When a whale dumps a favorite
A whale selling a 70–97¢ favorite should know something, and historically the dumped favorites did fail more than their price implied. Buying the other side at live prices lost money in both lanes.
The hypothesisWhale BUYS are mostly noise — but a whale SELLING a heavy favorite at 70–97¢ is informative, because dumping a near-certain winner forfeits guaranteed money unless the seller doubts the certainty. When it happens, buy the opposite side cheap.
The test
The mine came first: every resolved large SELL print, with sell-side settlement semantics carefully verified against the ledger (a subtle trap — a SELL "wins" when the sold outcome loses). Favorites dumped in the 70–97¢ band subsequently failed at 16.7% (non-sports) and 18.3% (sports) against price-implied rates of 12.6–14.6% — a persistent gap on $16.7M of prints. But the monthly cut showed the sports version of the signal died mid-season, so the live probe runs the two lanes separately and never pools them. Live rules: qualifying dumps trigger a paper-buy of the opposite token at its live ask, gated on price, spread, and a maximum premium over what the dump itself implies, settled net of fees.
The result
At our August update both lanes looked strongly positive (+125% and +36%) on 24 and 60 settlements. Both have now passed their gates, and both lose.
Non-sports: 166 settled signals on 94 markets. Counted per signal, the lane shows +58% — but that figure is an artifact: a few winning markets produced many signals each (one whale dumping in several prints), so their wins were counted many times. Counting one position per market, the opposite side won in 12 of 94 markets (12.8%) at an average price of 13.8¢ — −18.3% ROI at the ask. By month: July +104% (14 markets), August −31% (41), September −72% (26), October −4% (13).
Sports: 230 markets, 32 wins (13.9%) at an average price of 14.6¢ — −22.3%.
Conclusion
Negative. The historical pattern did not show up in live trading: the favorites we bet against failed about as often as the prices we paid implied, slightly less in the non-sports lane. It is also a lesson in counting. The per-signal average said +58%; one position per market said −18%. One informed seller printing ten fills is one bet, not ten.