# Large exchange deposits do not predict price drops

> When a few large holders move a token onto an exchange, the token does not fall over the next 24 hours any more than it does on an ordinary day.

E877 · Verdict: KILL · Cause: No signal · Market: Crypto & DeFi · Mechanism: Public intent · Horizon: 4h-24h · Tested: October 2026

Source: https://alphanotfound.eu/experiments/E877-exchange-deposits-do-not-predict-selling

On-chain trackers often flag large token transfers to exchanges as a warning that a holder is about to sell. We found 12,794 such deposits in 68 tokens and shorted the token two hours later for 24 hours. The short lost 41.9 bp on average, but shorting the same token at the same hour one week earlier lost 43.4 bp. Compared event by event with its own control, the deposit added +0.8 bp, with an interval from -6.9 to +8.2 bp: nothing.

## Key numbers

- **Deposit events:** 12,794 (68 tokens, Jul 2025 to Oct 2026)
- **Event minus control:** +0.8 bp (paired per event; 95% interval -6.9 to +8.2 bp)
- **Required to pass:** +30 bp (mean net per trade, pre-registered)
- **Worst single trade:** -8,671 bp (a short squeeze)

## The idea being tested

**Belief:** A large deposit of a token to an exchange is a public signal that its holder is about to sell.

**Why people believe it:** A wallet that moves tokens to an exchange deposit address usually does so to sell them. Blockchain transfers are public, while most derivatives traders do not watch the chain, so the selling pressure should arrive after the deposit is visible.

**Why it would persist:** The deposits are small relative to large-cap volume, shorting small tokens is expensive, and a large holder spreads the sale over hours, so the move would be slow and noisy enough to stay unexploited.

## How it was tested

We identified exchange hot wallets from the blockchain itself, recorded every hour in which a token's deposits were large and came from only a few addresses, and simulated a hedged short two hours later. Every result was compared with the same token at the same hour one week earlier, so the test measures what the deposit adds, not what shorting small tokens costs in general.

**Pre-registered:** yes

**Data:** ERC-20 Transfer event logs into exchange hot wallets (Ethereum public RPC node (anonymous); per transfer, aggregated per token-hour; 2025-07..2026-10); Perpetual futures 1-hour candles, quote volume and funding (largest USD-margined crypto perpetual venue, public market-data archive; 1 hour; 2025-07..2026-10)

**Universe:** 77 ERC-20 tokens with a USD-margined perpetual on two large venues; 68 had qualifying events. One contract per token: the one with the most transfers in the discovery sample.

**Signal:** Exchange hot wallets = addresses receiving from at least 20 distinct senders in three sample hours, at least 70% of which sent only to that address (deposit addresses forward everything to one wallet); 6 wallets found, at least two of them belonging to large centralized exchanges. Event = token-hour in which sweeps into those wallets are worth at least 1% of the token's trailing 24-hour perpetual volume and the top 3 depositing addresses account for at least 80% of it. One event per token per 24 hours.

**Execution:** Short the perpetual at the 1-hour close two hours after the event hour (sweep lag, no look-ahead). Hold 24 hours; 4 hours reported.

**Costs:** 5 bp fee plus a modelled half-spread and impact at $5,000, on both sides, plus the hedge leg; funding paid or received at historical rates.

**Controls:** Same token and hour 7 days earlier (decides the verdict). Placebo: hours with the same deposit size but the top 3 addresses below 30% of it (many small deposits). Hedge: BTC beta estimated over the prior 30 days.

**Sample split:** Rule fixed before any data was streamed; first and second half of the period reported.

**Pass criterion:** At least 300 events from 30 tokens; mean net at least +30 bp; token-clustered 95% interval above 0; median above 0; both halves above 0; placebo and 7-day control each below half the event mean.

**Kill criterion:** Mean below +10 bp, or the interval includes 0.

**Power:** Hedged 24-hour standard deviation about 700 bp; about 1,000 effective events give a minimum detectable effect of about 60 bp.

## Result

**Why it was rejected:** The event and its 7-day control lose the same amount, so the deposit carries no information about the next 24 hours.

**Observations:** 12,794

**Net mean:** -41.9 bp

**95% interval:** -48.9 bp to -35 bp

**Median:** -8.7 bp

**Worst:** -8,671 bp

- Shorting after a deposit lost 41.9 bp on average over 24 hours (4-hour hold: -35.8 bp).
- The same short one week earlier lost 43.4 bp. Paired over the 12,779 events that have a control, event minus control is +0.8 bp (interval -6.9 to +8.2 bp).
- Bigger deposits do not help: by deposit size, the three thirds lose 42.5, 47.0 and 36.3 bp.
- Both halves of the period lose (-35.1 and -48.8 bp); 48.5% of trades were profitable.
- The worst 1% of trades equal 51% of the net loss of all trades combined: shorting small tokens carries squeeze risk.

### The deposit adds nothing (bp)

Mean 24-hour net return of the short with its 95% token-clustered interval. The event and the same hour a week earlier are indistinguishable; their difference sits on zero, far below the +30 bp the test required.

|   | Mean | 95% low | 95% high |
| --- | --- | --- | --- |
| After a large deposit | -41.9 | -48.9 | -35.0 |
| Same hour, 7 days earlier | -43.4 | -49.2 | -37.8 |
| Event minus control | 0.8 | -6.9 | 8.2 |
| Placebo: many small deposits (n 122) | 17.0 | -48.7 | 89.3 |

### No dose response (bp)

If deposits signalled selling, the short should do better after larger deposits. Every size third loses as much as its own control.

|   | After deposit | 7 days earlier |
| --- | --- | --- |
| Smallest third (1.0-1.4% of volume) | -42.5 | -43.5 |
| Middle third (1.4-2.5%) | -47.0 | -47.4 |
| Largest third (above 2.5%) | -36.3 | -39.3 |

### Same distribution, deposit or not (% of trades)

Share of 24-hour short trades by net return in bp. The two shapes overlap almost exactly: wide, centred slightly below zero, with a long loss tail from squeezes.

| Bin | After deposit (count) | 7 days earlier (count) |
| --- | --- | --- |
| < -800 | 425 | 411 |
| -800..-400 | 1083 | 1187 |
| -400..-200 | 1645 | 1683 |
| -200..-100 | 1501 | 1447 |
| -100..-50 | 878 | 898 |
| -50..0 | 1052 | 992 |
| 0..50 | 1059 | 982 |
| 50..100 | 949 | 1007 |
| 100..200 | 1513 | 1488 |
| 200..400 | 1707 | 1679 |
| 400..800 | 828 | 849 |
| > 800 | 154 | 156 |

## Interpretation

**Why it failed (interpretation, not established):** Two explanations fit the data and both leave nothing to trade. Either the holder sells before the deposit is swept into the exchange hot wallet, so the visible transfer lags the sale, or the transfers are exchange and market-maker inventory moves that were never sales. The flat dose response points the same way: a larger deposit is not a larger sale.

**Lesson:** The raw event average measures the trade, not the event. Shorting these tokens for a day loses about 42 bp after a deposit and 43 bp on an ordinary day, through cost, funding and drift; only the difference, +0.8 bp, says anything about deposits. An event study that reports the event average alone, without a same-asset, same-hour control, describes the cost of shorting small tokens and attributes it to the event.

**Scope:** Covers deposits into six heuristically identified exchange hot wallets, ERC-20 tokens only, hourly resolution, July 2025 to October 2026. It does not cover other exchanges' wallets, stablecoin flows, other chains, or timing finer than one hour.

## Reproducibility

**Steps:** 1) Sample three hours of Transfer logs per token; label recipients with at least 20 distinct senders and at least 70% single-destination senders as hot wallets. 2) Stream Transfer logs into that wallet set for the period and aggregate per token-hour: swept USD, distinct deposit addresses, top-3 share. 3) Apply the event rule with 1-hour perpetual data; short at h+2h for 24 hours with a 30-day BTC-beta hedge and the stated costs. 4) Compare with the same hour 7 days earlier and the dispersed-deposit placebo; bootstrap by token.

**Grade:** COMPLETE

---
AI-generated research, checked but not peer-reviewed. Historical, experimental research, not investment advice. Dataset Terms of Use: https://alphanotfound.eu/terms