PredictIt is the small-stakes US political market that taught a generation of forecasters what real-money prediction looks like under hard caps and wide spreads. After Kalshi’s regulated depth and Polymarket’s global pools, PredictIt shows how product limits—not only beliefs—reshape expected value, Kelly sizing, and whether a venue can be an execution home or merely a signal.
The PredictIt model
PredictIt grew out of an academic research frame operated with university affiliation and CFTC no-action constraints that shaped its design. Retail traders see familiar YES/NO contracts on US politics—nominations, chamber control, policy props—but with features commercial exchanges often omit: a maximum of about 850 shares per contract per user, a fee on profits rather than only per-trade taker charges, and spreads that can be several cents wide even when the mid looks reasonable.
The platform is optimized for belief discovery and teaching, not for deploying the bankroll sizes serious macro traders use elsewhere. That is a feature for classrooms and a ceiling for professionals.
Why 850 shares changes everything
Kelly logic might suggest putting $3,000 on a six-cent edge. PredictIt says you may hold at most 850 shares—often under $500 notional at typical prices. At 30¢, cap notional is $255; at 55¢, about $468; at 80¢, about $680. The cap binds before Kelly does.
Implication: PredictIt is a signal venue and a small bet venue, not a growth engine for large conviction. Use it to see how retail political flow prices an event; execute size on Kalshi or Polymarket when rules match and depth allows. Cross-venue traders should log PredictIt as information, not capacity.
Order book at retail scale
PredictIt presents a book UI, but microstructure feels different from Kalshi marquee markets. You might see 24¢ bid and 30¢ ask on the same YES—a 6¢ spread—while a 29¢ mid tempts you to overstate edge. Always buy at the ask and sell at the bid when computing EV.
If your model says 35% and YES asks 30¢, EV per share is 0.35 − 0.30 = 5¢. Eight hundred fifty shares cost $255; expected gross profit is about 0.05 × 850 = $42.50 before the profit fee on winning branches. Kelly might recommend 7% of a $10,000 bankroll—$700—but the share cap stops you near $255 regardless.
Selling into a wide spread can erase edge on exit; holding to resolution is often rational when your thesis is slow-moving politics and the ask was fair at entry.
Fees and after-fee economics
PredictIt’s profit fee (verify the live schedule) taxes winners, not losers—unlike per-contract taker fees on regulated exchanges. Win $595 profit on a $255 stake and a illustrative 10% profit fee might take ~$60, netting ~$535 on the happy path; lose and you pay no profit fee on the loss branch. EV spreadsheets must multiply the win branch by (1 − fee rate on profit), not slap a flat fee on notional the way taker schedules do.
Wide spreads and profit fees compound: edge must clear both the ask you lift and the tax on being right. A market that looks +8¢ at mid might be +2¢ at ask and +0¢ after fee on modest win sizes.
Cross-venue gaps that are not arbitrage
Imagine PredictIt YES at 30¢ and Kalshi YES at 24¢. Tempting—but check resolution text, timing, and size. PredictIt might move from 30¢ to 31¢ with $255 while Kalshi still shows 24¢ on $50,000 of depth. Structural arbitrage needs scale on both legs and identical payoffs; PredictIt is almost always the weak leg.
The gap may still inform your forecast—signal without execution—but it is not free money. The arbitrage chapters’ checklist applies: same payoff definition, same resolution source, executable size both sides, fees and clocks aligned. PredictIt fails the size row for any serious arb book.
Automated liquidity and displayed price
PredictIt historically used market-maker subsidy concepts akin to automated liquidity—not identical to Polymarket’s on-chain pool, but the same lesson: your trade moves the price you are reacting to. The AMM-versus-book chapter applies in spirit even when the UI looks like a book: trust executable prices, not tiles.
When PredictIt still wins
For learning binaries cheaply, the cap protects against ruin while you practice reading rules. For niche US primary props PredictIt still lists when larger venues have not caught up, compare spreads-adjusted prices to Kalshi or Polymarket after you read each Rules tab. For publishable small bets in academic or journalism contexts, the cap is a feature. For deploying $20k on a six-cent edge, look elsewhere.
Calibration homework shines here: trade many capped markets, log p versus outcomes, and compute Brier scores. Discovering your model is good while EV after spread is negative is an execution lesson, not a forecasting insult.
Regulatory and platform risk
No-action letters can change; caps and user limits are regulatory design, not bugs. Platform wind-down or rule changes are tail risks—treat cannot trade or cannot withdraw as states in your planning, the same way you treat ambiguous resolution. Do not assume PredictIt is permanent infrastructure for a career bankroll.
Withdrawal delays and delisted markets can strand capital until resolution even when you no longer want the exposure—budget illiquidity as part of venue risk.
Historical role in forecasting culture
PredictIt prices appear in papers, blogs, and election postmortems because the cap kept the venue small but earnest. Treat published PredictIt charts like poll aggregates: useful with metadata on spread, cap, and rules version. When a primary delists, capital and attention migrate—your watchlist should migrate too.
NO leg and complement thinking
Retail traders overweight YES tickets; NO is often the cleaner expression when you believe an event fails. PredictIt’s wide spread hurts both legs—compute EV on the side you actually lift. If YES ask is expensive but NO ask is also expensive, the market may be saying uncertainty is high, not that your YES view is wrong.
Common mistakes
Equating PredictIt mid with Kalshi probability. Stacking ten 850-share correlated primary markets as if they were ten independent bets. Ignoring profit fee on the win branch when comparing to taker-fee venues. Declaring arbitrage from headlines without rules diff. Using PredictIt size to “prove” a thesis you then lever on Kalshi without re-running EV.
Role in the platform module
PredictIt is the same binary math with retail friction and a hard share cage. Edge exists when p minus ask clears fees on the winning path; serious capital belongs on venues covered in the adjacent chapters unless you consciously choose a tiny expressive stake.
Key ideas
PredictIt is real money with training wheels that stop at 850 shares. Spread and profit fee dominate. Signal yes, scale elsewhere. Calibration valuable; bankroll growth usually is not.
Pairing PredictIt with Kalshi in a research workflow
A practical workflow: use PredictIt mid adjusted by half the spread as a quick prior, then require Kalshi ask to confirm before sizing. Log the gap over time; persistent gaps after spread adjustment are institutional, not typos. Your thesis should explain why the gap exists—caps, fees, or different rules—not only “Kalshi is cheaper.”
When the platform itself is the story
Media coverage of PredictIt often treats it as the US prediction market. Practically it is one constrained venue in a growing menu. Prices can diverge from Kalshi because capital cannot arb the gap—that divergence is data about liquidity institutions, not proof that either price is “wrong” in isolation.
Manifold next shows what happens when stakes are play money first—and when creator resolution replaces exchange ops.
Next: Manifold Markets: Play Money to Real Money