Do Polymarket Whales Predict Outcomes? We Backtested It on Real Resolved Markets

Everyone says whales "move markets." Almost nobody checks. We own the dataset that can — every Polymarket wallet trading more than $10,000 in a single position, 1,000,000+ trades since February 2026 — so we ran the test on real, resolved markets and are publishing the number, including the part that is inconvenient for us.

This is a real study on real on-chain data, not a sample. Dataset and method are below; the per-market results are in the downloadable CSV. Run date: 2026-06-08. Window: 2026-02-21 to 2026-06-08.

The headline number

We took every resolved, decisive binary market in the dataset (483 of them: a clear Yes/No winner at resolution) and asked one question: when whale wallets held a net position before the market resolved, did the side they were on win?

Across the 95 resolved markets where whales held a pre-resolution position, the whale-favored side won 93 times — 98%.

That sounds like a crystal ball. It is not, and the honest reading is the entire point of this post.

Why 98% is mostly confirmation, not prophecy

Here is the catch we went looking for, and found. By the time whales are sitting on a position, the market has usually already moved to price it in. We measured the market's own price 48 hours before resolution for each of those 95 markets: the overwhelming majority were already trading at an extreme (above 0.90 or below 0.10). In other words, the whale was holding the side the crowd had already concluded would win. The 98% is real, but it is heavily look-ahead biased — it confirms the obvious rather than predicting the uncertain.

So we ran the hard version of the test: restrict to markets that were still genuinely uncertain (priced between 0.15 and 0.85) at the moment we measured whale positioning. The result is the honest one:

• •Of all 483 resolved markets, only 12 were ever a real contest — markets that started in the uncertain band and later resolved decisively. The rest were lopsided from the start.
• •In the handful of those where whales took a meaningful side, the outcome was a coin flip on a sample too small to call (2 of 3). We checked at 48-hour, 7-day, and 14-day horizons before resolution; every horizon gave the same answer — the clean, uncertain-market sample is too thin to claim a predictive edge.

The honest conclusion: in this data, whale flow is a high-fidelity confirmation and tracking signal — it tells you, with 98% reliability, which side the smart money is sitting on as a market matures — but it is not yet a demonstrated early-prediction signal on genuinely uncertain markets. We are the only ones who can run this test, and that is what it says today.

Why that is still the useful answer

A confirmation signal is not a consolation prize. What the data does support is exactly what apimie is built for:

• •See who is on the winning side, and how hard. 98% reliability on whale-held outcomes means whale positioning is a clean read on conviction.
• •Track concentration before resolution events. When one wallet holds most of a market's whale exposure, you see it (top_whale_concentration_pct) before the move.
• •Follow specific wallets, not vibes. Real addresses, real entry prices, real sizes — the audit trail behind every position.

apimie sells the tracking, transparently. We are not selling a prophecy we cannot prove. When the data can prove early prediction, we will publish that number too — same way, with the CSV attached.

Real positions from the run

Three real, verifiable examples from the resolved set (all in the CSV):

• •Haaland top scorer (2025–26 EPL). Whales held $31,869 net on Yes. Resolved Yes. They were on it while it was still a contest.
• •PSG to win the 2025–26 Champions League. Whale entries averaged 0.58 — a genuine coin-flip price — and the market resolved Yes (0.999). One of the few clean foresight cases.
• •Salah top scorer (2025–26 EPL). Whales held $135,532 net on No. Resolved No. High-conviction confirmation of the obvious.

Method (reproducible)

1. •Resolved set: binary markets past their end date whose latest snapshot shows a decisive winner (an outcome priced ≥ 0.95).
2. •Whale exposure, no look-ahead: for each market, sum signed USDC (buys positive, sells negative) per outcome from whale wallets (is_whale), counting only trades that occurred before a fixed cutoff (we used 48h / 7d / 14d before the market's end date).
3. •Favored side & index: the outcome with the larger net whale exposure; the reversal index is |yes_exposure − no_exposure| / total_exposure.
4. •Uncertainty control: record the market price at the cutoff. A real predictive test only counts markets still in the uncertain band (0.15–0.85) at that moment.
5. •Score: did the whale-favored side match the resolved winner.

The same shape is reproducible against the public actor with the market_whales and market_history actions (Pro tier):

import requests

APIFY_TOKEN = "your_apify_token_here"
ACTOR = "N5u8wVgS1bKzibZdf"

def market_whales(market_id):
    r = requests.post(
        f"https://api.apify.com/v2/acts/{ACTOR}/run-sync-get-dataset-items",
        params={"token": APIFY_TOKEN},
        json={"action": "market_whales", "marketId": market_id},
        timeout=120,
    )
    r.raise_for_status()
    return r.json()

# Net whale exposure per outcome → favored side + reversal index.
# To remove look-ahead, only count this against the market price at the
# time the position was taken (market_history), not the resolved price.

Download the full per-market results (CSV) → — 95 real resolved markets with whale exposure, the favored side, the market price 48h before resolution, and whether the whale side won.

What this won't tell you

A 98% confirmation rate is not a 98% trading edge — you cannot enter at the resolved price. The genuinely predictive question (do whales lead price while it is still uncertain?) is not yet answerable from this window of data with statistical confidence; there simply are not enough contested markets with heavy whale money in it yet. We are collecting forward to answer it properly. Until then, treat whale flow as what the evidence supports: the clearest available read on where conviction sits — see the live data. The free tier is enough to start checking our work.