on monthly averages is not a backtest of your grid. It is a backtest of a different strategy that only gets to act once a month. Run the identical simulator on daily benchmark fixes and the same configuration returns 0.38 percentage points more a year for the 2% grid and 1.10 points more for the 0.5% grid — while taking a deeper drawdown. This page measures the gap and separates it from the price-source choice, so you can tell which part of a backtest is real.

1968-04 to 2026-10 · 703 months · 14,716 daily bars in window · four grid configurations × three price paths

The number that matters

CAGR gap, 2% grid 0.38 pp daily path vs monthly backtest
Drawdown gap, 2% grid 5.15 pp daily path is deeper
Extra fills, 0.5% grid 4.62× 595 → 2,749
Price-source effect, 2% grid 0.031 pp control path vs Pink Sheet

The last tile is the one that makes the rest interpretable. Switching from one published gold price series to another moves the annualised return by 0.031 points. Switching from monthly sampling to daily sampling moves it by 0.42 points — roughly 14× as much. The frequency of the data, not the source of it, is what dominates a grid backtest.

Correction, 16 November 2026

This page previously stated that a cost scan from $0 to $20 per ounce found no crossover at which the daily path falls below the monthly path, in any of the four configurations, and that the daily path is always the higher-returning one. That was wrong. The crossover scan was measuring the daily 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 annualised return collapsed to nearly zero — which made the gap negative at every cost level and hid the sign change. The scan has been fixed and the dataset regenerated; the crossovers now appear in the cost table above and in the CSV. Two drawdown figures in the prose also quoted the control-versus-daily gap while the table beside them quoted the monthly-versus-daily gap, and one FAQ stated the return change with its sign inverted. All are corrected here. Every published metric other than the four crossover values is unchanged by the fix.

Source and method

Monthly price source
World Bank Commodity Price Data (The Pink Sheet), Monthly Prices sheet, column 70 (Gold)
Monthly source updated
November 03, 2026
Daily price source
LBMA PM gold price (1968-04 to 2026-02-25) and World Gold Council FSAPI (2026-02-26 onward)
Daily file
scripts/daily_gold_lbma_pm.csv (14,723 rows, 1968-04-01 to 2026-11-10)
Common sample
1968-04 to 2026-10 (703 months, 14,716 daily bars, 20.93 bars per month on average)
What is compared
The same grid simulator and the same four grid configurations are run on three price paths: the monthly average path, the daily path, and the monthly path with each month re-sampled from its own daily bars.
Why the paths differ
A grid level is filled when price touches the level and closed when price reaches entry*(1+spacing). On monthly bars both touches are detected only at the monthly close, so fills happen at the month close price. On daily bars the touch is found earlier in the path. The two paths are therefore not the same strategy with more or less detail - they are different strategies, and the difference is systematic rather than random.
Equity reporting
Equity is reported once per calendar month on every path, so the three paths are directly comparable.
Cost model
A flat round-turn cost of $0.30 per troy ounce is charged on every fill, matching the published multi-account dataset. No spread, slippage, financing, margin or tax is modelled.
What this is not
This is NOT a live account. There is no broker statement, no server-side execution, no requotes, no swaps and no intraday path inside a daily bar behind these numbers. It isolates the price-path granularity effect only.
Downloads
gold-ea-live-vs-backtest.csv (configuration, path, months, CAGR, max drawdown, Sharpe, fills, exposure)
Reproduction
scripts/gold_ea_live_vs_backtest_analysis.py; independently recomputed from the published CSV with a separate numpy implementation by scripts/verify_gold_ea_live_vs_backtest.py

Read this before quoting any number on the page

This is NOT a live account. There is no broker statement, no server-side execution, no requotes, no swaps and no intraday path inside a daily bar behind these numbers. It isolates the price-path granularity effect only.

The page title uses "live" in the sense of a live price path, not a live account. There is no broker statement, no server-side execution, no requotes and no swaps behind any figure here. What the page isolates is the price-path granularity effect: how much of a grid backtest is an artefact of the sampling frequency.

Three paths, one simulator

The same grid code, the same four configurations and the same cost assumption run on three price paths. The third path is the control: a monthly average recomputed from the daily series, which isolates the sampling frequency from the choice of price source.

Path What it is Fill opportunity What it measures
Monthly World Bank Pink Sheet monthly average one close per month the conventional backtest path
Monthly control monthly average recomputed from the daily series one close per month the same frequency, a different series
Daily LBMA PM benchmark fixes, daily 20.93 bars per month the price path a running EA actually sees

The control path is what makes the decomposition possible. Without it, a difference between two backtests cannot be attributed to the data frequency rather than to whichever price series happened to be used.

Every configuration, every path

Configuration Path CAGR Max DD Total multiple Sharpe (rf 0) Fills Mean exposure
1% spacing, 20 levels Monthly average 1.656% -49.68% 2.614× 0.220 279 59.5%
1% spacing, 20 levels Monthly control 1.645% -50.66% 2.598× 0.219 272 59.7%
1% spacing, 20 levels Daily fixes 2.373% -55.92% 3.943× 0.252 1,125 67.0%
2% spacing, 10 levels Monthly average 1.276% -36.26% 2.100× 0.229 129 39.6%
2% spacing, 10 levels Monthly control 1.245% -37.10% 2.063× 0.227 122 39.1%
2% spacing, 10 levels Daily fixes 1.661% -41.41% 2.621× 0.243 405 42.9%
3% spacing, 10 levels Monthly average 1.033% -34.71% 1.824× 0.201 94 37.2%
3% spacing, 10 levels Monthly control 1.015% -35.08% 1.806× 0.199 96 37.3%
3% spacing, 10 levels Daily fixes 1.506% -38.23% 2.397× 0.236 270 39.4%
0.5% spacing, 40 levels Monthly average 1.866% -57.49% 2.949× 0.216 595 73.4%
0.5% spacing, 40 levels Monthly control 1.843% -58.24% 2.910× 0.215 579 73.8%
0.5% spacing, 40 levels Daily fixes 2.965% -63.23% 5.525× 0.269 2,749 83.4%

Sharpe here is computed against a zero risk-free rate; the published multi-account dataset on this site uses 4%, so Sharpe figures are not comparable across the two pages. CAGR, maximum drawdown, total multiple, fill count and longest underwater period are directly comparable and are checked by the verification script.

Where the gap comes from

The difference decomposes into two independent parts, and the page reports both for each configuration.

Configuration Series effect (pp) Granularity effect (pp) Total CAGR gap (pp) CAGR ratio Fill ratio Exposure gap (pp)
1% spacing, 20 levels -0.011 0.728 0.717 1.433× 4.032× 7.482
2% spacing, 10 levels -0.031 0.415 0.385 1.302× 3.140× 3.274
3% spacing, 10 levels -0.017 0.490 0.473 1.458× 2.872× 2.146
0.5% spacing, 40 levels -0.023 1.122 1.100 1.589× 4.620× 10.007

The series effect is the difference between the Pink Sheet monthly path and the monthly control path — same frequency, different price series. It is small: 0.011, 0.031, 0.017 and 0.023 points for the four configurations. The granularity effect is the difference between the control path and the daily path — same series, different frequency. It is 0.73, 0.42, 0.49 and 1.12 points, in every case the larger of the two.

The fill ratio explains why. The tightest grid (0.5% × 40 levels) sees 4.62× as many fills on daily bars, the widest grid (3% × 10 levels) only 2.87×. A monthly bar collapses an entire month of price movement into one number, and a grid level that was touched and recovered inside that month leaves no trace at all.

Drawdown always gets worse

The extra fills the daily path finds are not free. Every configuration has a deeper maximum drawdown on daily data, by 6.24, 5.15, 3.52 and 5.74 points respectively.

Configuration Max DD, monthly Max DD, daily Gap Worst month, monthly Worst month, daily Longest underwater, daily
1% spacing, 20 levels -49.68% -55.92% 6.24 pts -12.36% -19.24% 332 mo
2% spacing, 10 levels -36.26% -41.41% 5.15 pts -8.93% -13.89% 332 mo
3% spacing, 10 levels -34.71% -38.23% 3.52 pts -8.47% -13.54% 316 mo
0.5% spacing, 40 levels -57.49% -63.23% 5.74 pts -14.19% -21.97% 332 mo

The direction is the point

A monthly grid backtest that reports a drawdown is reporting a floor. The same rule on daily data gives up 3.52 to 6.24 points of additional drawdown. Sizing a live grid off a monthly backtest's maximum drawdown means sizing it off a number that is too small by more than the spread between grid configurations.

Trading costs decide whether the daily path survives

The daily path trades far more often, so it pays costs far more often. The page re-runs each configuration across a range of round-turn costs from zero to 4 US dollars per troy ounce. At the zero-CAGR point the daily path is much more fragile than the monthly one.

Configuration Cost ($/oz) CAGR, monthly CAGR, daily End multiple, monthly End multiple, daily
1% spacing, 20 levels 0.00 1.737% 2.652% 2.739× 4.623×
1% spacing, 20 levels 0.15 1.697% 2.512% 2.676× 4.269×
1% spacing, 20 levels 0.30 1.656% 2.373% 2.614× 3.943×
1% spacing, 20 levels 0.60 1.575% 2.095% 2.495× 3.364×
1% spacing, 20 levels 1.00 1.467% 1.727% 2.344× 2.722×
1% spacing, 20 levels 1.50 1.332% 1.268% 2.169× 2.090×
1% spacing, 20 levels 2.50 1.061% 0.360% 1.854× 1.234×
1% spacing, 20 levels 4.00 0.655% -0.984% 1.465× 0.561×
2% spacing, 10 levels 0.00 1.327% 1.791% 2.162× 2.825×
2% spacing, 10 levels 0.15 1.301% 1.726% 2.130× 2.721×
2% spacing, 10 levels 0.30 1.276% 1.661% 2.100× 2.621×
2% spacing, 10 levels 0.60 1.225% 1.530% 2.039× 2.431×
2% spacing, 10 levels 1.00 1.158% 1.356% 1.961× 2.199×
2% spacing, 10 levels 1.50 1.073% 1.139% 1.867× 1.939×
2% spacing, 10 levels 2.50 0.903% 0.704% 1.692× 1.507×
2% spacing, 10 levels 4.00 0.647% 0.051% 1.458× 1.031×
3% spacing, 10 levels 0.00 1.066% 1.591% 1.860× 2.517×
3% spacing, 10 levels 0.15 1.049% 1.548% 1.842× 2.456×
3% spacing, 10 levels 0.30 1.033% 1.506% 1.824× 2.397×
3% spacing, 10 levels 0.60 0.999% 1.420% 1.789× 2.282×
3% spacing, 10 levels 1.00 0.954% 1.307% 1.743× 2.137×
3% spacing, 10 levels 1.50 0.898% 1.165% 1.687× 1.969×
3% spacing, 10 levels 2.50 0.785% 0.880% 1.580× 1.670×
3% spacing, 10 levels 4.00 0.615% 0.453% 1.431× 1.303×
0.5% spacing, 40 levels 0.00 1.979% 3.421% 3.147× 7.154×
0.5% spacing, 40 levels 0.15 1.922% 3.192% 3.046× 6.287×
0.5% spacing, 40 levels 0.30 1.866% 2.965% 2.949× 5.525×
0.5% spacing, 40 levels 0.60 1.752% 2.513% 2.762× 4.271×
0.5% spacing, 40 levels 1.00 1.601% 1.916% 2.532× 3.035×
0.5% spacing, 40 levels 1.50 1.412% 1.178% 2.271× 1.984×
0.5% spacing, 40 levels 2.50 1.036% -0.270% 1.828× 0.854×
0.5% spacing, 40 levels 4.00 0.474% -2.377% 1.319× 0.245×

The zero-CAGR break-even cost is where a grid stops making money altogether. For the 2% grid it is 7.73 dollars per ounce on the monthly path and 4.12 dollars on the daily path. For the tightest 0.5% grid it is only 5.27 and 2.31 dollars.

The crossover is the cost at which the monthly path stops being the conservative one. At the modelled 0.30 dollars per ounce the daily path is ahead in all four configurations — its extra fills paid off on this historical path — but each configuration has a cost level above which the monthly path overtakes it: 1.29 dollars per ounce for the tightest 0.5% grid, 1.40 for the 1% grid, 1.75 for the 2% grid and 3.05 for the 3% grid. The reason is mechanical: the daily path fills more often, so it is charged the round-turn cost more often. The break-even column above is the separate, larger question — where the strategy dies entirely.

Buy-and-hold barely moves

The control experiment for the whole page. Buy-and-hold never needs a fill signal, so the data frequency should not matter much to it — and it does not.

Measure Monthly path Daily path Difference
CAGR 8.333% 8.259% 0.074 pts
Max drawdown -62.07% -61.78% -0.29 pts
Annualised volatility 16.892% 19.410% 2.518 pts
Worst month -16.69% -22.37% -5.68 pts
Total multiple 108.03× 103.80× -4.23×

Buy-and-hold returns 8.333% a year while the best grid on the page returns 2.965% on the daily path. Every grid configuration on this page sacrifices the large majority of the return to cut the drawdown from -62.07% to somewhere between -34.71% and -63.23%. That trade is the reason grid results exist; the frequency finding does not change it.

Reconciling the buy-and-hold total multiple with the intra-month dataset page. This page's monthly path is the World Bank Pink Sheet monthly series; the intra-month dataset page computes its monthly path as the average of the daily benchmark fixes instead. Both cover the identical window (1968-04 to 2026-10, 703 months), so the 108.03× on this page and the 108.46× on that page are the same investment measured on two different price series, not two different time periods. The gap is 0.43×, which comes almost entirely from the first month of the sample: the Pink Sheet opens at $38.00/oz while the average of that month's daily fixes is $37.86/oz, a -0.36% difference that is amplified by 58 years of compounding. The daily path is the same on both pages — it comes from the same fix series collapsed to month-end — and both report it at 103.80×. Neither number is wrong; they answer the same question on two different price sources.

Cross-validation against the published dataset

This script's monthly path is the full Pink Sheet sample, and it must reproduce the figures already published on this site's multi-account dataset. If it does not, one of the two is wrong.

Figure Reproduced here Published on this site Match
Buy-and-hold CAGR 7.399% 7.399% yes
Buy-and-hold max DD -62.074% -62.074% yes
Buy-and-hold total multiple 117.29× 117.29× yes
Buy-and-hold longest underwater 315 mo 315 mo yes
G2 CAGR 1.117% 1.117% yes
G2 max DD -36.264% -36.264% yes
G2 fill count 129 129 yes

The full-sample rows of this script must reproduce the published multi-account dataset. A mismatch means one of the two is wrong. BH Sharpe is computed against a zero risk-free rate here while the published dataset uses 4%, so Sharpe is expected to differ and is not part of the equality check.

What this dataset supports, and what it does not

Supported: the choice of sampling frequency is a first-order driver of grid backtest results, and it dominates the choice of price series. On the same 703-month sample, monthly to daily sampling moves the annualised return of a 2% grid by 0.38 points and its maximum drawdown by 5.15 points, while changing the price source moves it by 0.031 points. It also shows that the direction of the frequency effect is systematic across all four configurations tested: more resolution means more fills, higher return on this path, and a deeper drawdown.

Not supported: any claim that a daily backtest is what you would have earned live. This page has no broker statement, no server-side execution, no requotes and no swaps. A benchmark fix is an auction price, not a bid/ask. The daily path is conservative on fills — a level fills at the first fix at or below it, never at an intraday low a real account might have caught — and optimistic on execution, since only a flat 0.30 dollars per ounce round-turn cost is modelled. And nothing here is a forecast: a single historical path is not a distribution, and the daily path's advantage on this path is not a guarantee about any other.

The practical takeaway: if you are comparing a grid backtest to live results, check what frequency the backtest ran on first. A monthly backtest is not a conservative version of a daily one — it is a different strategy, and the difference is larger than the gap between most grid configurations.

Frequently asked questions

How much does the price-path frequency change a grid backtest?

It changes it a lot, and always in the same direction. Running the identical grid simulator on monthly averages instead of daily benchmark fixes LOWERS the annualised return by 0.38 percentage points for the 2% grid and 1.1 points for the 0.5% grid, while deepening the maximum drawdown by 5.15 and 5.74 points respectively. The same pattern holds for all four configurations tested.

Why does daily data change the result at all? Is it just more accuracy?

It is a different strategy, not a better measurement of the same one. A grid level is filled when price touches it and closed when price reaches entry times one plus the spacing. On monthly bars both touches can only be detected at the monthly close, so a level fills at the month-close price or not at all. On daily bars the touch is found earlier in the path. That changes the fill price, the fill count, the average exposure and the sequence of equity events. The comparison isolates exactly this granularity effect; it does not claim either path is what a broker would have filled.

How many more trades does the daily path generate?

For the 2% grid, 129 fills on monthly bars against 405 on daily bars, a ratio of 3.14 times. For the 0.5% grid the ratio is 4.62 times (595 against 2,749). The 3% grid, with the widest spacing, has the smallest ratio at 2.87 times. Over the 703 months in the sample a monthly bar gives one fill opportunity per level per month; a daily bar gives 20.93 opportunities on average, so a monthly backtest structurally understates how often a tight grid would have traded.

Is the difference caused by the price series itself, or only by the sampling frequency?

Only by the sampling frequency. The page separates the two effects with a control path: a monthly average recomputed from the same daily series. Against the World Bank Pink Sheet monthly series, that control differs by only 0.031 percentage points of annualised return for the 2% grid and 0.023 points for the 0.5% grid. Moving from the control path to the daily path then accounts for 0.42 and 1.12 points respectively. The price source is essentially neutral; the frequency is what moves the numbers.

Does the daily path ever underperform the monthly backtest once trading costs are realistic?

Yes, at a cost level above the one modelled here, and the exact point depends on the configuration. At the modelled 0.3 US dollars per troy ounce round-turn cost the daily path is ahead in all four configurations. Scanning costs from zero to 20 US dollars per troy ounce, each configuration has a crossover where the monthly path overtakes the daily path: 1.29 dollars per ounce for the 0.5% grid, 1.4 for the 1% grid, 1.75 for the 2% grid and 3.05 for the 3% grid. The mechanism is mechanical rather than mysterious: the daily path fills more often, so every extra dollar of round-turn cost is charged more often to it. That is a statement about this historical path, not a general rule.

At what round-turn cost does each grid stop making money?

The break-even cost is where annualised return reaches zero. For the 2% grid it is 7.73 US dollars per ounce on the monthly path and 4.12 dollars on the daily path. For the tightest 0.5% grid it is only 5.27 and 2.31 dollars. The daily path is more fragile here because it trades roughly 4.6 times as often, so each extra dollar of cost is charged many more times. A backtest that undercounts fills also undercounts the cost of those fills.

How does buy-and-hold compare with the grids on the same data?

Buy-and-hold returns 8.333% a year on monthly averages and 8.259% on daily fixes, with maximum drawdowns of -62.07% and -61.78%. Every grid configuration on this page returns a small fraction of that, between 1.033% and 2.965% a year, while cutting the drawdown to between -34.71% and -63.23%. Switching to daily data barely moves buy-and-hold (0.074 points of annualised return and 0.289 points of maximum drawdown) because it never needs a fill signal. The grids are the ones whose backtest depends on the data frequency.

What are the limits of this analysis?

Three, and they matter. First, this is not a live account. There is no broker statement, no server-side execution, no requotes, no swaps and no intraday path inside a daily bar, so it measures the price-path granularity effect only and cannot be quoted as a live-versus-backtest result. Second, a daily bar is an auction price, not an executable bid/ask, so the daily path understates intraday excursion; no spread, slippage, financing or tax is modelled beyond a flat 0.3 US dollars per ounce round-turn grid cost. Third, the monthly path is a monthly average, which is not a price any account traded, and this is a single historical path, not a distribution.

Sources

  • World Bank Commodity Price Data (The Pink Sheet), Monthly Prices sheet, column 70 (Gold), updated November 03, 2026 — the monthly average price path and the full-sample cross-check.
  • LBMA Gold Price PM benchmark fixes and World Gold Council FSAPI, 1968-04-01 to 2026-11-10 (14,723 rows) — the daily price path, built by scripts/build_daily_gold.py.
  • Analysis script: scripts/gold_ea_live_vs_backtest_analysis.py. Independent recomputation from the published CSV with a separate numpy implementation: scripts/verify_gold_ea_live_vs_backtest.py (260 checks).
  • For the grid simulator's structure and its corrections, see Grid vs. trend following: the evidence and Multi-account capital management. For the frequency argument on drawdown measurement, see Gold intra-month drawdown.
  • A daily bar is a fix/auction price, not a tradable bid/ask, so it understates intraday excursion.
  • No spread or slippage is modelled, so all four configurations are optimistic in absolute terms.
  • The monthly path is a monthly AVERAGE, which is not a price any account traded.
  • Fills are modelled as if the whole order filled at the touched price; real fills can be partial.

Prices are nominal US dollars per troy ounce and include no carry, financing or storage cost. Nothing on this page is investment advice or a forecast.

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