The seven factors that move XAU/USD — the mechanism behind each, where to read it, the horizon it applies to, and how to weight them into one directional call.
Usable factors7
Mid-term anchorReal yields
60s 1σ move~0.02%
Conclusion firstA framework, not a view
7 / 7
Only four factors reliably explain gold's direction beyond one quarter: real yields, the dollar, central-bank buying, and investment demand. Inflation expectations are the flip side of real yields, risk aversion is an impulse, and supply plus cost only define the long-run floor.
Any single narrative — safe haven, inflation, central banks — breaks at some point, and it will not tell you in advance.
I. The seven factors, one by oneBy weight
01Real yields · the mid-term anchorLeads 0–3 months
Conclusion: the 10-year TIPS real yield is the single most stable explanatory variable for gold, negatively correlated, with a historical regression elasticity of roughly −8% to −12% per +100bp.
Gold pays no coupon. The cost of holding it is the risk-free real return you forgo. Note it is the real yield, not the nominal one — nominal rates hit double digits in the 1970s and gold still rallied, because inflation was higher.
That anchor loosened after 2022: between October 2022 and October 2024 the 10-year real yield rose from about +1.5% to +1.9%, while gold went from 1,650 to 2,740. Both moved up together, which means something stronger than rates was buying. So use it as a deviation measure: the more gold trades above the rate-implied value, the heavier the non-rate bid — and when that bid turns, the drawdown is fast and deep.
Read
FRED: DFII10 (daily, T+1); 10-year TIPS auction stop-out real yield and bid-to-cover
Fails when
In a dollar-liquidity crisis, forced liquidation drives both legs and the correlation scrambles
02The dollar · a pricing effectIntraday – weekly
Conclusion: the rolling 60-day correlation between DXY and gold usually sits at −0.6 to −0.8, the fastest factor on intraday-to-weekly horizons — but it fails in a dollar squeeze.
This is a pricing effect, not causation: a stronger dollar makes gold more expensive in local currency for non-US buyers, which caps marginal demand. DXY is 57.6% euro, 13.6% yen, 11.9% sterling — it is "USD versus Europe and Japan", not versus the yuan, so pricing gold off USD/CNY misplaces the signal.
Two failures worth memorising: March 2020 and September 2022. The world scrambled for dollars and gold became the easiest thing to sell for cash. Watch the funding indicators: EUR/USD basis, FRA-OIS, Fed repo take-up.
Read
DXY, EUR/USD, US–Germany 10-year spread, USD/JPY
03Central-bank buying · the floorQuarterly+
Conclusion: the most important structural variable since 2022. This bid is price-insensitive; annual purchases have exceeded 1,000 tonnes for three consecutive years, about 28% of global mine supply.
The motive is not return, it is reserve safety — avoiding the risk of reserves being frozen. So it ignores short-term price: it buys on the way down and on the way up, which raises the floor rather than pushing a ceiling.
Scale check: 1,000 tonnes is 32.15 million ounces, roughly USD 77bn a year at 2,400/oz. Modest against the total ETF complex, but it is one-way, price-inelastic and never takes profit — which is exactly how it can overpower real yields.
04Investment demand · ETFs and COMEXWeekly – monthly
Conclusion: ETF tonnage is the direction of slow money — trade with it. COMEX non-commercial net length is the crowding of fast money — trade against it.
SPDR GLD tonnage is the most-quoted sentiment gauge: a peak near 1,350t in 2012, a low near 830t in mid-2024. Watch continuity — four or more consecutive weeks of inflows usually accompanies trend continuation; gold making new highs while tonnage stalls is the classic price-volume divergence.
CFTC COT non-commercial net length prints Friday 15:30 ET, measured as of Tuesday — structurally three days stale. Normal range is 100k–250k contracts; above 300k signals a crowded long and rising drawdown risk; below 50k is usually capitulation. 100k contracts is 10 million ounces, about 311 tonnes.
Read
GLD / IAU daily tonnage, CFTC COT, COMEX inventories and the EFP spread
05Inflation and inflation expectationsData window → weekly
Conclusion: nominal rate minus inflation expectation equals the real rate. Inflation alone does not lift gold — a surprise does, and only after being filtered through "what will the central bank do about it".
The strongest macro mix is stagflation: high inflation, weak growth, a central bank that cannot hike. The 1970s and H2 2022 ran that script. The counter-example is H1 2022: US CPI above 8% and gold fell from 2,000 to 1,680 — the market was not trading inflation, it was trading "the Fed will hike until recession", which makes inflation the fuel for tightening.
Conclusion: gold rallies driven by a VIX spike are impulses, not trends — historically most of the move is given back within 5 to 15 sessions.
Haven flows buy "the most liquid non-USD asset", and gold qualifies. But when the shock itself is a dollar shortage, gold also becomes the easiest thing to sell — October 2008 and March 2020 both rallied first and then sold off hard.
For short horizons the useful question is not the VIX level but whether the first 30 minutes of direction holds. The knee-jerk gap on a data print is mostly retraced within two hours, which is why "wait 30 minutes" often has better expectancy than "take the first second".
Spreads routinely widen 2–5× at the print; slippage eats the entire edge
07Supply, cost and physical demandYearly
Conclusion: slow variable. It sets the long-run floor, not the direction, and explains no two-week move.
Mine supply runs about 3,600–3,700 tonnes a year, recycling adds roughly 1,200t, total supply near 4,800t — and it is extremely rigid, since a mine takes 10+ years from exploration to production. All-in sustaining cost is about 1,500/oz, the cost floor: trading below marginal cost is unsustainable, but "unsustainable" can last three years.
On the physical side: China and India together are more than half of jewellery demand, and the Shanghai premium over London is a real-time tightness gauge — a widening premium means strong Asian physical buying, a discount means selling.
Read
WGC quarterly supply-demand tables, Swiss and Hong Kong export data, Shanghai–London spread, India import premium
II. Weight matrix: same factor, different horizon
Factor60sDay–weekQuarter+
Event window & liquidity60——
Real yields102530
US dollar102015
Central-bank buying0525
ETF + COMEX positioning52010
Inflation expectations01010
Risk aversion / VIX15150
Supply & physical demand0510
Total100100100
Percent. Weights are research priors, not a backtested optimum. The three horizons must not be mixed — that is the single most common way retail traders lose money.
III. Three steps to one directional call
1
Pick the horizon, then the model
Sixty seconds and six months are two different models. At 60s the factors are nearly irrelevant — the event window and book depth set the price. Over a quarter, events are noise and real yields plus central banks are the story.
2
Score each factor −2 to +2, weight, sum
Strong bullish +2, neutral 0, strong bearish −2. The weighted sum lands between −2 and +2. Above +0.5 is long-biased, below −0.5 short-biased, and the middle is a noise band — no position.
3
Calibrate to payout, not to conviction
The composite is only a directional probability; the payout decides whether it is worth taking. A 68% payout needs a 59.5% win rate to break even; 90% needs 52.6% (math on the transparency page). Your historical win rate minus break-even is the expectancy. Negative means no trade.
To be blunt: at a 60-second horizon the seven factors together still fail to explain most of the move — 1σ is about 0.02%, roughly USD 0.5 on a 2,400 ounce, so noise dominates signal. What makes money at that scale is execution quality (spread, slippage, fill) and discipline, not forecasting.
IV. Three methodological errors
01Treating correlation as causation
Gold up and VIX up does not mean "because safe haven" — a third leg, dollar liquidity, may be pushing both. Test causation by asking what moved first and what jumped across the event, not by how similar two lines look.
02Using stale data for real-time decisions
COT lags three days, WGC supply-demand lags six weeks, central-bank reserves lag one to two months. They explain how we got here; they do not predict the next minute.
03Forgetting the words "real" and "denominated"
Reading nominal instead of real rates, and USD gold instead of local-currency gold, accounts for most gold misreads. Subtract inflation from rates, multiply gold by the exchange rate; skip one step and the conclusion inverts.
Beijing time. Before the slash: US daylight saving (2nd Sunday in March to 1st Sunday in November); after: standard time. ★★★ marks windows where 5-minute gold volatility can run 3–6× normal.
Factors give direction. The payout decides whether it is worth taking.
XAUAD does one thing: turns a directional call into a settleable contract. 60-second expiry, 68%–90% payout, zero spread and zero slippage with fills at the public mid, USDT funding from 10.
Basis: all prices are XAU/USD spot mid; 1 tonne = 32,150.7 troy ounces; volatility annualised at 15% and scaled to 60 seconds. Elasticities and weights are research ranges, not promised returns. Any model can fail; nothing here is investment advice.
English
简体中文