Informed flow in prediction markets: what whales actually tell you

Prediction markets are often described as machines for discovering the probability of future events. That reputation rests on a simple idea: people with information and capital will bid prices toward accuracy, and the resulting quote is a usable forecast. Decades of work support the core claim that markets can aggregate dispersed knowledge more effectively than polls alone (Wolfers & Zitzewitz, 2004; Arrow et al., 2008). But the leap from “markets can be informative” to “this large fill is informed” is where most desks get sloppy.

On venues like Polymarket and Kalshi, a whale print can mean conviction - or inventory, hedging, attention chasing, or a market-maker rebalancing. Algomarket’s whale layer exists to make that ambiguity inspectable: size, persistence, venue, and whether multiple accounts lean the same way. This note lays out how we reason about that layer, grounded in public research rather than trader Twitter.

“Prediction markets are the best technology we have for extracting information from people who have it and putting it into a form that is useful for decision-makers.”Arrow et al., Science (2008)

What the literature actually says

Classic surveys argue that prediction markets excel when traders are heterogeneous, incentives are real, and contracts are well-defined (Wolfers & Zitzewitz, 2004). The Iowa Electronic Markets literature, among others, showed that market prices can track election outcomes competitively with polls under many conditions (Berg, Nelson, & Rietz, 2008). Separately, work on interpreting prices warns against reading a quoted probability as a pure belief: risk preferences, liquidity, and wealth effects distort the mapping from beliefs to prices (Manski, 2006).

Market-microstructure theory adds another constraint. Informed trading models (Kyle, 1985) imply that large orders move prices precisely because they may contain information - but they also attract camouflage. Noise traders, hedgers, and market makers coexist with informed agents. Observing a large trade without context is therefore not a signal; it is a hypothesis generator.

A practical reading stack for whale prints

Algomarket does not claim to identify “correct” whales. We organize the desk so you can ask better questions faster:

• Notional before narrative - rank exposure by dollars at risk, not by how loud a market is on social feeds.
• Persistence over prints - prefer accounts and books that reappear across sessions to one-off spikes.
• Cross-venue identity - the same event on Polymarket and Kalshi can host different participant sets; align contracts before comparing flow.
• Resolution clarity - poorly specified contracts weaken the information-aggregation story the literature celebrates (Arrow et al., 2008).

Limits we state plainly

We do not publish win rates for whales. We do not invent how often large traders are right. Prediction markets remain uncertain environments; the product thesis is situational awareness - a clearer map of who is pressing which books - not automated alpha. That humility is deliberate: the same papers that praise markets also document bias, thin liquidity, and misinterpretation of prices as probabilities (Manski, 2006).

For Algomarket, the research implication is operational. Free users get the whale dashboard and trader views to study structure. Pro tools (live flow, exposure rankings) tighten the loop when you need higher-frequency context. In both cases, the literature’s warning stands: treat flow as evidence to weigh, not as an oracle.