▶ WHY GOLD MOVES: AN INTERACTIVE TUTORIAL
2026-03-19 | Live market parameters
Here's the puzzle: gold pays you nothing. No interest, no dividend, no coupon. The only reason to hold it is when everything else pays you even less in real terms — or when you're afraid everything else might fall apart. Right now, bonds are paying more than inflation, which means there's a concrete cost to sitting in gold instead. On top of that, gold is priced in dollars. When the dollar gets stronger (DXY 104.2), the same ounce just costs fewer of them — the price has to fall to clear the market worldwide. Two separate mechanisms, both pushing the same direction, compounding. The real rate is 1.53%. Move the sliders below and watch them interact.
Real Rate = Nominal Rate − Expected Inflation Gold ≈ G₀ × exp(−λ × ΔRealRate) [λ ≈ 0.12] Interpretation: every 1% rise in real rates → ~12% fall in gold
The market doesn't wait for the Fed to act. By the time a rate cut actually happens, gold has often already moved — because traders are constantly pricing in what they think the Fed will do next. The real signal isn't the current rate; it's the gap between where rates are now and where the market thinks they're heading. That gap is what this model captures.
CB HISTORICAL SCENARIOS
Gold(USD) = G₀ × (DXY₀ / DXY)^γ [γ ≈ 1.5] If DXY rises 10% → gold falls ~14.5% (power-law amplification)
log(Gold) = α − β×RealRate − γ×log(DXY) + ε Gold = A × exp(−β×RealRate) × DXY^(−γ) [α=8.1, β=0.12, γ=1.5]
| FACTOR | PROXY | TYPICAL β | INTERPRETATION |
|---|---|---|---|
| Real Rate Sensitivity | TIPS 10Y | -9.0 | Duration risk of "zero-coupon perpetual bond" |
| Currency Beta | DXY | -1.8 | Inverse dollar exposure |
| Risk Premium | VIX | +0.3 to +0.5 | Convexity during tail events |
TARGET: Make gold reach $---
Use the combined model sliders above. Your current model price:
Progress toward target (50% = at baseline)
You moved gold to the target price!
You understand opportunity cost and dollar dynamics.
The real test of any model isn't how well it fits the data it was trained on — it's whether it makes sense of history you haven't shown it yet. Load each scenario and see where the model was right, and where reality surprised it.
Before you look at the numbers, take a guess: which matters more — interest rates, inflation, or the dollar? It's not obvious, and the answer shifts depending on the starting values. Move each slider by +1 unit and see how your intuition holds up.
| → SAFE-HAVEN | → INFL-HAWK | |
|---|---|---|
| FROM SAFE-HAVEN | — | — |
| FROM INFL-HAWK | — | — |
Gold and silver have traded together for thousands of years — dug from the same ground, bought by the same people, driven by the same monetary fears. That shared history creates a gravitational pull: when the ratio between their prices drifts too far from its long-run average, something tends to snap it back. The residual $z_t$ measures exactly how far we currently are from that anchor. When $|z_t| > 2\sigma$, history says reversion becomes likely.
Think of gold as a bond that pays no interest and never expires.
Every other asset competes with it for your money.
When real rates are positive — when your savings account
actually beats inflation — there's a concrete cost to holding
gold instead. You're giving something up.
The exponential formula follows directly: if opportunity cost
rises, demand falls, and price adjusts. Why exponential and
not linear? Because percentage changes compound — a 1% shift
in real rates has a larger absolute price impact at $3,000/oz
than it did at $1,000/oz.
λ ≈ 0.12: every 1% rise in real rates → ~12% fall in gold.
Not derived from theory. Measured. Rolling 10-year regressions
on 24 years of London PM Fix data vs 10-yr TIPS yields.
The number is stable across different sub-periods, which is
the only reason to trust it.
Gold trades everywhere but gets quoted in dollars. A saver in
India doesn't think in dollars — she converts. When the dollar
strengthens, that same ounce costs her more rupees. Enough
people making that calculation and global demand falls. The
dollar price drops to clear the market. The ounce itself hasn't
changed; only the unit of measurement has.
Pure PPP math predicts γ = 1. The data says γ ≈ 1.5. That
extra sensitivity probably comes from two sources:
1. Traders who specifically pile into gold when the dollar
weakens — a correlated speculative bet.
2. Emerging-market central banks that buy more gold when the
dollar strengthens (diversifying away from it) — a
feedback loop that amplifies the basic effect.
Neither appears in the theory. Both show up in the data.
Take 24 years of monthly gold prices. Add real interest rates
and the dollar index. Run the regression.
Fitted: α=8.1, β=0.12, γ=1.5
R²≈0.89, Adj-R²≈0.88
Two variables explain 89% of gold's price variance. The other
11% is everything else: wars, sanctions, central bank buying,
the invention of Bitcoin, a pandemic. The ε term catches all
of it in one catch-all residual.
Why log form? Because the relationship is multiplicative.
A 10% dollar move has the same proportional effect at $1,000
as at $3,000. In log space, that becomes a straight line that
a regression can actually fit.
Be honest about the limits: ±8% error is typical. In a crisis,
±15%. The model is a compass, not a GPS.
The problem with simple regression: it assumes causality flows
one way. Rates affect gold. End of story. But gold also affects
inflation expectations, which affect the Fed, which affects
rates, which comes back to gold. Everything connects.
A VAR lets every variable explain every other, using each
one's own past. Gold, oil, the dollar, the Fed rate, and
inflation all go in. The math finds the connections itself.
Y_t = [Gold_t, DXY_t, Brent_t, FedFunds_t, InflationExp_t]'
The useful output is the impulse response: if oil spikes by
1 standard deviation today, what happens to gold over the
next 12 months?
The answer depends entirely on the regime:
Inflationary regime:
oil → Fed hikes faster → higher real rates → gold falls
Panic regime:
oil → stagflation fear → safe-haven demand → gold rises
Same shock. Opposite response. This is why regime detection
from Section 4.5 isn't optional — it determines the sign.
R² within-sample: ~0.82 at 1 month. Degrades quickly beyond
3 months. Don't over-trust it.
Volatility doesn't behave like a fresh random draw each day.
A wild day is usually followed by another wild day. A quiet
week tends to stay quiet. This clustering is the whole point.
Parameters (London PM Fix daily returns):
ω ≈ 0.00003 (long-run variance floor)
α ≈ 0.08 (how much yesterday's shock raises today's vol)
β ≈ 0.90 (how much yesterday's vol carries forward)
α + β ≈ 0.98 → nearly integrated: shocks decay very slowly
With β = 0.90, a spike in volatility has a half-life of about
20 trading days. The system is still agitated long after the
event that caused it.
Gold's return distribution has negative skew — the left tail
is fatter than the right. Big losses are more likely than big
gains of the same magnitude. The reason: margin calls cascade.
Forced selling compounds. Lognormal models that treat the
distribution as symmetric are systematically underpricing the
downside. Skew-adjusted models don't have that blind spot.
Every ounce of gold ever mined is still around. It doesn't
rust, burn, corrode, or get consumed. That single physical
fact changes everything about supply dynamics.
Total above-ground stock: ~212,000 tonnes
Annual mine production: ~3,500 tonnes (S2F ≈ 60 years)
Annual recycled scrap: ~1,200 tonnes
Compare to oil: S2F ≈ 1.5 years. You burn through the world's
entire oil stockpile in 18 months. For gold it takes 60 years.
Doubling mine output would add a rounding error to existing
supply. Gold's price cannot be inflated away by digging more.
Demand breakdown (WGC 2024):
Jewelry: ~46% (price-sensitive)
Investment: ~25% (ETFs, coins — very price-sensitive)
Central Banks: ~20% (NOT price-sensitive)
Technology: ~9% (industrial — inelastic)
The central bank slice is the structural shift most models
miss. They're not buying because they think gold is cheap.
They're buying because they want reserves that can't be frozen
by a sanctions order. That logic doesn't change when gold
is expensive, and it creates a demand floor the rate-and-dollar
model was never designed to capture.
Options pricing reveals what the market is secretly thinking.
The 25-delta call profits if gold rallies significantly. The
25-delta put profits on a large drop. If calls cost more than
puts, the market is paying extra for upside — it's more afraid
of missing a rally than of a crash. That's a positive risk
reversal, and it's a bullish structural signal.
Negative risk reversal: the smart money is buying downside
protection. Often a smarter signal than watching spot price.
The detective read:
Gold FALLING but RR stays POSITIVE:
Sophisticated money is still buying calls.
The dip is tactical, not structural. Don't panic.
Gold RISING but RR turns NEGATIVE:
Owners are quietly insuring against reversal.
The rally may be borrowed time.
COMEX options matter too. Heavy put buying at round numbers
like $3,000 creates hedging clusters — they act as resistance
on the way up and support on the way down, because dealers
have to dynamically hedge their books around those strikes.
The current Fed rate is public knowledge. Gold already priced
it in yesterday. What the market is constantly arguing about
is what the Fed will do next — and that argument moves prices.
By the time a cut is announced, gold has often already rallied.
The event is anticlimactic. The move happened weeks earlier,
when futures markets repriced the expected path.
The dot surprise is the key number to watch:
Dot Plot > Market Expected Rate:
Fed is more hawkish than assumed → forward real rate
rises → gold falls
Dot Plot < Market Expected Rate:
Fed blinks first → forward real rate falls → gold rallies
before a single cut happens
Historical evidence:
Q4 2023: No cuts had happened. But the market started
pricing aggressive cuts for 2024. Gold rose +15% on pure
expectation repricing. The actual first cut in Sept 2024
barely moved the needle — already priced.
Jan 2019 Powell Pivot: Fed signaled pause after Dec 2018
hike. Expected rate fell ~100bp below dot plot. Gold +12%
in 60 days without a single actual rate change.
Jackson Hole 2023: Powell surprised hawkish. Dot plot
above market consensus. Forward real rates surged.
Gold -7% in 6 weeks.
Formula: pathGold = G₀ × exp(−λ × (forwardReal − spotReal))
Same λ ≈ 0.12 as Model 1. The mechanism is identical;
only the rate input changes from spot to forward.
No single model is right. But each captures something real. This section stacks them: start with the fundamental anchor (rates and the dollar), layer in the regime adjustment (which market environment are we in right now?), then add the tactical signals (what are traders and geopolitics doing today?). Each layer is a bet about which forces dominate at which time horizon. Adjust the components and watch how they compound.
Educational tool only. Not financial advice. Model parameters estimated from historical data and may not predict future prices.