Mention markets ask whether a speaker will say a word: weekly social-media posts, earnings calls, podcasts, sports broadcasts. The thesis: retail overestimates mentions, and transcript research beats them.
Is there a market?
110–160 events a month with about 15 words each — but the median word trades only about $770 over its whole life. Plenty of sample size, very little capacity.
Is “Yes” overpriced?
Yes, from 5¢ to 80¢. Buying “No” whenever “Yes” traded at 20–50¢ showed +8% to +12% per trade, t ≈ 3–4 after fees. But the edge disappeared in markets with more than $1k of volume, it was negative on single speeches — the case the transcript idea was built for — and live order books had a median spread of 37¢. The historical edge was the mid-point of near-empty books.
The other side of the trade
If retail overpays, someone sells to them. Rebuilding 117,000 historical trades from the resting order’s point of view:
| Maker position | Return on capital at risk | t-statistic |
|---|---|---|
| Selling cheap “Yes” | +7.1% | 5.8 |
| Selling “Yes” during the live event | −42% | −4.2 |
| Buying “Yes” | −16% | −2.6 |
Selling unlikely outcomes paid, positive every month — as long as the orders were gone before the event went live. This became Selling longshots.