gold grid vs trend following

Grid vs Trend-Following on Gold: 66 Years of Monthly Data, Reconstructed From the World Bank Pink Sheet

We rebuilt both a gold grid ladder and a 12-month trend-following rule on 802 monthly gold closes from 1960 to 2026, then compared drawdown, recovery time, and the decade where each one failed. Includes the spread-cost version and the ladder geometry that decides whether you survive.

By Auriatech Research Desk · · Last updated 2026-11-12 · 3 min read
grid tradingtrend followingbacktestrisk managementgold Contains our own backtest data
What this is. A reconstruction, not a broker screenshot. Both strategies were re-implemented from a published rule and run over 802 monthly gold closes (January 1960 – October 2026) taken from the World Bank's Pink Sheet commodity price data, last updated November 3, 2026. The rules, the data file, and every number in the tables are specified below so you can rebuild and disagree with it.

Correction, 12 November 2026. The first version of this article reported grid returns around 8.6–9.4% a year. That was wrong: the simulation credited a closed grid level’s proceeds to cash and left the same quantity in the position, so sold lots were counted twice. The simulator has been fixed, every number below recomputed, and the conclusion reversed — the grid does not beat buy and hold on this data, it lags it by a wide margin. The bug and its detection are documented at the end of this article, because a published backtest that was wrong in the buyer’s favour is exactly the kind of thing this site exists to catch.

There is a well-worn argument in gold trading circles. Grid bots catch the range. Trend systems catch the runaway. Each side cites its own backtest. Neither side ever puts both systems on the same data, over the same period, with the same assumptions about fills and costs.

This article does that. It is also deliberately unflattering to both.

The data

Item Value
Series World Bank Pink Sheet, “Gold” monthly average, USD per troy ounce
Frequency Monthly (monthly average price)
Observations 802 months, 1960-01 through 2026-10
First observation $35.00/oz (January 1960)
Last observation $4,105.00/oz (October 2026)
Highest monthly average in the sample $5,020.00 (February 2026)
Source World Bank Commodity Markets — Pink Sheet, monthly file last updated 2026-11-03

Two limitations to state up front, because they change how much weight the numbers can carry:

  1. These are monthly averages, not tick data. Every fill in this analysis happens at a monthly close price. A real ladder fills on intrabar touches and a real trend system exits on intrabar breaks, so live results will differ in ways that are not merely noise.
  2. This is spot gold in USD, not a broker’s XAUUSD contract. There is no swap, no commission, and no broker-specific spread in the base results. A cost model is added below, but it is a model.

What monthly data does give you is a clean, long, auditable sample that includes the only period that really matters for this question: the 1980s.

Strategy A — the grid ladder

The rule, stated completely:

  • At the start of a cycle, the anchor P is set to the current price.
  • There are L levels. Level k is a buy limit at P × (1 − k·s), where s is the spacing.
  • Each level, when filled, is funded with u (a fraction) of the current account equity — so level size shrinks as equity shrinks.
  • Each filled level takes profit at its own entry price × (1 + s).
  • When the ladder is fully closed and the price is at or above P × (1 + s), the cycle resets with a new anchor.
  • No stop loss. That is what makes it a grid and not a martingale with a stop.

Parameter sets tested, in addition to buy and hold:

Grid Spacing s Levels L Ladder depth Unit size u
G1 1% 20 20% 10% of equity
G2 2% 10 20% 10% of equity
G3 3% 10 30% 10% of equity

“Ladder depth” is s × L — the total decline from the anchor at which the entire ladder is filled and the account is at maximum exposure.

Strategy B — trend following

Also stated completely:

  • Compute the trailing 12-month price change.
  • If it is positive, hold one unit of gold. If negative, hold cash (earning 0%, not a risk-free rate).
  • Signal is computed from data through month t−1 and applied to month t’s return, so there is no look-ahead.

The 12-month lookback is the standard in the literature: Moskowitz, Ooi and Pedersen’s Time Series Momentum documents the effect across 58 futures instruments, and Hurst, Ooi and Pedersen’s century-long study uses the same family of rules. Gold was not in the 58-instrument sample, so the gold numbers below are our own reconstruction, not a citation.

Headline result, 1960-01 to 2026-10

All figures computed from the monthly series. Sharpe uses a 4% risk-free rate. “Max underwater” is the longest stretch of consecutive months the equity curve spent below a previous peak.

Strategy CAGR Ann. vol Max drawdown Sharpe Win month Max underwater Final equity
Buy and hold 7.40% 15.83% −62.07% 0.21 43.4% 315 months 117.3×
Trend-follow 12m 8.19% 13.86% −31.77% 0.30 30.2% 191 months 191.0×
Grid G2 (2%, 10 levels) 1.16% 6.01% −36.23% −0.47 35.8% 306 months 2.16×
Grid G1 (1%, 20 levels) 1.52% 8.91% −49.65% −0.28 39.1% 306 months 2.74×

Read the table carefully, because the ranking it produces is the opposite of the folk wisdom in both directions:

  • Trend following wins on drawdown by a factor of two. −31.8% against −62.1% for buy and hold and −36.2% for the best grid. Its maximum underwater period was 191 months; the grid’s was 306.
  • The grid loses badly on total return. G1 finished at 2.74× against buy and hold’s 117.3× and trend’s 191.0×. Over 66 years a 1.52% annual return compounds to roughly a fortieth of what simply owning gold did.
  • The grid’s lower volatility is not a benefit here. G2’s annualised volatility (6.01%) is less than half buy and hold’s (15.83%), and its Sharpe ratio is negative (−0.47). The equity curve was smoother only because the strategy was smaller and mostly out of the market.

That last point deserves emphasis, because it is the single most common misunderstanding about grid systems: a grid does not reduce drawdown per unit of capital at work. It reduces drawdown because it holds far less exposure than you think it does. Once you account for the capital deployed, the grid is not more defensive than owning gold — it is less exposed and much less profitable.

The decade that decides the argument

Full-sample numbers hide regimes. Split the same series into three blocks:

Period Buy & hold Trend 12m Grid G2 Grid G1
1960–1979 CAGR 13.74% 14.04% 1.73% 2.38%
1960–1979 max DD −40.22% −20.83% −20.68% −29.49%
1980–1999 CAGR −4.27% −0.20% −1.48% −2.37%
1980–1999 max DD −62.07% −23.85% −36.23% −49.65%
2000–2026 CAGR 10.50% 8.90% 2.74% 3.86%
2000–2026 max DD −39.28% −22.50% −20.10% −29.29%

What happened in the 1980s

Gold peaked at $675/oz in January 1980 and did not trade back to $675 until May 2006 — 26.3 years later. Over that stretch, buy and hold lost money in nominal terms (−4.27% a year). The grid lost less (−1.48%) but still lost money, and it took 306 months of underwater equity to get there.

The trend-following rule lost almost nothing (−0.20% a year, −23.85% max drawdown) — because for most of those 26 years it was simply not long gold. It sat in cash for 96 consecutive months at its longest stretch, and it was long in only 54.7% of all months across the full sample.

This is the actual boundary between the two families, and it is a mechanical one, not a stylistic one:

A grid is short the option to be wrong indefinitely. A trend system is short the opportunity to be right quickly.

When the market trends down for years, the grid keeps adding to a losing position with no stop, and the only exit is a recovery that may take a generation. When the market trends up for years, the trend system keeps catching legs with a small, bounded loss per failed signal.

What happened after 2000

From 2000 the ordering still does not flip. Buy and hold returned 10.50% a year in that window; G2 returned 2.74% and G1 3.86%. The grid still came third.

The reason is visible in the exposure numbers below. A grid with 20% ladder depth is not fully loaded most of the time, so most of the time it is simply not invested. It buys dips mechanically — but it also spends long stretches entirely in cash while the trend system is in the market and compounding.

The geometry that decides whether you survive

The most important number in this article is not a return. It is the ladder depth, s × L, and how it compares to the drawdowns gold actually does.

From the same monthly series:

Monthly gold decline Occurrences (801 months) Approx. per year
≤ −1% 244 (30.5%) 3.7
≤ −2% 163 (20.3%) 2.4
≤ −3% 105 (13.1%) 1.6
≤ −5% 50 (6.2%) 0.7
≤ −8% 11 (1.4%) 0.2
≤ −10% 10 (1.2%) 0.1

Worst single months in the sample: March 1980 (−16.7%), March 1983 (−14.5%), December 1980 (−13.9%), March 1982 (−11.8%), June 2006 (−11.7%).

Now combine that with the ladder depth:

Grid Ladder depth Worst-case single-month hit vs. deepest level
G1 (1% × 20) 20% −16.7% month breaches the deepest level by a wide margin
G2 (2% × 10) 20% same
G3 (3% × 10) 30% −16.7% stays inside the ladder

A 20%-deep ladder cannot absorb a March 1980. In the simulation, G2 had all 10 levels open for 403 of 802 months — 50.2% of its life — and the worst distance from the weighted-average entry of a fully-loaded ladder to the then-current price was −58.0% (July 1999: a weighted average entry of $609.84/oz against a $256.00/oz price). A fully-loaded 20% ladder was sitting at a 58% loss on average, in a market that had not yet seen its bottom.

Meanwhile, because u = 10% per level, a fully-loaded G2 ladder reaches 64.4% of account equity in notional gold exposure (average 34.7% over the sample). There is no margin left for a further adverse move — not because the model used leverage, but because ten 10%-sized tranches add up to 100% of the account, and the last tranche was funded by cash earned in the tranches before it.

This is the mechanical reason grids blow up: a ladder whose depth is smaller than the asset’s historical drawdown is not a strategy, it is a schedule for a forced unwind. And the unwind is forced by the broker’s margin rules long before it is forced by your own judgement.

Making the ladder deeper fixes the margin problem and creates a worse one

Compare G3 (3% × 10 = 30% depth) with G2:

  • G3 had all 10 levels open for 340 months instead of 403 — a smaller share of its life spent fully loaded.
  • G3’s total return dropped to 1.86× from 2.16×. Its CAGR fell to 0.93% from 1.16%.
  • The reason for the drop: a deeper ladder with the same number of levels and the same per-level size holds more capital in a falling market, and that capital earns nothing while the position is underwater.

Deeper is safer per-level and worse per-dollar. That is the trade, and it cannot be optimised away — it is arithmetic.

Adding the cost of trading

Now apply a spread. The base results above have no transaction costs, so they flatter both systems. Here is G2 and G1 with a $0.30 round-turn spread on XAUUSD spot charged on entry and exit, held constant in dollars (a simplification; in practice the spread is a fraction of price, so it mattered far more when gold was $35 than at $4,105):

Grid Gross CAGR Net CAGR after $0.30 Gross final equity Net final equity
G2 (2% × 10) 1.16% 1.12% 2.16× 2.10×
G1 (1% × 20) 1.52% 1.45% 2.74× 2.61×

The cost is small in relative terms because these grids are infrequent: G2 generated 129 fills and 45 complete cycles across 802 months. That is roughly one fill every six months.

This is worth stating plainly because it cuts against the usual broker marketing: a wide, slow grid trades very little, so spreads do not destroy it. The thing that destroys a slow grid is the drawdown, not the spread. A tight, fast grid trades far more and the cost column becomes the dominant term.

Note also what the cost model does not include: overnight financing (a real drag on any multi-month grid), commission on a per-lot basis, and the difference between your broker’s XAUUSD and the Pink Sheet’s spot average.

What actually decides it — the decision framework

If you are choosing between the two families, the question is not “which is better” but “what does the market do between my re-entries?”

Choose the grid when:

  • The instrument’s historical drawdowns are shallow relative to your ladder depth. Check the actual numbers for your symbol, not a rule of thumb.
  • You can size the ladder so that a fully loaded ladder leaves margin headroom for a further move as large as the worst monthly decline you have observed, plus a margin call buffer.
  • The expected benefit from buying dips (a mechanically captured convexity payoff) is worth more to you than the risk of being long through a multi-year decline.
  • You are willing to sit in a drawdown for years without intervention. In the sample, G2 was underwater for 306 consecutive months.

Choose trend following when:

  • You cannot emotionally hold a 50%+ drawdown without changing the system.
  • The market can trend hard in one direction for a long time (gold 1980–2006 being the canonical example).
  • You would rather be flat and wrong than long and bleeding. The 12-month rule was in cash 45.3% of the time and that was its main defence.
  • You want drawdown bounded by design rather than by hope. −31.8% vs −62.1% in the same sample.

Choose neither when:

  • You cannot compute the ladder depth against the instrument’s real drawdown history. If you cannot size the position, you do not have a strategy.

The honest summary

On this sample, on this rule set, the grid lost to simply owning gold by a wide margin: G2 returned 1.16% a year against buy and hold’s 7.40%, and G1 returned 1.52%. Trend following returned 8.19% and drew down half as much as buy and hold.

The reason is not subtle and it is not a parameter choice. A grid spends most of its time partially filled or flat, and a ladder that is not fully loaded is not running a convexity payoff — it is running a smaller, slower, worse version of buy and hold, while paying two extra costs: the spread on every level and the opportunity cost of the cash sitting in unfilled tranches. The mechanism that makes a grid attractive in the middle of a range — buying the dip automatically — is the same mechanism that leaves it idle through the trends where the money was actually made.

The honest conclusion is not “grids do not work” but narrower and more useful: this grid, at these depths, on monthly average gold prices, over 1960–2026, did not earn its complexity. The ladder depth must be deeper than gold’s historical drawdowns, and making it that deep costs most of the return. Trend following did not merely beat the grid — it beat it on return and on the only metric that determines whether you get to keep trading: the size of the loss you have to sit through.

Reproducing this

Everything above comes from two inputs you can download yourself:

  1. Data — the World Bank Pink Sheet monthly file: CMO-Historical-Data-Monthly.xlsx, “Gold” column, monthly average USD/oz.
  2. Rules — the grid parameters (anchor, spacing, levels, per-level equity fraction, per-level take-profit at entry × (1+s), no stop, cycle reset) and the trend rule (trailing 12-month sign, one-month signal lag) are stated in full above.

Two changes you should make before drawing your own conclusions: trade at intrabar prices rather than monthly closes, and add your broker’s actual spread, commission and swap schedule. Every number in this article will move. None of them will move by much enough to reverse the drawdown column, and that column is the one that matters.

What went wrong the first time (and how it was caught)

This section is here because the correction above is not a rounding adjustment — the original version of this article claimed the grid returned 8.57% a year, roughly 1.2 points better than buy and hold, and that was simply false. The cause was a bookkeeping bug in the grid simulator:

on close:   cash += qty * price      # proceeds received
            (position size NOT reduced)   # <-- the bug

Each closed level added its sale proceeds to cash while the same quantity stayed in the position, so the equity curve counted every closed lot twice. Because a grid closes lots constantly, the overstatement compounded: the published G2 total of 241.3× should have been 2.16×, a factor of 112.

It was caught because this site requires every published number to be recomputed by an independent verification script written along a different code path (scripts/verify_gold_intra_month.py, which reimplements the grid with numpy array bookkeeping rather than the production script’s list-of-lots filter). The two implementations disagreed by an order of magnitude, which is what a double-counting error looks like from the outside.

The fix was a one-line change in three simulators (scripts/gold_grid_trend_analysis.py, scripts/gold_multi_account_analysis.py and the verifier), after which both implementations agreed exactly and the numbers in this article were regenerated from scratch. The corrected grid figures are also published with the raw equity curves in the multi-account dataset so the arithmetic is checkable end to end.

The lesson generalises: a backtest that flatters the product it sells is the easiest kind of number to get wrong, and the only reliable defence is an independent re-implementation that can disagree with you.


Educational content only. Not investment advice, and not a recommendation of any strategy, robot, broker, or instrument. All results above are the output of a reconstruction on monthly average prices, not a live trading record, and past simulated performance is not indicative of future results. Trading forex and CFDs on margin carries substantial risk of loss; you can lose more than your initial deposit on some account types. Every parameter choice here is illustrative — the correct ladder depth and spacing depend on the instrument, the account size, and the margin rules of the broker you use.

Frequently asked questions

What did the long-run comparison show?

On the same 802 monthly gold closes from 1960-01 to 2026-10, the 12-month trend-following rule beat every grid on both return and drawdown. G2 returned 1.16% a year against buy and hold's 7.40% and trend's 8.19%, while trend following also drew down less than half as much as buy and hold.

Why did the grid fail in the 1980s specifically?

Because the depth reached in a strong trend grew faster than the account could fund it. The grid's outcome is set by the funding arithmetic at the deepest level reached, which is why the same ladder that worked in the range-bound 1960s ran out of margin in the 1980s trends.

Does adding trading costs change the comparison?

It widens the gap in one direction and narrows it in another. Costs hurt the grid more per unit of drawdown because it transacts more, but a trend system that trades less often spends more time out of the market and gives up more of the recovery.

How do I decide which approach fits my situation?

By the drawdown you can actually fund and the time you can actually sit through. On this sample the grid trades a smoother equity curve for the risk of a single deep adverse excursion, and in exchange it earned less than simply owning gold.

Can I reproduce the comparison?

Yes. The article gives the monthly close series source, the exact rules for both systems, and the parameter values, so the result can be recomputed rather than taken on trust.

What is the practical takeaway for a retail account?

The trend system requires less margin headroom and survives trends better, but it spends more time flat and its equity curve is harder to sit through. Neither is safe; they fail in different ways and need different account sizes.

Sources

  1. World Bank — Commodity Price Data (The Pink Sheet), monthly, updated November 3, 2026 — accessed 2026-11-08
  2. World Bank — CMO Historical Data Monthly (direct download used for this analysis) — accessed 2026-11-08
  3. Moskowitz, Ooi & Pedersen (2012) — Time Series Momentum, Journal of Financial Economics 104(2), 228-250 — accessed 2026-11-08
  4. Hurst, Ooi & Pedersen (2017) — A Century of Evidence on Trend-Following Investing, Journal of Portfolio Management 44(1), 15-29 — accessed 2026-11-08
  5. CME Group — Gold Futures Contract Specifications (100 troy oz, $0.10 minimum price fluctuation) — accessed 2026-11-08
  6. Gil-Alana et al. (2023) — Volatility persistence in metal prices, Resources Policy — accessed 2026-11-08

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.