Does spot lead perpetual futures?

Five-minute trade prices appeared to show a leader. Order-book midpoints cut the effect by about 85%, and no market led consistently across BTC, ETH and SOL.

The direct answer

Neither market led consistently at a five-minute horizon.

When we used last-trade candles, an unusually large spot-versus-perpetual return gap closed by about 0.9 basis points in the next five minutes. That looked like a measurable lead-lag effect.

Most of it disappeared when the same test used synchronized order-book midpoints. Midpoint gaps closed by only 0.10 to 0.15 basis points. The apparent effect was roughly 85% smaller once bid-ask bounce and asynchronous last prints were reduced.

The remaining convergence did not identify a stable leader. In BTC, ETH and SOL, sometimes the lagging market caught up and sometimes the market that moved first reversed. The decomposition changed by asset and by which market was ahead.

Spot can lead a specific event. Perpetuals can lead another. The data did not support the general rule that one venue class reliably discovers price first over the next five minutes.

What "leading" should mean

Two markets moving at the same time is not leadership.

Spot and perpetual futures trade claims connected to the same underlying asset. Arbitrageurs monitor the price difference and can buy the cheaper market while selling the more expensive one. That link keeps the prices close even though the instruments and participants differ.

A useful lead claim requires an order:

1. One market moves first.
2. The other market has not fully moved yet.
3. The lagging market subsequently adjusts toward the first one.

If both move inside the same five-minute bar, the data only shows co-movement. If the gap closes because the first market reverses, the first move was not necessarily price discovery. The market that appeared to lead may simply have printed away from shared value and come back.

That distinction is the center of this test. We measure the next move of the lagging market separately from the reversal of the market that moved first.

Why spot and perpetual markets can disagree

Spot represents immediate asset exchange

A spot trade exchanges the asset for quote currency. Persistent spot demand can reflect buyers willing to hold the underlying rather than only its price exposure.

That makes spot flow important, but it does not make every spot buyer informed. Market makers, treasury transfers, execution algorithms and cross-venue hedges all trade spot too.

Perpetuals concentrate leverage and speed

Perpetual futures make it easy to add directional exposure without moving the underlying asset. They often have more volume, lower capital friction and dense algorithmic participation. News and liquidations can therefore appear in the perpetual tape quickly.

Again, speed is not the same as information. A leveraged market can overshoot, force liquidations and reverse.

Different last trades create a false gap

A candle close is the final trade recorded inside its interval. The spot close may be a buy at the ask while the perpetual close is a sell at the bid. Those two prints can show a price difference even when both order books center on the same value.

The trades may also occur at different moments. One market can print near the start of the final second and the other near the end. A last-trade comparison then mixes price change with timestamp and spread noise.

An order-book midpoint averages the best bid and ask. It is not a perfect fair value, but it removes much of the last-trade side effect. A lead that disappears on midpoints was not a strong price-discovery result.

What a lead-lag setup looks like

Suppose spot rises 0.20% in five minutes while the perpetual rises 0.10%. The spot-minus-perpetual return gap is +0.10 percentage points.

Three things can happen next:

• The perpetual rises while spot holds. That supports spot leading and the perpetual catching up.
• Spot falls while the perpetual holds. The gap closes, but spot mostly reverses. That does not support spot leadership.
• Both move, with some catch-up and some reversal. The lead is shared or unresolved.

The same logic applies when the perpetual moves first. Simply observing that the prices reconverged is not enough. You need to know which leg did the work.

Order flow adds another check. If spot price moves first while spot aggressive flow is also more one-sided, the event has a stronger spot-origin footprint. But flow classification still cannot identify the passive counterparty, and in our test flow alignment did not create a universal leader.

A practical way to read a temporary gap

The study below did not establish a general spot-first or perpetual-first strategy. The midpoint effect was tiny and the leader decomposition was inconsistent. These steps are for diagnosing a live event, not following a validated fixed rule.

1. Compare synchronized prices. Use the same interval and, where possible, midpoints rather than unrelated last trades. A visual gap from timestamps or bid-ask placement is not price discovery.

2. Separate the move from the catch-up. Mark which market moved first, then watch the next interval. Leadership requires the lagging market to move toward the leader, not only the leader to reverse.

3. Check normalized flow on both markets. Compare aggressive imbalance as a share of each market's gross flow. Raw CVD dollars are not comparable when the perpetual trades several times more volume than spot.

4. Check leverage conditions. A perpetual-led move during rising open interest differs from one driven by liquidations and falling OI. Both can lead price briefly, but one adds exposure while the other removes it.

5. Watch the basis and funding response. A widening perpetual premium can show leveraged demand outrunning spot. A gap that closes immediately without a lasting basis change carries less structural information.

6. Use the event's price level as invalidation. If one market breaks a level and the other confirms, the move has broadened. If the first market falls back through the level, the apparent lead failed its simplest test.

How we tested leadership

We built synchronized, non-overlapping five-minute bars for Binance spot and perpetual BTC, ETH, SOL, XRP and DOGE. For every bar we calculated:

spot return - perpetual return

For the midpoint test, each market needed one snapshot in each of five distinct minutes. The first snapshot set the bar open and the last set the close. Spot and perpetual open timestamps had to be within one second, as did their close timestamps. We also required both spot and perpetual flow bars for the same interval. We did not take a median midpoint inside the bucket or use the next available observation after a gap.

At the start of each UTC day, the extreme-gap thresholds were estimated from 90 qualifying prior days, all within the preceding 180 calendar days. The current day never helped set its own threshold.

The top decile contains bars where spot moved farther. The bottom decile contains bars where the perpetual moved farther. For every selected bar, the next five-minute returns were split into:

• the lagging market's catch-up move;
• the first market's reversal;
• total convergence, which is the sum of those two components.

Uncertainty was estimated by resampling qualifying UTC days rather than treating every five-minute bar as independent. A qualifying day needed at least 280 synchronized five-minute price-and-flow bars.

The long-history test uses last-trade candles. It contains 295,759 spot-ahead events and 298,166 perpetual-ahead events across 2,409 UTC days.

The more important robustness test uses synchronized Binance spot and perpetual order-book midpoints for BTC, ETH and SOL from 2024 through March

1. It contains 24,904 spot-ahead events and 24,714 perpetual-ahead events across 559 UTC days.

The trade-close result looked much stronger

Those values are close to 0.9 basis points. With hundreds of thousands of events, the intervals are narrow. A large sample can make a small measurement look extremely certain.

The midpoint control changes the interpretation.

The midpoint result

The midpoint convergence is only 0.10 to 0.15 basis points. The separate lagger and leader components each have day-block intervals that cross zero.

The asset results refuse a simple winner:

• When spot was ahead, BTC convergence came mainly from spot reversing, ETH was nearly even, and SOL came mainly from the perpetual catching up.
• When the perpetual was ahead, BTC and SOL came mainly from spot catching up, while ETH came mainly from the perpetual reversing.

The prices do reconverge slightly. What does not survive is a stable statement about which market discovered the correct price first.

Common mistakes

• Calling same-bar correlation leadership. Simultaneous movement does not establish an order.
• Using unrelated last trades. Bid-ask side and timestamp differences can create a false price gap.
• Counting convergence without decomposing it. The gap can close because the lagger catches up or because the apparent leader reverses.
• Assuming spot flow is always informed. Spot contains hedging, market making and execution noise too.
• Assuming perpetual speed means correctness. Leverage can accelerate both price discovery and overshooting.
• Comparing raw flow magnitudes. Normalize each market by its own gross volume before comparing imbalance.
• Promoting a tiny statistical effect into a trade. The midpoint effect is a fraction of one basis point before costs.

What this study cannot answer

• Five-minute bars can miss leadership that resolves in milliseconds or seconds.
• Order-book midpoints reduce last-trade noise but do not measure executable size or hidden liquidity.
• The midpoint archive covers BTC, ETH and SOL, not every asset in the longer candle sample.
• Binance spot and perpetual markets are one venue family. Cross-venue price discovery can differ.
• The test does not identify which participant or order caused a gap.
• A specific news, liquidation or flow event can have a real leader even though no market leads reliably on average.