Trump went to Beijing. Bitcoin reacted to Wall Street.
During the May 13-15 2026 Trump-Xi summit, BTC sold off twice and bottomed within $100 of the same level ($78,730 and $78,632). The news cycle blamed Xi's Taiwan warning and the summit's anticlimactic close. Order flow tells a different story: both cascades started within 30 minutes of NYSE cash open, alongside SPY, QQQ, gold, and oil.
Three days of Trump in Beijing, two crashes on the BTCUSDT perp, both bottoming within $100 of the same level ($78,730 on May 13, $78,632 on May 15). CoinDesk attributed the May 14 dip to Xi's Taiwan warning, and crypto.news called the May 15 close-of-summit drop a "strategic stability disappointment." Both takes match the headline narrative.
Neither matches the order book.
Both cascades started within eight minutes of NYSE cash open. Both coincided with SPY, QQQ, gold, and oil printing their session lows at 13:45 UTC. Funding rates on the BTC perp were already in the third percentile of a trailing trailing two-year window thirteen hours before the May 15 cascade fired. And during the cascade itself, $51.1 million of longs were liquidated on Binance against $626 thousand of shorts.
This is what those three days looked like from inside the matching engine.
The trip in three sentences
Trump landed in Beijing late on May 12 with a 17-CEO delegation that included Musk, Cook, Huang, and Fink. The summit covered trade, AI infrastructure, and critical minerals, with a "strategic stability framework" announced at close on May 15. The word "crypto" was not mentioned once in any public summit material.
What the BTC perp did across those three days:
Day Open High Low Close Δ
May 13 (arrival) $80,470 $81,270 $78,730 $79,318 −1.4%
May 14 (Xi warning) $79,318 $81,626 $79,184 $81,201 +2.4%
May 15 (closing) $81,368 $81,604 $78,632 $79,078 −2.8%
Source: BTCUSDT perpetual mid-price aggregated from our cross-venue microstructure feed. Times UTC.
The headlines wrote about the directions of the moves. We are going to write about the timing.
Day 3: the capitulation, by the minute
The May 15 drop was the cleanest of the three. Below is the price path between 13:30 and 14:30 UTC, downsampled to two-minute resolution. NYSE cash open is 13:30.
13:30 ████████████████████████ $79,852
13:38 ██████████████·········· $79,364
13:46 ███████················· $79,000
13:52 ██······················ $78,772 ← peak CVD-30m −$216.7M
14:00 ························ $78,632 ← LOW
14:08 ██████·················· $78,960 ← reversal starts
14:22 ██████████·············· $79,146 ← peak CVD-30m +$285.0M
Two metrics, one window:
- Cumulative Volume Delta (CVD) on a 30-minute trailing window shifted from −$216.7 million at 13:52 UTC to +$285.0 million at 14:22 UTC. The rolling sum of aggressive sells flipped to a rolling sum of aggressive buys in thirty minutes. (CVD signs every taker trade by the side of the aggressor. Negative means more aggressive selling than buying.)
- Order-book imbalance (OBI) on Binance held in the +0.20 to +0.35 range throughout the decline, meaning the bid side of the book was up to 35 percent deeper than the ask side even as price was falling. Bids were being stacked below the market while sellers swept them.
The cascade burned itself out in 32 minutes. Then aggressive buyers came in.
Where the long-liquidations stack
In that one-hour window (13:30-14:30 UTC), aggregated across BTC, ETH, SOL, BNB, XRP, and DOGE perps:
Venue events longs liquidated shorts liquidated total
Binance 1,012 $51,076,348 $626,342 $51.7M
OKX 617 $4,276,270 $170,719 $4.4M
(Bybit excluded from this table; see Methodology.)
The single biggest minute was 13:51 UTC: 169 events on Binance, $19.65 million of longs in 60 seconds, zero shorts. That minute is also the one where CVD-30m crossed −$200M for the first time.
This is not a wash-out of crowded longs in the textbook sense. By the time the cascade fired, longs were already a minority (more on funding below). What got liquidated was the leftover one-sided positioning that the market had been bleeding off for the previous twelve hours.
What was happening: NYSE 13:30 UTC
Switching to Yahoo Finance 5-minute bars for the same window on May 15:
Asset NY open (13:30) Low Low at Δ to low
SPY (S&P 500) $740.17 $737.96 13:45 UTC −0.30%
QQQ (Nasdaq 100) $709.54 $705.55 13:45 UTC −0.56%
GC=F (Gold) $4,546.70 $4,513.80 13:45 UTC −0.72%
CL=F (WTI Crude) $99.41 $99.18 13:35 UTC −0.23%
BTC (Binance perp) $80,301 $78,632 14:00 UTC −2.08%
Three observations:
- Everything sold off together at NYSE open. SPY, QQQ, gold, and oil all hit their session lows within fifteen minutes of 13:30 UTC, all between −0.23% and −0.72%. Whatever drove that move was not a crypto story.
- BTC tracked the direction but with roughly seven times the magnitude. SPY dropped 0.30%, BTC perp dropped 2.08% from the same starting moment. Crypto's leverage stack converted a sub-half-percent equity dip into a 2% perp drawdown.
- BTC's low printed fifteen minutes after the equity low. SPY bottomed at 13:45, then recovered through 14:00. BTC bottomed at 14:00, then recovered through 14:22. That fifteen-minute lag is the cascading-liquidation tail: long positions that needed to be force-closed before the bid could lift.
Total BTC perp volume at the 14:00 UTC candle was $1.03 billion, against a roughly $30 million baseline for the preceding hour. The capitulation print.
This was a macro flow event, amplified by leverage. It was not a crypto-specific reaction to anything announced at the summit.
Day 1 mirrored Day 3
The May 13 drawdown got coded as "market unimpressed by arrival ceremony" in most coverage. The order flow on May 13 looked almost identical to May 15.
Same Yahoo Finance pull for May 13:
Asset NY open (13:30) Low Low at Δ to low
SPY $738.47 $735.47 13:45 UTC −0.41%
QQQ $709.82 $704.83 13:45 UTC −0.70%
BTC (Binance perp) $79,957 $78,730 16:05 UTC −1.53%
Same trigger time (NYSE 13:30 UTC). Same equity-low window (13:45 UTC). BTC bottomed at $78,730, $98 above the May 15 low.
The difference between the two days was how long BTC bled after stocks recovered:
- May 15: BTC lagged the equity bottom by ~15 minutes.
- May 13: BTC lagged by ~2 hours and 20 minutes.
Peak CVD-30m on May 13 hit −$175.8 million at 16:10 UTC, again with OBI holding positive throughout the decline (+0.28 average across the worst hour). Same shape. Slower burn.
Both days, the BTC perp bottomed within $100 of the same level. Both days, that bottom printed in the window that started with NYSE cash open. Neither day's bottom coincided with a specific Trump-Xi headline.
The funding tape, before and after
Perp funding rates show which side of the book is paying for positioning. Positive funding means longs pay shorts (longs crowded, paying for the privilege of being long). Negative means shorts pay longs.
Three relevant readings.
Pre-cascade, May 15 00:00 UTC (13.5 hours before the flush):
- Bybit BTC: −0.0056%, third percentile of its trailing trailing two-year window.
- Binance BTC: −0.0042%, fifth percentile of trailing trailing two-year window.
- Hyperliquid hourly funding had spent five consecutive hours in the bottom seven percent of its trailing two-year window on May 14 evening (17:00-21:00 UTC).
By the time the cascade fired on May 15 afternoon, the market had already been short-positioned through the overnight session. The liquidations that hit the tape during 13:30-14:30 UTC were not over-extended new longs. They were the last remaining longs being squeezed out at the bottom.
Post-cascade, May 16 00:00 UTC onward (after the bid-lift and recovery):
Hyperliquid funding went deeper negative, not lighter. From 00:00 to 14:00 UTC on May 16, every hourly reading printed in the bottom seven percent of the trailing two-year window. The deepest reading was −0.0194% at 10:00 UTC May 16, the 0.9th percentile of all hourly funding samples in the trailing two-year window.
OKX went the other way. The 16:00 UTC May 16 funding cycle printed +0.0100%, the 96.5th percentile of OKX's data window. (Caveat: OKX's per-venue history in our pipeline runs about 95 days, so the percentile is against a shorter base than Bybit and Hyperliquid. Direction is still clear.)
Two reads of the same instrument, two opposite stories. The cascade did not flush positioning into a consensus. It split the market in two.
The cross-venue book during the flush
On the order-book side during the 13:30-14:30 UTC cascade itself, the three venues we track (Binance, Bybit, OKX) showed a consistent split:
Time Binance OBI Bybit OBI OKX OBI Note
13:30 +0.112 +0.002 +0.044 pre-flush
13:35 +0.242 +0.109 +0.061 Binance stacking bids hard
13:45 +0.127 +0.093 −0.045 OKX flips ask-heavy
13:50 +0.022 −0.047 −0.137 all softening
14:00 +0.107 +0.042 −0.075 Binance re-stacks at low
14:05 +0.112 +0.105 −0.044 Binance + Bybit aligned
14:25 +0.079 −0.182 −0.197 post-low, asks return
Caveat: OBI is computed per venue over a price band around mid. The exact band depth and weighting can differ between venues, so absolute magnitudes aren't 1:1 comparable. Direction (sign and trend) is reliable.
What's consistent across all three venues: bids stacked positive on Binance, neutral-to-positive on Bybit, and negative on OKX during the cascade. We won't speculate on the trader-base differences that produce this split, but the directional read is in the data.
What this changes about reading market reaction
The summit happened. Bitcoin moved. The two facts are loosely correlated through the trading calendar. They are not connected through the order book.
When we strip out the macro frame, what's left:
- The trigger was global risk-off at NYSE open, twice. Both Day 1 and Day 3.
- Crypto amplified equities by 4-7x on the full drawdown, not via news flow but via leverage cascade. The first 15 minutes were proportional (1.5x on Day 1). The next two hours weren't.
- Funding had pre-positioned bearish by the time the May 15 cascade fired. The dump didn't surprise the perp market. It confirmed positioning that had built overnight.
- Liquidations were almost entirely one-sided: roughly $55 million of longs against under $1 million of shorts on Binance and OKX combined during the May 15 hour. The cascade was a forced exit, not a bidirectional shakeout.
- Post-cascade, the market split. Hyperliquid funding pushed to top-1% bearish extremes; OKX flipped to top-5% bullish on its 95-day window.
None of this required reading what Xi said about Taiwan, or what Trump said about Boeing's 200-jet order. It required watching the NYSE cash session open at 13:30 UTC.
What to watch next
Xi is scheduled to visit Washington on September 24, 2026. If the BTC perp moves in the surrounding window, three places to look first:
- NYSE cash session timing. If the move starts at 13:30 UTC, the catalyst is macro flow, not the summit.
- Funding rate percentile in the 12 hours before any large move. If Bybit BTC funding drops below the fifth percentile of its trailing window, the perp market is already short-positioned and any further drop will be a forced-long exit, not new selling.
- OBI direction during the decline. Positive OBI during a price drop is passive accumulation. Negative OBI during a price drop is supply meeting bids. The two have different implications for whether the move continues.
The live versions of all three signals are on the funding scoreboard (term structure + percentile across Binance, Bybit, OKX, Hyperliquid) and the positioning quadrant (OBI × CVD). Or via the feed if you'd rather wait for the next write-up.
FAQ
Did Bitcoin crash because of Xi's Taiwan warning during Trump's Beijing visit?
Not in the way the news coverage suggested. On May 14, when Xi reportedly warned Trump on Taiwan, BTC actually rallied from $79,500 to $81,626 (+2.6%) with $99 million of aggressive buying in a single hour around 16:00 UTC. The two BTC sell-offs during the May 13-15 summit window, Day 1 (May 13) and Day 3 (May 15), both started within eight minutes of NYSE cash open at 13:30 UTC, alongside simultaneous drops in SPY, QQQ, gold, and oil. The trigger was global risk-off, not summit-specific news.
What time did Bitcoin actually bottom on May 15 2026?
$78,632 at 14:00 UTC on May 15. The cascade began at 13:38 UTC, peaked in sell-side aggression at 13:52 UTC (CVD-30m at −$216.7 million), and bottomed eight minutes later. Aggressive buyers reversed it almost immediately: by 14:22 UTC, the same rolling CVD-30m metric had swung to +$285 million. The 14:00 UTC candle alone printed $1.03 billion of BTC perpetual volume against a $30 million baseline for the preceding hour.
Why were only long positions liquidated during the May 15 Bitcoin cascade?
Funding rates had already pre-positioned the market bearish by the time the cascade fired. At 00:00 UTC on May 15, thirteen hours before the flush, Bybit BTC funding hit −0.0056% (third percentile of its trailing trailing two-year window) and Binance hit −0.0042% (fifth percentile). Negative funding means shorts pay longs, so by overnight the short side was already crowded. By the May 15 afternoon, the remaining longs were a minority being squeezed out. Binance and OKX combined recorded $55.4 million of long liquidations against $0.8 million of shorts in the 60-minute cascade window.
Did stocks and gold also drop when Bitcoin crashed on May 15 2026?
Yes, simultaneously. From NYSE cash open at 13:30 UTC, SPY dropped 0.30% to its low at 13:45 UTC, QQQ dropped 0.56%, gold (GC=F) dropped 0.72%, and WTI crude (CL=F) dropped 0.23%. The Binance BTC perp dropped 2.08% over the same starting moment, roughly seven times the magnitude. Stocks recovered by 14:00 UTC, but BTC kept bleeding to its low at 14:00 UTC because its remaining longs needed to be force-closed first via the cascading liquidation tail.
What does a negative funding rate signal for Bitcoin perpetual futures?
When BTC perpetual funding is negative, short positions pay long positions at each settlement (every 8 hours on Binance, every 1 hour on Hyperliquid). It signals that shorts are crowded enough to pay for the privilege of being short. On May 15 at 00:00 UTC, Bybit BTC funding sat at the third percentile of its trailing trailing two-year window; on Hyperliquid, the May 14 evening saw five consecutive hourly readings in the bottom seven percent. Both confirmed the market was already short-positioned overnight, well before the afternoon cascade.
What should I watch in BTC microstructure during the next Trump-Xi event?
Xi's reciprocal visit to Washington is scheduled for September 24, 2026. Three signals separate summit-driven moves from macro-driven ones: NYSE cash session timing (if BTC starts moving at 13:30 UTC, it's macro flow), funding-rate percentile in the 12 hours before any large move (below the fifth percentile means shorts are already crowded and any further drop will be a forced-long exit), and OBI direction during a decline (positive OBI is passive accumulation; negative OBI is supply meeting bids).
Methodology and sources
- BTC mid-price and Binance microstructure (OBI, CVD-30m, CVD-2h) from our live trade-flow feed on Binance USDⓈ-M perpetuals, aggregated to 1-minute bars. All BTC prices and percentages in this post are from the Binance perpetual; Yahoo's
BTC-USDaggregate would show slightly different numbers (smaller drawdowns because spot lags perp on cascades). - Bybit and OKX OBI from each exchange's depth WebSocket, aggregated to 1-minute resolution. OBI is computed per-venue over a price band around mid, so the magnitudes are normalized within a venue but not cross-comparable in absolute units. Direction is.
- Liquidation events from Binance, Bybit, and OKX public liquidation streams. Side normalized to "long liquidated" / "short liquidated" convention across all three venues. Our current Bybit normalization disagrees with Binance and OKX for the May 15 window; we treat that as an open data-pipeline question and exclude Bybit totals from the cited cascade numbers.
- Funding history for Binance, Bybit, OKX, and Hyperliquid pulled from each venue's REST funding-history endpoint. Bybit and Hyperliquid archives are populated to a 730-day trailing window per nightly ingest (so the start date drifts forward over time; today the Hyperliquid archive spans 2024-05-15 → 2026-05-18). OKX has roughly 95 days in our archive.
- "Bybit excluded" from the May 15 cascade liquidation totals: our current side-normalization for Bybit's V5
allLiquidationfeed produces the inverse long/short split from Binance and OKX for that window. Until we reproduce the cause (daemon convention bug versus a real venue-specific tape difference), we exclude Bybit totals from the cited cascade numbers rather than report a story we can't defend. - Equity, gold, and oil data via Yahoo Finance 5-minute bars (
SPY,QQQ,GC=F,CL=F). - News-cycle attributions from CoinDesk, news.Bitcoin.com, crypto.news, CryptoSlate, Sunday Guardian, and CNBC published May 13-16, 2026.
Related
- What does a negative funding rate mean? → primer on the funding-rate sign convention used above.
- Cumulative Volume Delta (CVD): the crypto perps trader's guide to order flow → how CVD is computed and what it does and doesn't show.
- Two weeks, six majors, $715M in Binance perp liquidations → broader context on the liquidations dataset used here.