Does a stacked bid side actually mean support?
Bid-heavy books predicted higher prices across five Binance perpetual markets, with a spread of 9.4 points that survived a block bootstrap. The gross return still stayed below taker fees.
The direct answer
Yes, on average. A bid-heavy book was followed by a higher price more often than an ask-heavy book in all five Binance perpetual markets we tested.
BTC gives the clearest result. In non-overlapping hourly observations, price was higher one hour later 46.08% of the time after the most ask-heavy decile and 55.52% after the most bid-heavy decile. The 9.43 percentage-point spread has a standard error of 1.28 points, or a t-statistic of 7.4. A day-level block bootstrap, which lets hours inside the same day move together instead of assuming they are independent, gives a 1.22-point standard error and a 95% interval from 7.1 to 11.9 points. None of the 2,000 resamples produced a negative spread.
That is strong evidence of a directional relationship. It is not evidence of a tradeable strategy. The most bid-heavy BTC decile earned only 0.35 basis points on average before fees, funding and slippage. A Binance taker fee is roughly 4 to 5 basis points per side, depending on tier.
The largest bid walls do not extend the pattern. The most bid-heavy 1% was followed by a higher price 52.81% of the time, against 54.13% in the four percentile points immediately beneath it, and its average one-hour return was negative. That tail holds only 303 hours, and the gap between the two bands is not separable from noise. The honest reading is not that huge walls are worse. It is that the relationship stops strengthening well before the extreme, which is already at odds with treating wall size as a measure of conviction.
The useful lesson is narrower than "more bids means buy." Imbalance tells you where displayed liquidity is concentrated. Whether that liquidity remains when price reaches it is a separate question.
Picture the order book
The order book is a price ladder of unfilled limit orders. Bids sit below the market and offer to buy. Asks sit above the market and offer to sell. The best bid and best ask form the inside market, while deeper orders show how much displayed liquidity waits farther away.
A stacked bid side means more size is visible below the market than above it over the depth range being measured. In this study, the range is one percent on each side of the mid price:
skew = (bid depth - ask depth) / (bid depth + ask depth)
A positive value is bid-heavy. A negative value is ask-heavy. A balanced book sits near zero.
This is displayed intent, not completed trading. A bid can trade, move or disappear before price reaches it. The order book shows what participants are currently offering to do under current conditions.
What imbalance looks like on a chart
On a depth ladder, a bid-heavy book appears as larger quantities beneath the current price. On a heatmap, it appears as brighter or more persistent bands below price than above it.
Do not confuse a broad imbalance with one wall. A single large order at one price can dominate the eye while the rest of the bid side remains thin. A depth band combines many levels and answers a different question: how much displayed liquidity exists across the nearby region?
Time matters as much as size. A wall that persists through repeated trades is different from one that appears only while price is far away. A useful chart therefore shows the history of the liquidity, not just the latest snapshot.
When price finally reaches a bid, three outcomes are possible:
- It holds. Sell orders trade into the bid and price fails to continue lower.
- It is consumed. Sellers trade through the displayed quantity and price continues.
- It is pulled. The bid is cancelled or moved before meaningful trading reaches it.
The snapshot looks supportive in all three cases. Only the test reveals which one was real.
Different actions create the same skew
A positive imbalance can grow because bids were added, because asks were removed, or both. The formula does not distinguish them.
Those paths can reflect different conditions:
- A participant may genuinely want to buy below the market.
- A market maker may quote more size on one side while managing inventory.
- Ask liquidity may retreat during an upward move, making the bid side look larger without any new bid demand.
- A large displayed order may be cancelled as price approaches.
- One venue may look bid-heavy while liquidity on other venues points the other way.
The book can show the change. It cannot show the owner's motive.
A practical way to read imbalance
The historical study below tests hourly averages, not this discretionary framework. It found a directional relationship but no gross return large enough to cover ordinary taker fees. The steps here are for interpreting a live book, not a strategy validated by the table.
1. Define the book you are reading. Name the venue, market and depth band. "The book is bid-heavy" is incomplete if one exchange or one narrow level is doing all the work.
2. Identify what changed. Check whether bids were added, asks were removed or price simply moved into a different part of the book. The same final skew can arrive through each path.
3. Track persistence. A level that remains through several approaches carries more information about current quoting behavior than a wall that flashes once. Persistence still does not guarantee it will trade.
4. Watch the test. When price reaches the level, separate holding, being consumed and being pulled. Completed trades and the price response matter more than the size that was advertised beforehand.
5. Compare the tape and other venues. A bid that absorbs aggressive selling without losing price is different from a bid sitting untouched while the wider market moves elsewhere.
6. Check the economics. A directional hit rate can be statistically real while the average move remains too small for fees and slippage. The measured payoff, not the percentage of winning observations, decides whether the edge is usable.
What the historical sample showed
We used QuantumFlow's first-party depth capture for Binance perpetuals from 2023-01-01 through 2026-08-01. The source records depth at least every 30 seconds, and considerably more often since the start of 2026. We averaged the one-percent bid and ask skew inside each clock hour and used the final mid price in that hour. The outcome is the mid price exactly one clock hour later.
Hours without a matching next clock hour were excluded. This leaves 30,359 BTC observations. Because each outcome window occupies a separate hour, the sample does not count the same one-hour move twelve times.
BTC perpetual, sorted by hourly skew:
| decile | hours | avg skew | higher 1h later | mean 1h return |
|---|---|---|---|---|
| 1, most ask-heavy | 3,036 | -26.8% | 46.08% | -0.66 bp |
| 2 | 3,036 | -12.6% | 46.31% | -0.68 bp |
| 3 | 3,036 | -7.4% | 48.95% | +1.43 bp |
| 4 | 3,036 | -3.2% | 48.52% | +0.19 bp |
| 5 | 3,036 | +0.4% | 50.10% | +0.63 bp |
| 6 | 3,036 | +3.8% | 51.78% | +1.76 bp |
| 7 | 3,036 | +7.4% | 52.50% | +0.16 bp |
| 8 | 3,036 | +11.5% | 52.80% | +1.96 bp |
| 9 | 3,036 | +16.9% | 53.10% | +0.59 bp |
| 10, most bid-heavy | 3,035 | +28.2% | 55.52% | +0.35 bp |
The hit rate is almost ordered from one side of the distribution to the other. The mean return is not. That distinction is why a directional relationship can be statistically clear and financially weak at the same time.
The same comparison across five markets
| market | usable hours | ask-heavy decile | bid-heavy decile | spread |
|---|---|---|---|---|
| BTC | 30,359 | 46.1% | 55.5% | +9.4 pp |
| ETH | 30,384 | 47.0% | 55.4% | +8.5 pp |
| SOL | 30,309 | 48.3% | 54.0% | +5.7 pp |
| XRP | 30,334 | 50.6% | 53.5% | +3.0 pp |
| DOGE | 30,355 | 51.1% | 51.6% | +0.5 pp |
The ordering is positive in every market, but the markets are not independent replications. They trade through the same crypto regimes, and asset identity is mixed together with volatility, tick size, depth and participant composition. The table establishes the ordering we observed. It does not establish why BTC and ETH show a larger spread than DOGE.
The pattern stops strengthening at the extreme
The hourly percentile tails show something the deciles hide:
| band | hours | avg skew | higher 1h later | mean 1h return |
|---|---|---|---|---|
| most ask-heavy 1% | 304 | -64.9% | 44.41% | -6.59 bp |
| next 4% ask-heavy | 1,216 | -27.3% | 46.96% | -0.60 bp |
| middle 90% | 27,324 | +2.1% | 50.61% | +0.71 bp |
| prior 4% bid-heavy | 1,212 | +30.7% | 54.13% | +0.66 bp |
| most bid-heavy 1% | 303 | +44.4% | 52.81% | -0.21 bp |
The extreme ask-heavy tail remains strongly directional. The extreme bid-heavy tail flattens instead.
Read those two tail rows carefully, because each holds only about 300 hours. A day-level block bootstrap puts the gap between the most bid-heavy 1% and the band beneath it at -1.3 percentage points, with a 95% interval running from -7.8 to +5.2. That interval contains zero with room to spare, so this is not evidence that the biggest walls are worse than the merely large ones. What survives is the flattening itself: the steady climb visible across the deciles does not continue into the top percentile.
Large displayed bid size may be genuine, temporary, defensive or intended to influence other participants. This test cannot distinguish those explanations.
Book imbalance is only readable if you can see the whole book over time rather than a snapshot. QuantumFlow renders depth as a heatmap alongside the tape, so resting liquidity is visible as it builds and disappears.
Common mistakes
- Treating displayed size as a commitment. Orders can be cancelled or moved before they trade.
- Calling one wall the whole book. A single price level and a broad depth imbalance are different measurements.
- Ignoring how the skew changed. Added bids and removed asks can produce the same final number.
- Reading one venue as the market. Crypto liquidity is distributed across exchanges and instruments.
- Looking only at a snapshot. Persistence, pulling and consumption require a time history.
- Confusing hit rate with profitability. The strongest BTC decile still averaged less than one basis point before costs.
- Assuming a bigger wall is a stronger signal. The relationship stopped strengthening before the extreme tail, and that tail is too small to resolve on its own.
What this study does not establish
- The test uses depth within one percent of mid. Other bands may behave differently.
- The cross-market ordering does not identify a liquidity-provision mechanism.
- The extreme-tail bands hold about 300 hours each, too few to resolve small differences, and the test cannot say whether spoofing, defensive quoting or incoming flow shaped them.
- Nothing here predicts whether the wall visible in front of you will hold.
---
*Written by Dom at QuantumFlow. Published 5 August 2026. Figures use QuantumFlow's first-party Binance perpetual depth capture from 2023-01-01 to 2026-08-01.*