Research · Dataset

How bad was it really, before the month ended?

Almost every drawdown number ever published about gold is measured at a closing price. The month closes, the chart prints a low, and the drawdown number gets written down as if that low were the deepest point. It was not. In 88.6% of the 703 months in this sample, gold went deeper before the month-end close and recovered into it.

1968-04-01 to 2026-10-30 · 14,716 daily benchmark fixes · 703 months in the common sample

The number that matters

Months worse intra-month 88.6% 623 of 703 months
Mean gap 1.93 pp median 1.28 points
Worst drawdown, daily -70.26% vs -62.08% measured monthly
Volatility gap 2.55 pp 16.76% monthly → 19.31% daily

The direction is not in doubt. The magnitude depends on the strategy, and for every strategy on this page the daily drawdown is worse than the monthly one — by 8.2 points for buy-and-hold and 6.1 points for the tightest grid.

Source and method

Price source
LBMA Gold Price PM benchmark fix, USD/oz
Daily coverage
1968-04-01 to 2026-10-30 (14,716 fixes)
Monthly coverage
1968-04 to 2026-10 (703 months)
Common sample
Both the monthly and the daily run are restricted to the 703 months present in both series (1968-04 to 2026-10), so the two cover identical price history.
Monthly basis
Monthly averages of the daily benchmark fixes. The World Bank Pink Sheet monthly series is used to cross-validate these averages and to reproduce the site's published grid numbers.
Grid simulator
Static anchor per cycle (never trailing), each level takes profit at its own entry x (1 + spacing), a level is never filled twice. This is the same third-version simulator used across the existing grid pages.
Fill price
On daily data a level fills on the first benchmark fix at or below the level, and the fill happens at that fix price. The advantage of a real intraday low is not captured, which makes the daily result conservative rather than flattering.
Costs
$0.30 per troy ounce round-turn charged on every grid fill. No commission, financing, swap or taxes.
Annualisation
Monthly equity curves annualise with 12 periods per year; daily equity curves with 252. Volatility is the population standard deviation of periodic returns scaled by the square root of the period.
Risk-free rate
Sharpe ratio uses a 4% annual risk-free rate, matching the rest of the site.
Downloads
gold-intra-month-daily.csv (date, price, source, drawdown from peak)
gold-intra-month-strategies.csv (date, strategy, equity)
Reproduction
scripts/gold_intra_month_analysis.py; independently recomputed with a separate numpy implementation by scripts/verify_gold_intra_month.py

A benchmark fix is a daily auction price, not a tick-by-tick quote and not an executable bid/ask. It bounds the sequence of prices a trader could have transacted at, it does not reproduce one.

Where the drawdown actually was

For each of the 703 months, this page measures the worst drawdown reached at any point inside the month, and compares it with the drawdown implied by that month's closing price. The gap is the amount of pain that a monthly-close-based risk report simply does not show.

Depth threshold Months deeper intra-month Share Months deeper at month-end Share Understatement
≥ 5% 645 91.75% 530 75.39% 16.36 pts
≥ 10% 574 81.65% 473 67.28% 14.37 pts
≥ 15% 513 72.97% 431 61.31% 11.66 pts
≥ 20% 466 66.29% 400 56.90% 9.39 pts
≥ 25% 438 62.30% 375 53.34% 8.96 pts

The understatement widens with depth. At a 5% threshold the two measures disagree on 16.4 points of months; at 20% they disagree on 9.0 points. The drawdowns that matter most to a risk framework are the ones a closing-price report is most likely to understate.

The two worst months

Month Worst intra-month drawdown Month-end drawdown Gap Trading days
1999-07 -70.26% -69.93% 0.33 pts 22
1999-08 -70.25% -70.02% 0.23 pts 21

Both rows come from the same 1999 episode — the deepest gold drawdown in the series. The deepest episode ran from 1980-01-21 to 1999-07-20, -70.26% peak to trough over 4,891 days. The gap between the two worst months is small, which tells you the extremes were the same episode seen from two ends rather than two separate events.

Every strategy, both frequencies

Strategy CAGR, monthly CAGR, daily Max DD, monthly Max DD, daily DD understatement Sharpe, daily
Buy and hold 8.328% 8.274% -62.08% -70.26% 8.18 pts 0.221
12-month time-series momentum 9.123% 8.574% -33.47% -45.43% 11.96 pts 0.280
Grid G2 (2% spacing, 10 levels) 1.244% 1.664% -37.10% -41.59% 4.49 pts -0.264
Grid G1 (1% spacing, 20 levels) 1.643% 2.377% -50.67% -56.79% 6.13 pts -0.121
Grid G3 (3% spacing, 10 levels) 1.014% 1.508% -35.08% -40.08% 5.00 pts -0.305

Annualised returns, maximum drawdown and Sharpe for the same 1968-04–2026-10 window, computed once on monthly averages (12 periods per year) and once on daily benchmark fixes (252 periods per year). Buy-and-hold total return is 108.46× monthly and 103.80× daily. Sharpe uses a 4% annual risk-free rate.

Why the live-versus-backtest dataset page quotes 108.03× instead. Its monthly path is the World Bank Pink Sheet monthly series; this page's monthly path is the average of the daily benchmark fixes. Both cover the identical 1968-04–2026-10 window (703 months), so the two totals measure the same investment on two different price series rather than over two different periods. The gap of 0.43× comes almost entirely from the first month: 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 amplified by 58 years of compounding. The daily total is identical on both pages (103.80×) because both derive it from the same fix series collapsed to month-end.

What does not change

The ranking. Time-series momentum is the only strategy here with a positive daily Sharpe (0.280) and the smallest drawdown among the long-only rules (-45.43% daily). Buy-and-hold returns more (8.274% daily) and takes a far deeper drawdown (-70.26%). The grids return 1.664% and 2.377% a year with drawdowns of -41.59% and -56.79%. Switching from monthly to daily data changes the magnitudes, not the conclusions.

Volatility, decade by decade

The frequency gap is not a constant offset. It varies by year and by decade, which is why a single volatility assumption cannot be applied across a full cycle.

Decade G2 CAGR, monthly G2 CAGR, daily G2 max DD, monthly G2 max DD, daily
1970-1979 2.956% 4.338% -20.78% -26.08%
1980-1989 -1.200% -3.304% -32.32% -38.75%
1990-1999 -0.896% -0.446% -19.70% -20.77%
2000-2009 1.600% 4.271% -2.84% -12.00%
2010-2019 0.086% 0.573% -20.37% -24.13%

Grid G2 returns and drawdowns per decade, monthly basis versus daily basis. The 1980s are the only decade in which both bases are negative — the decade gold fell, and no sampling frequency makes a long grid profitable in it. This table is recomputed independently by the verification script, which caught an earlier version of it that had reused the full-sample daily equity curve for every decade.

Volatility by year

Annualised volatility computed from monthly averages versus from daily fixes, over 59 years. Daily data gave the wider reading in 56 of them (94.9%).

Year Monthly basis Daily realised Gap
2013 9.794% 21.715% 11.921 pts
1981 17.200% 27.932% 10.731 pts
2008 20.880% 31.587% 10.706 pts
1975 11.459% 20.273% 8.813 pts
2020 11.269% 20.002% 8.733 pts

The five widest gaps. The largest was 2013, where monthly averaging understated realised volatility by 11.921 points. Position sizing that assumes the monthly figure is therefore most likely to be wrong precisely in the years where the realised path was most violent.

Cross-validation

The monthly series is checked against the World Bank Pink Sheet over 703 months (1968-04 to 2026-10). Mean absolute relative difference 0.1514%, median 0.0649%, with 663 months within 0.5% and 685 within 1%.

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
G2 CAGR, no cost 1.162% 1.162% yes
G2 max DD, no cost -36.234% -36.234% yes
G2 CAGR, with spread 1.117% 1.117% yes

The same simulator run on the published Pink Sheet sample reproduces the figures already published on this site, which is what ties this page to the rest of the research. The grid simulator previously credited a closed level’s proceeds to cash while leaving the same quantity in the position, double-counting sold lots. Grid CAGR was inflated by roughly 10x. Figures below are the corrected published values.

What this dataset supports, and what it does not

Supported: monthly-close drawdown figures understate the pain in the large majority of months, by a median of 1.28 points and worse as drawdowns deepen. Any risk report built on closing prices alone is a lower bound on the drawdown actually experienced. And the same understatement applies across all five strategies tested, including the grid rules, because it is a property of the measurement frequency rather than of any one strategy.

Not supported: any claim that the daily numbers here are what you would have earned. A benchmark fix is an auction price, not a fill. The daily grid result is conservative on fills — a level fills at the first fix at or below it, never at an intraday low you would have actually caught — but it is optimistic on execution, since no bid/ask spread, commission, financing or swap is modelled beyond the flat $0.30 per ounce round-turn grid cost. And nothing here is a forecast: the current drawdown at the end of the sample is -27.60%, and a single historical path is not a distribution.

The practical takeaway is narrow but actionable: if your gold position size is derived from a monthly closing price, it is derived from a number that was too flattering in 88.6% of the months in the last 59 years.

Frequently asked questions

How much deeper does a drawdown get inside a month than at the month-end price?

Across 703 months, the worst intra-month drawdown is deeper than the drawdown measured at the month-end price in 623 of them, which is 88.6% of the sample. The mean gap is 1.93 percentage points and the median gap is 1.28 points. This is the single most practically important number on the page: a position sized on a monthly closing price can experience a drawdown materially larger than the one the closing price reported.

What was the worst single month for drawdown, and was it worse intra-month or at month-end?

The worst intra-month drawdown was in 1999-07, at -70.26% against -69.93% at that month's close — a gap of 0.33 points over 22 trading days. The worst month-end drawdown was in 1999-08, at -70.02% at the close against -70.25% intra-month. Note that both sit inside the same 1999 episode, which is the deepest gold drawdown in the series.

How large is the drawdown understatement at different depth thresholds?

At a 5% threshold, 645 months (91.75%) went deeper than 5% intra-month, against 530 months (75.39%) at the month-end price. At 10% the split is 81.65% against 67.28%. At 20% it is 66.29% against 56.9%. The understatement is systematic and widens with depth, which is exactly where it matters.

How much does daily data change the volatility estimate?

Annualised volatility is 16.759% when computed from monthly averages and 19.307% from daily benchmark fixes — a gap of 2.548 percentage points, or 15.2% of the monthly figure. Daily data produced the wider estimate in 56 of 59 years (94.9%). The largest single-year gap was 2013, at 11.921 points.

Does the choice of data frequency change which strategy looks best?

It changes the numbers but not the ranking among the strategies that matter. Buy-and-hold returns 8.328% a year on monthly data and 8.274% on daily data. 12-month time-series momentum returns 9.123% monthly and 8.574% daily, with drawdowns of -33.47% and -45.43%. The grids return 1.244% and 1.664% (G2), and 1.643% and 2.377% (G1). Daily data makes the grids look slightly better on return and slightly worse on drawdown, because it resolves the ladder fills that monthly averages hide.

How much worse is the drawdown when measured on daily data?

Buy-and-hold's maximum drawdown is -62.08% measured monthly and -70.26% measured daily — a difference of 8.18 points. Momentum's is -33.47% against -45.43%. G2 is -37.1% against -41.59%, and G1 is -50.66% against -56.79%. Every strategy on this page has a materially worse drawdown when you look at it daily.

How reliable is the monthly price series?

The monthly series is cross-validated against the World Bank Pink Sheet over 703 months (1968-04 to 2026-10). The mean absolute relative difference is 0.1514% and the median is 0.0649%, with 663 months within 0.5% and 685 within 1%. The same simulator run on the published Pink Sheet sample reproduces the site's existing grid figures exactly (7.399% buy-and-hold CAGR, 1.162% for G2 before costs), which is the check that ties this page to the rest of the site.

What are the limits of this analysis?

Three, and they are real. First, a benchmark fix is a daily auction price, not an executable bid/ask — it bounds the sequence of prices a trader could have transacted at but does not reproduce one, and no spread, commission, financing or swap is modelled beyond the flat $0.30 per ounce round-turn grid cost. Second, the grid simulator uses a static anchor per cycle and a level is never filled twice, which is a specific model of a specific grid, not a claim about grid trading in general. Third, this is a single historical path; a drawdown of -70.26% happened once and would not necessarily happen again in the same shape. The current drawdown at the end of the sample is -27.6%.

Sources

  • LBMA Gold Price PM benchmark fixes, 1968-04-01 to 2026-10-30 — 14,716 daily observations, built by scripts/build_daily_gold.py. scripts/build_daily_gold.py splices two independently sourced feeds and verifies the overlap (73 rows, 0 mismatches > $0.005) before merging.
  • World Bank Pink Sheet monthly gold series, updated November 03, 2026 — used to cross-validate the monthly averages and to reproduce this site's existing grid figures.
  • Analysis script: scripts/gold_intra_month_analysis.py. Independent recomputation from the published CSVs with a separate numpy implementation: scripts/verify_gold_intra_month.py (142 checks).
  • For the grid simulator's structure and its corrections, see Grid vs. trend following: the evidence and Multi-account capital management. For the volatility-targeting comparison, see Volatility targeting and the truth about drawdown.
  • For the same calendar question asked at daily resolution instead of monthly — whether gold has a day of the week — see Gold day of week.

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.