The Impulse Response Function of Global Markets to a Single Truth Social Post
Danielle Fong — March 28, 2026
In condensed matter physics, the linear response function χ(ω) relates an external perturbation to the system's response. Apply a delta function (an instantaneous kick), and the output is the impulse response h(t)—the Green's function of the system.
where βi is asset i's loading on the geopolitical risk factor, τi is its characteristic timescale, Qi is its quality factor (how much it overshoots), and δi is the information propagation delay.
The hypothesis: A Trump Truth Social post about Iran is, to first approximation, a delta function in the space of geopolitical risk. Every asset's price response reveals its transfer function. The shape of that response tells you whether the market is dominated by algorithms (high Q, fast overshoot, rapid reversion) or humans (low Q, slow buildup, persistent offset).
The test: If the system is approximately linear, then:
Normalized returns relative to t=0 (one minute before impact). All 11 instruments, 50 minutes post-impulse.
TACO sell Anti-TACO-2
↓ relief off ↓ relief
ES 6420─6588 ▆▆▅▅▅▆▆▆▆▆▆▆▅▆▆▇▆▆▆▆▅▅▄▅▄▄▄▃▄▃▂▂▃▂▃▂▂▂▁
BZ 98─106 ▄▄▄▂▃▃▃▃▃▃▃▃▃▃▃▃▄▄▅▅▅▆▅▅▆▆▆▆▆▅▆▆▅▆▆▆▇▇
GC 4381─4584 ▂▁▂▁▂▃▂▂▂▂▂▃▃▃▃▄▄▄▄▃▃▃▃▃▃▃▄▄▆▇▇▇▇▆▆▅
BTC 65533─69308 ▆▇▇▆▇▆▇▇▇▆▆▆▇▇▆▆▆▅▅▄▃▂▃▂▂▂▁▁▁▁▂▂▁▁▁▁▁
Mar 26 Mar 27 Mar 28
| NQ | RTY | GC | BZ | BTC | ZN | |
|---|---|---|---|---|---|---|
| ES | 0.996 | 0.989 | 0.343 | −0.910 | 0.892 | 0.924 |
| BZ | −0.907 | −0.925 | −0.423 | — | −0.806 | −0.907 |
One principal component explains ~90% of variance. Oil is the anti-factor at r = −0.91.
Each market's response can be classified by its analogy to a damped harmonic oscillator:
ALGO DOMINATED
Sharp peak, fast exponential ringdown. 70–90% of the initial move reverts within 10 minutes.
Assets: ES, NQ, RTY, YM, BZ, ZN, ZB
RETAIL / MIXED
Quick rise, slow settling. No overshoot. ~50% of the move persists.
Assets: BTC, ETH
MACRO / HUMAN
Slow buildup over 30 minutes. Not an impulse response at all—it's responding to what the impulse means, which takes time for humans to compute.
Assets: GC (gold)
HUMAN ONLY
2-minute dead time. Three-lot trades. One human at a time placing orders after reading the news.
Assets: SI (silver)
| Asset | Model | Peak | Residual | Overshoot | τfast | tpeak | Type |
|---|---|---|---|---|---|---|---|
| BZ | 2-phase | −3.37% | −0.34% | 89.8% | 3.1 min | 1 min | ALGO |
| ZN | 2-phase | +0.24% | +0.04% | 83.5% | 0.8 min | 1 min | ALGO |
| ES | 2-phase | +0.94% | +0.24% | 74.9% | 1.3 min | 1 min | ALGO |
| NQ | 2-phase | +0.85% | +0.22% | 74.1% | 1.6 min | 1 min | ALGO |
| YM | 2-phase | +0.99% | +0.25% | 74.7% | 1.5 min | 1 min | ALGO |
| RTY | 2-phase | +1.66% | +0.46% | 72.0% | 1.2 min | 1 min | ALGO |
| ZB | 2-phase | +0.36% | +0.11% | 68.5% | 0.9 min | 1 min | ALGO |
| GC | driven | +1.45% | +0.54% | 62.4% | τrise=7.4 | 29 min | DRIVEN |
| BTC | simple | +1.15% | +0.61% | 44.6% | τ=60 | 3 min | MIXED |
| ETH | simple | +1.32% | +0.70% | 47.4% | τ=60 | 13 min | MIXED |
| SI | step+delay | +2.40% | +2.08% | 13.0% | δ=2 min | 5 min | HUMAN |
t=0s ZN: 79,700 contracts Bonds LEAD. Always. ES: 17,279 NQ: 5,343 Equity algos same second BZ: 548 (100× normal) Oil algos selling BTC: +1.15% ETH: +1.32% Crypto follows t=+2m BZ low hits −4.45% Algo selling CONTINUED RTY high +1.95% Small caps still overshooting t=+3m GC vol peaks at 258 Gold ACCELERATING SI: FIRST TRADE. 3 lots. Silver was DEAD for 2 minutes t=+5m SI peaks at +2.40% One contract at a time. Humans. t=+29m GC peaks at +1.45% Gold STILL building
| t (min) | ES (×avg) | NQ | RTY | BZ | ZN | GC |
|---|---|---|---|---|---|---|
| 1 | 3.5× | 4.1× | 6.2× | 97× | 26.7× | 7.6× |
| 2 | 3.1× | 2.9× | 3.4× | 24.3× | 19.3× | 6.7× |
| 3 | 1.5× | 1.5× | 1.8× | 33.1× | 10.9× | 7.9× |
| 5 | 1.2× | 1.1× | 1.4× | 19.1× | 4.0× | 2.9× |
| 10 | 0.4× | 0.4× | 0.8× | 9.0× | 1.1× | 1.7× |
| 15 | 0.7× | 0.7× | 0.9× | 23.6× | 0.8× | 1.9× |
BZ has secondary volume surges at t=3 and t=15 — multiple algo tiers firing on different timescales. ZN has the cleanest exponential decay (t½ = 1.5 min). Gold volume peaks at t=3-4, not t=1 — human rebalancing.
Brent crude at t=1 (the impulse minute): Open: 101.47 (pre-impulse) High: 101.39 (−0.08% — barely dipped before the crash) Low: 97.40 (−4.10% — algo hit the entire book) Close: 98.14 (−3.28% — partial recovery within same minute) Full intra-minute range: 4.02% At t=2: low hit 97.04 (−4.45%) — even lower. Algo still selling. By t=39: BZ = −0.45%. 89% of a 4.5% crash was algorithmic.
Hyperliquid is a DeFi perpetual futures exchange with 229 markets, $1.8B BTC open interest, and 24/7 trading. Three data streams prove critical:
| Coin | t=1 min | Volume |
|---|---|---|
| BTC | +1.12% | 400.7 BTC |
| ETH | +1.25% | 4,988 ETH |
| SOL | +1.33% | 31,421 SOL |
| TRUMP | +1.40% | 11,887 lots |
TRUMP perp is the purest sentiment proxy. +1.4% on de-escalation.
| Coin | t=1 min | Volume |
|---|---|---|
| SOL | −1.08% | 101,438 SOL |
| ETH | −0.65% | 10,152 ETH |
| TRUMP | −0.69% | 18,909 lots |
| BTC | −0.22% | — |
SOL volume explosion: liquidation cascade on the DEX. Crypto LED this selloff.
The Anti-TACO-1 event (Mar 27 10:37 UTC) reveals something the TACO didn't: crypto markets are the leading indicator for geopolitical events that hit during off-hours.
Anti-TACO-1 propagation at t=1 minute: ETH: −0.702% ← CRYPTO LEADS (24/7 market) BTC: −0.224% ES: −0.015% ← equities barely moved NQ: −0.011% BZ: −0.019% At t=5 minutes: ETH: −1.382% ← STILL ACCELERATING (liquidation cascade) BTC: −1.039% ES: −0.084% ← now following crypto down NQ: −0.097%
The Hormuz news hit crypto first, triggered perp liquidations on Hyperliquid (101K SOL in one minute), and equities followed 5 minutes later. During US market hours, the hierarchy is Bonds→Equities→Crypto. During off-hours, it inverts: Crypto→Equities→Bonds.
| Coin | Price | Funding | Open Interest | 24h Vol |
|---|---|---|---|---|
| BTC | $66,916 | +0.0013% | $1,814M | $1,044M |
| ETH | $2,022 | +0.0013% | $1,174M | $553M |
| SOL | $83.36 | −0.0028% | $308M | $122M |
| TRUMP | $2.99 | −0.0056% | $13.2M | $1.0M |
TRUMP funding deeply negative — heavy short interest. SOL also net short. Market positioning for continued escalation.
We captured three distinct impulse events in 48 hours, each with different physics:
| TACO (de-escalation) | Anti-TACO-1 (Hormuz) | Anti-TACO-2 (relief) | |||||||
|---|---|---|---|---|---|---|---|---|---|
| Asset | t=1 | t=10 | Revert | t=1 | t=10 | Revert | t=1 | t=10 | Revert |
| ES | +0.94 | +0.28 | 70% | −0.02 | −0.07 | ACCEL | +0.28 | +0.35 | ACCEL |
| BZ | −3.37 | −1.14 | 66% | −0.02 | +0.20 | REV | −0.76 | −0.60 | 22% |
| GC | +0.51 | +0.95 | ACCEL | −0.03 | −0.22 | ACCEL | +0.23 | +0.10 | 57% |
| BTC | +1.15 | +0.81 | 30% | −0.22 | −0.96 | ACCEL | +0.69 | +0.68 | 3% |
| ETH | +1.32 | +1.05 | 21% | −0.70 | −1.24 | ACCEL | +0.81 | +0.75 | 7% |
The TACO (de-escalation) overshoots and reverts. Algos buy the news instantly, creating a spike. Then the spike fades as humans assess the actual significance. 70-90% reversion.
The Anti-TACO (escalation) ACCELERATES. The initial move is small, but it keeps going for 10+ minutes. No reversion. This is because selling into fear is fundamentally different from buying into relief:
The system is NOT linear. The response to +Δ is not the mirror of −Δ. De-escalation creates an impulse that reverts. Escalation creates a cascade that amplifies. This asymmetry is the key insight for trading.
Google's TimesFM (500M parameter, pre-trained on 100B+ time points) was given 60 minutes of pre-impulse context + 3 minutes of impulse data and asked to predict the remaining 37 minutes. Zero training on financial data from this event.
| Asset | Revealed (t=3) | Actual Residual | Predicted Residual | Direction | MAE |
|---|---|---|---|---|---|
| YM | +0.56% | +0.34% | +0.21% | ✓ CORRECT | 0.16% |
| ZB | +0.31% | +0.18% | +0.04% | ✓ CORRECT | 0.12% |
| ES | +0.49% | +0.33% | −0.05% | ✓ CORRECT | 0.35% |
| NQ | +0.50% | +0.33% | −0.11% | ✓ CORRECT | 0.40% |
| BZ | −2.32% | −0.59% | +0.11% | ✓ CORRECT | 0.71% |
| GC | +1.25% | +1.13% | +0.11% | ✓ CORRECT | 0.71% |
| BTC | +1.16% | +0.88% | −0.16% | ✓ CORRECT | 0.86% |
| ETH | +1.32% | +0.98% | −0.14% | ✓ CORRECT | 0.88% |
| SI | +1.97% | +2.08% | −0.37% | ✗ WRONG | 2.39% |
10/11 correct on reversion direction (91%). The model's failure mode: it over-predicts reversion for BTC/ETH (which retained more than expected) and completely fails on silver (a step function, not an impulse). TimesFM's prior is "spikes revert"—correct for algo markets, wrong for human-driven step changes.
| Asset | Edge (% of peak) | τfast | Hold Time | $/contract |
|---|---|---|---|---|
| BZ (Brent) | 89% × 3.37% = 3.0% | 3.1 min | 9 min | $30,000 |
| RTY (Russell) | 72% × 1.66% = 1.2% | 1.2 min | 4 min | $15,000 |
| ES (S&P) | 75% × 0.94% = 0.7% | 1.3 min | 4 min | $2,300 |
| ZN (10Y) | 84% × 0.24% = 0.2% | 0.8 min | 3 min | $2,000 |
You only fade de-escalation impulses. Escalation events don't revert—they cascade. The asymmetry is the strategy's moat: most quantitative models assume symmetry. The market isn't symmetric. Fear feeds forward. Relief overshoots and fades.
We treated a Truth Social post as a physics experiment and discovered:
"every single commodity and currency moving like a drum. impulse response of the market."
— @DanielleFong, March 26, 2026