Backtesting Big Trades
By Geoff · · Updated
Do large aggressive trades predict short-term price direction?
We tested big trades in crude oil (CL) and the E-mini S&P 500 (ES) from August 31 – September 4, 2026. A big trade here is one aggressive order of the size below or more, rebuilt from its split fills the same way the Big Trades indicator does.
Short answer: not on their own. Following the aggressor gave no consistent direction in either market. Big trades into thin liquidity tended to reverse, and trading that reversal made money in CL over the five days but lost in ES after commission.
- CL:
- 50+ contracts · 430 trades
- ES:
- 120+ contracts · 525 trades
Moves are in each instrument's ticks: +1 means one tick up, −1 means one tick down. n is the number of trades.
What happens after a big trade?
| Side | Market | n | 10s | 30s | 1m | 5m |
|---|---|---|---|---|---|---|
| Buys | CL | 220 | -1.1 | -0.9 | -0.8 | +1.4 |
| ES | 258 | +0.4 | -0.5 | -0.5 | +1.0 | |
| Sells | CL | 210 | +0.5 | +1.3 | +1.0 | +2.3 |
| ES | 267 | +2.2 | +1.0 | +0.1 | +1.3 |
Following the aggressor did not give a consistent directional signal.
Does range location matter?
We grouped trades by their position in the preceding 30 minute range.
Buys
| Location | Market | n | 10s | 30s | 1m | 5m |
|---|---|---|---|---|---|---|
| Top 20% | CL | 123 | -0.7 | -0.4 | -0.1 | +4.3 |
| ES | 100 | 0.0 | -2.4 | -3.1 | -1.1 | |
| Middle 60% | CL | 68 | -1.9 | -2.4 | -3.1 | -3.4 |
| ES | 125 | +0.2 | +0.4 | +1.1 | +3.4 | |
| Bottom 20% | CL | 29 | -0.8 | +0.3 | +1.4 | +0.3 |
| ES | 33 | +2.3 | +2.2 | +1.3 | -2.0 |
Sells
| Location | Market | n | 10s | 30s | 1m | 5m |
|---|---|---|---|---|---|---|
| Top 20% | CL | 39 | -0.2 | +1.0 | -0.1 | +4.0 |
| ES | 59 | +1.9 | -2.7 | -4.6 | +1.1 | |
| Middle 60% | CL | 90 | +0.2 | +1.5 | +1.3 | -2.8 |
| ES | 148 | +2.1 | +1.7 | +1.5 | +1.6 | |
| Bottom 20% | CL | 81 | +1.3 | +1.3 | +1.2 | +0.8 |
| ES | 60 | +2.7 | +2.7 | +1.5 | +0.9 |
No clear pattern. Large buys do have a tendency to occur at the top of a range.
Does size matter?
We grouped trades by contract size in each market. Q1 contains the smallest qualifying trades; Q4 the largest.
Buys
| Quartile | Market | Contract range | n | 10s | 30s | 1m | 5m |
|---|---|---|---|---|---|---|---|
| Q1 | CL | 50–54 | 55 | -1.1 | +0.4 | +0.5 | +3.0 |
| ES | 120–137 | 65 | -0.2 | -1.7 | -1.3 | +0.2 | |
| Q2 | CL | 55–74 | 55 | -0.6 | -1.4 | -1.9 | -0.7 |
| ES | 138–165 | 64 | +0.6 | -0.7 | -1.6 | -0.4 | |
| Q3 | CL | 75–101 | 57 | -1.3 | -1.5 | -0.6 | +1.5 |
| ES | 166–240 | 65 | +1.6 | +0.6 | +0.3 | +0.9 | |
| Q4 | CL | 102–617 | 53 | -1.4 | -1.2 | -1.4 | +1.8 |
| ES | 241–1344 | 64 | -0.4 | 0.0 | +0.5 | +3.2 |
Sells
| Quartile | Market | Contract range | n | 10s | 30s | 1m | 5m |
|---|---|---|---|---|---|---|---|
| Q1 | CL | 50–54 | 54 | -0.1 | +0.8 | +0.8 | -1.3 |
| ES | 120–137 | 70 | +1.8 | +0.1 | -0.7 | +0.1 | |
| Q2 | CL | 55–74 | 51 | +1.3 | +2.1 | +2.4 | +4.9 |
| ES | 138–165 | 64 | +2.5 | +1.8 | +0.1 | +1.6 | |
| Q3 | CL | 75–101 | 57 | +0.1 | +1.0 | +0.9 | +3.6 |
| ES | 166–240 | 66 | +1.3 | -0.3 | -1.0 | -1.3 | |
| Q4 | CL | 102–617 | 48 | +0.9 | +1.5 | -0.2 | +1.9 |
| ES | 241–1344 | 67 | +3.1 | +2.3 | +2.1 | +4.9 |
Larger trades did not consistently produce stronger follow-through. ES's largest buys were followed by higher prices at 5 minutes, but so were its largest sells.
Does liquidity matter?
We split buys and sells in each market into quartiles by average contracts per level swept. Q1 is the thinnest liquidity; Q4 is the most concentrated.
Buys
| Quartile | Market | Avg contracts / level | Avg levels swept | 10s | 30s | 1m | 5m |
|---|---|---|---|---|---|---|---|
| Q1 | CL | 11.4 | 7.3 | -2.6 | -2.1 | -1.2 | -4.9 |
| ES | 36.3 | 6.1 | -1.6 | -3.0 | -2.8 | -2.5 | |
| Q2 | CL | 16.0 | 5.3 | -1.6 | -2.0 | -3.0 | -0.3 |
| ES | 59.7 | 3.2 | +0.3 | -1.2 | -0.5 | +2.6 | |
| Q3 | CL | 37.0 | 3.4 | -0.9 | -0.6 | -0.7 | +4.3 |
| ES | 97.9 | 2.5 | +0.9 | +0.7 | 0.0 | +1.7 | |
| Q4 | CL | 121.1 | 1.1 | +0.8 | +1.1 | +1.6 | +6.6 |
| ES | 199.3 | 1.3 | +2.0 | +1.6 | +1.3 | +2.1 |
Sells
| Quartile | Market | Avg contracts / level | Avg levels swept | 10s | 30s | 1m | 5m |
|---|---|---|---|---|---|---|---|
| Q1 | CL | 11.7 | 5.9 | +2.3 | +4.0 | +4.2 | +6.5 |
| ES | 38.4 | 7.4 | +1.5 | +1.3 | +0.8 | +1.7 | |
| Q2 | CL | 15.6 | 5.1 | +0.4 | +0.6 | +0.6 | +2.5 |
| ES | 69.1 | 2.9 | +3.4 | +1.8 | +1.7 | +1.2 | |
| Q3 | CL | 27.7 | 4.2 | +0.5 | +0.6 | -0.5 | -0.8 |
| ES | 118.3 | 1.9 | +1.7 | -0.3 | -1.3 | +1.2 | |
| Q4 | CL | 104.1 | 1.0 | -1.1 | 0.0 | -0.3 | +0.9 |
| ES | 219.7 | 1.2 | +2.2 | +1.1 | -0.7 | +1.1 |
Concentrated buys were followed by higher prices in both markets; concentrated sells did not lead to lower prices at 5 minutes.
Big buys and sells that hit thin liquidity tended to reverse. The separation was clearer in CL: ES sell quartiles all finished with similar positive 5 minute moves.
Is there more size than the book shows?
A depth test: a big trade hits a price and more trades there than the book was showing. The extra came from hidden orders or from size that refilled as it traded. We compared each big trade with the size shown at the prices it hit just before it.
Our stored data gives each fill its own timestamp, so for this section a big trade is the run of same side fills before the book next changes, the same rule the Big Trades indicator uses. That reproduces this post's trade counts to within 5%.
Buys
| Size taken vs book | Market | n | 10s | 30s | 1m | 5m |
|---|---|---|---|---|---|---|
| More than shown | CL | 130 | -1.9 | -2.0 | -2.0 | -0.6 |
| ES | 22 | -3.0 | -3.0 | -3.5 | +0.5 | |
| Within what was shown | CL | 90 | -0.4 | -0.1 | +0.3 | +3.8 |
| ES | 237 | +0.1 | -0.8 | -0.8 | +0.3 |
Sells
| Size taken vs book | Market | n | 10s | 30s | 1m | 5m |
|---|---|---|---|---|---|---|
| More than shown | CL | 112 | +1.8 | +2.7 | +2.4 | +4.7 |
| ES | 34 | +1.5 | +1.6 | +0.7 | +3.9 | |
| Within what was shown | CL | 89 | -0.2 | -0.1 | -0.3 | -0.1 |
| ES | 233 | +2.9 | +1.4 | +0.7 | +1.6 |
In CL, the hidden size usually won. Buys that took more than the book showed fell about 2 ticks within a minute, and sells rose 2 to 3. Big trades that stayed within the shown book did not reverse.
In ES, about 9 in 10 big trades stayed within the shown book. The 22 buys that took more than shown fell about 3 ticks within a minute. The sells do not stand apart: ES sells rose afterwards in every group.
Can we trade it?
We turned the thin liquidity pattern into trading rules, exiting after 5 minutes with no stop or target. Both results include commission. The backtest walks every trade rather than every bar; see how to backtest order flow on tick data for how that changes fills.
| Market | Round trips | Wins / losses | Win rate | Net PnL |
|---|---|---|---|---|
| CL | 90 | 53 / 37 | 58.9% | +$1,636.40 |
| ES | 49 | 17 / 32 | 34.7% | -$73.70 |
While ES finished negative, each round trip is $3.80 in commission - without including commissions this would be profitable ($112.50). This could be worth pursuing.
Try and modify the algo yourself
Paste this into Algos for CL or ES. This version adds a 7-tick stop and limits trading to 09:30–15:55 New York time.
// Fades big trades that sweep thin liquidity.
// https://marketbyorder.com/docs/api/algos
interface SymbolParams {
minSize: number
thinThreshold: number
}
const SYMBOL_PARAMS: Record<string, SymbolParams> = {
ES: { minSize: 120, thinThreshold: 45 },
CL: { minSize: 50, thinThreshold: 15 },
}
const DEFAULT_PARAMS = SYMBOL_PARAMS.ES
function paramsFor(sym: string): SymbolParams {
const key = Object.keys(SYMBOL_PARAMS).find(k => sym.toUpperCase().startsWith(k))
if (!key) {
print(`Thin Liquidity Fade: no tuned params for symbol "${sym}", falling back to ES defaults`)
return DEFAULT_PARAMS
}
return SYMBOL_PARAMS[key]
}
const { minSize: MIN_SIZE, thinThreshold: THIN_THRESHOLD } = paramsFor(symbol)
const QTY = 1
const HOLD_MS = 5 * 60_000
const SL_TICKS = 7
const RTH_START = 9 * 60 + 30 // minutes after midnight, New York time
const RTH_END = 16 * 60 // minutes after midnight, New York time
const FLATTEN_BUFFER_MIN = 5
const exchangeTime = new Intl.DateTimeFormat('en-US', {
timeZone: 'America/New_York',
hourCycle: 'h23',
weekday: 'short',
hour: 'numeric',
minute: 'numeric',
})
const sessionByMinute = new Map<number, { clock: number; weekday: string }>()
const sessionInfo = (ms: number) => {
const minute = Math.floor(ms / 60000)
const cached = sessionByMinute.get(minute)
if (cached) return cached
const parts = exchangeTime.formatToParts(ms)
const get = (type: string) => parts.find((part) => part.type === type)?.value ?? ''
const info = { clock: Number(get('hour')) * 60 + Number(get('minute')), weekday: get('weekday') }
sessionByMinute.set(minute, info)
return info
}
const isRth = (ms: number) => {
const { clock, weekday } = sessionInfo(ms)
return weekday !== 'Sat' && weekday !== 'Sun' && clock >= RTH_START && clock < RTH_END
}
const canEnter = (ms: number) => {
const { clock, weekday } = sessionInfo(ms)
return weekday !== 'Sat' && weekday !== 'Sun' && clock >= RTH_START && clock < RTH_END - FLATTEN_BUFFER_MIN
}
let exitTime: number | null = null
onUpdate(({ time, bid, ask, trades }) => {
const pos = orders.position()
if (pos.side !== "flat" && !canEnter(time)) {
orders.flatten()
exitTime = null
return
}
if (orders.working().length > 0 && !canEnter(time)) {
orders.cancelAll()
}
if (pos.side !== "flat" && exitTime !== null && time >= exitTime) {
orders.flatten()
exitTime = null
return
}
if (pos.side !== "flat" || orders.working().length > 0 || !canEnter(time)) return
for (const trade of trades) {
if (trade.side === "none" || trade.size < MIN_SIZE) continue
const levels = trade.levels.length
const cpl = trade.size / levels
if (cpl > THIN_THRESHOLD) continue
if (trade.side === "buy") {
const sl = bid + SL_TICKS * tickSize
orders.sell(QTY, { sl })
print(`FADE SELL: big BUY ${trade.size} swept ${levels} levels (cpl=${cpl.toFixed(1)}) - thin liquidity, sl=${sl.toFixed(2)}`)
} else {
const sl = ask - SL_TICKS * tickSize
orders.buy(QTY, { sl })
print(`FADE BUY: big SELL ${trade.size} swept ${levels} levels (cpl=${cpl.toFixed(1)}) - thin liquidity, sl=${sl.toFixed(2)}`)
}
exitTime = time + HOLD_MS
break
}
})To see these trades on a chart, turn on the Big Trades indicator, set Filter mode to Manual and Minimum size to 50 on CL or 120 on ES.