Opinions
Analysis, tutorials, and essays on prediction markets, causal models, and agent-driven trading.
Information Finance Has Arrived: A Material Map of Prediction Markets in Q2 2026
Combined Kalshi + Polymarket volume hit $66B in just four months of 2026 — already greater than the entire 2025 industry total. Bernstein projects $1T by 2030. Two venues hold 95% of US share. The distribution layer fragmented across nine retail surfaces. AI agents are 30% of Polymarket wallet activity and 3-5× more profitable than humans. The CFTC is the most powerful financial-market regulator no one is watching. The next Supreme Court case is already inevitable. This is not a curiosity layer anymore — it's the second financial-markets layer.
The $3 Trillion Lock-In: Why AI and Geopolitics Are One Correlated Mega-Trade
Hyperscaler capex hit $650B/year. UAE committed $1.4T. HBM sold out through 2026. CoreWeave bonds at 538bps. Iran's ceasefire is structurally required, not contingent. Stack the material data and a single shape comes out — every major actor making the same irreversible bet, at the same time, in the same direction. That isn't a window. It's a window's overshoot.
Market Making on Polymarket: Why Maker Status Cuts Loss Probability by 36 Points — and Why Spreads Persist Anyway
Akey et al.'s most economically significant finding: moving from pure taker to pure maker status reduces the probability of losing money by ~36 percentage points on Polymarket. Resolution-spec risk is why cross-platform spreads persist at 1.5–4.5% and why even Susquehanna and Jump can't fully arb them.
The Maduro Indictment Is the Boesky Moment: Information-Edge Arbitrage Reaches Polymarket
April 25, 2026 produced the first U.S. criminal indictment for prediction-market insider trading. Akey et al. estimate $143M in aggregate anomalous profit across 93,000 markets. The 1986–1988 Boesky-to-ITSFEA arc looks like the structural template for the next three years.
Volatility Arbitrage in Prediction Markets: Why Political Favorites Above 60¢ Are Systematically Underconfident
A binary contract at price p has variance p(1-p). Le 2026 (292M trades) finds political markets underconfident at slopes 0.93-1.83 — meaning a 70¢ contract corresponds to a true probability near 83%. Four systematic vol-arb trades follow from the calibration evidence.
The 1992 Pound Trade Has a $82M Polymarket Twin: Conviction Macro Bets in the Age of On-Chain Liquidity
Soros and Druckenmiller scaled their pound short to $10B in a single September 1992 day. Théo deployed $80M across 11 Polymarket accounts on Trump 2024 and netted ~$82.3M. The mechanic — proprietary signal, conviction sizing, asymmetric payoff — is identical. The instrument is new.
Term-Structure Trading on Polymarket: Bootstrapping Hazard-Rate Curves the LTCM Way
Meriwether's Salomon arb group reportedly produced 80–100% of the firm's earnings on yield-curve inconsistency trades. The same arithmetic applies to nested ceasefire and Fed-decision markets on Polymarket — the most theoretically rich and least empirically explored alpha pocket as of April 2026.
Multi-Outcome Overround on Polymarket: Citadel's First Convertible Trade, Translated to Election Markets
Ken Griffin started Citadel in November 1990 with $4.6M trading converts. Saguillo et al. estimate ~$40M extracted from Polymarket multi-outcome arbitrage. The math is the same: when the components of a composite security misprice relative to the whole, a complete-set trade captures the residual.
Pairs Trading on Polymarket and Kalshi: From Tartaglia's APT Group to Cross-Venue Spread Bots
Bamberger originated single-stock pairs trading at Morgan Stanley in 1982. Tartaglia's APT group reportedly produced $50M in 1987. The three modern PM analogs — within-platform structural pairs, cross-venue spreads, and mean reversion — work mechanically. Settlement-spec divergence is the binary-payoff killer.
Why Distressed-Debt Alpha Doesn't Port: The Longshot Pile Is a Value Trap, Not Apollo's First Trade
Apollo and Elliott built fortunes buying senior claims at 10-30¢ on the dollar. The naive prediction-market analog — buying 1-5¢ longshots hoping for 5x reversals — loses ~60% of capital invested per Bürgi/Deng/Whelan's 300,000-contract Kalshi study. Here is why the playbooks don't transfer.
Greenblatt's Spinoff Playbook on Polymarket: Why Child Contracts Mis-Reprice After Parent Resolution
Joel Greenblatt made 50% gross by buying spinoffs that index funds were forced to dump. The same forced-seller mechanic now appears on Polymarket whenever a parent market resolves and dozens of child contracts have to reprice in sequence — measurable in single-digit minutes against wire reports.
Tail-End Trading on Polymarket: Goldman's 1940s Merger-Arb Playbook, Reborn at 95-Cent Spreads
Buying YES at $0.95–$0.99 after the event has resolved is the same trade Gus Levy formalized at Goldman in the 1940s and Buffett scaled at Berkshire in the 1980s. The deal-break risk is now UMA dispute risk — and it has produced documented eight-figure wipeouts on what looked like riskless spreads.
Search Attention Is the Last Closed Macro
Eight and a half billion searches a day on Google, none of it visible. Why tech people should care about democratizing public search attention — and what the join actually looks like.
The Legislation-Market Gap: Congress Moves Slowly, Markets Don't
When a bill gets referred to committee, the market reprices in minutes. The bill itself might not move for months. That gap is where the edge lives.
LMSR vs CLOB vs Continuous Double Auction: Prediction Market Architectures
Three liquidity models, three trade-offs. LMSR for cold-start, CLOB for active markets, CDA for specific event-driven cases. Why Polymarket runs both at the same time.
Brier Score vs Log Loss vs Quadratic Score: Picking a Calibration Metric
Three proper scoring rules, three different jobs. Brier for public reporting, log loss for ML training, quadratic for cross-distribution comparison. The same five forecasts scored under each.
Kalshi vs PredictIt: What Changed When PredictIt Closed
PredictIt shut down in 2024 after the CFTC withdrew its no-action letter. Kalshi inherited most of the audience but not all of the use cases. Here is the migration map.
Kalshi vs Polymarket: Mechanics, Fees, Regulation, Liquidity (2026)
A side-by-side of the two largest prediction-market venues in 2026. Neither one wins outright. Kalshi wins on legality and tax paperwork; Polymarket wins on fees and listing speed.
Null Data Is Not Missing Data: How to Read EE=null, LAS=null, PIV≈0
The most common bug in new prediction-market traders is treating a null indicator as a system error. Each null state is a positive entry condition for a specific strategy. Stop filing tickets and start reading the field.
The Overround Illusion: Why EE > 0 Isn't Free Money
A new prediction-market trader sees Event Overround at +0.06 and thinks they have found a 6% riskless arbitrage. Almost never. Fees, slippage, and the moving target of the orderbook eat the apparent edge before the order gets posted.
Why Cycle Clustering Is the Most Prediction-Market-Native Indicator
Yield-curve thinking borrowed from bond traders is great. CYC is the indicator no equity desk has ever needed, because no equity has the discrete event-resolution structure that makes term-structure trading on a single underlying possible.
Liquidity Availability Is the Real Edge in Prediction Markets
Every other indicator describes which contract to trade. Liquidity Availability Score describes whether you can trade it at all. Most strategies that look beautiful on paper die at LAS, and that is why LAS is the only edge that matters.
The CRI Test for Stale Positions: When Your Market Hasn't Moved, You're Holding Noise
High Cliff Risk Index plus a flat P&L is not patience. It is the universe telling you the move is happening to other contracts and you are sitting on the wrong one. Exit.
Implied Yield vs Raw Probability: Why Bond-Adjacent Prediction Markets Need a Different Lens
Two Fed-decision contracts at the same mid-price can be wildly different trades. Raw probability hides the difference. Implied yield surfaces it in the unit fixed-income traders already use.
Monitoring the Situation
How three words that mean nothing became the defining phrase of 2025-2026 — from government holding statements to Jeff Bezos memes to a Polymarket pop-up bar where the lights went out.
The Divergence Thesis: When Markets, Sentiment, and Actions Disagree
Every edge in prediction markets comes from disagreement between signal types. Here's the framework for finding them.
Building a prediction market monitoring system: heartbeat architecture for 24/7 edge tracking
Markets move at 3am and your edge decays while you sleep — here is the architecture for a system that never stops watching.
The complete guide to prediction market order types: market, limit, and thesis-informed
How I decide between market and limit orders on Kalshi, and why a causal model changes the math on both.
Prediction market liquidity: why depth matters more than volume for serious traders
Volume tells you how many people showed up; depth tells you whether you can actually trade.
Causal trees for prediction markets: turning macro intuition into tradeable structure
A practical walkthrough of building hierarchical probabilistic models that map directly to binary contracts on Kalshi and Polymarket.
Prediction market edge detection: a practical framework for finding mispriced contracts
Most prediction market traders have opinions but no framework for measuring whether those opinions are worth trading — here is a systematic approach to finding and sizing edge.
Adversarial search: how I try to kill my own thesis before trading on it
The single most valuable feature in my trading system is the one that actively tries to prove me wrong every 15 minutes.
Limit orders on Kalshi: why thesis-informed makers outperform blind spread collectors
The edge isn't in being a maker — it's in knowing where to place the bid before the book tells you.
How I track my macro thesis across 49 Kalshi contracts without checking the screen
A causal tree, 12 edges, and a heartbeat that runs every 15 minutes so I don't have to.
Your prediction market thesis is in your head. That's a problem.
Most prediction market traders carry their thesis as an unwritten feeling — and bleed money when that feeling quietly shifts without them noticing.
The case for agentic market making on Kalshi
Traditional market makers won't touch prediction markets — but thesis-informed agents with catalyst awareness can provide liquidity and profit from it.
Making vs taking in prediction markets: two completely different games
Most traders don't realize they're playing the wrong game — market making and market taking in prediction markets require opposite personalities, opposite edges, and opposite relationships with time.
I automated my Kalshi thesis with a causal tree. Here's what I learned in 3 months.
Externalizing your thesis into a trackable causal structure changes how you think — not just how you trade.
Why prediction markets break traditional quant models — and what works instead
Statistical models that crush equities fall apart on prediction markets — because there's no history, no continuity, and exactly one instance of every event.
Why Prediction Market Orderbooks Are Nothing Like Stock Orderbooks
Every price is a probability. Every order is a belief statement. Every spread is a disagreement about the future. Prediction market microstructure operates on fundamentally different logic than equities — and the traders who understand that difference are the ones extracting alpha.
Prediction Markets Are Underpriced Insurance
If you are long oil equities, buying "Recession YES" at 35 cents is a cheaper hedge than any options strategy your broker will show you.
Why Your Trading Bot Needs a Thesis, Not Just a Signal
Signal-chasing bots lose money in prediction markets because they confuse price movement with probability changes. Here is the fix.
The Case for Automated Market Making on Kalshi
Most Kalshi markets have wide spreads because nobody is making them. That is both a problem and an opportunity.
Prediction Markets vs Polls: 6 Cycles of Head-to-Head Calibration
Six US election cycles, two forecasting methodologies, one Brier score per source per cycle. Markets win on the headline call. Polls win on demographic crosstabs. Neither wins on tail risk.
AI Agents Don't Need More Data. They Need Judgment.
The bottleneck for AI agents in financial markets isn't data access — it's the ability to structure beliefs, track causation, and know when they're wrong.
Prediction Markets Are Already Pricing the Post-Terminal-Value World
Chamath says terminal value is collapsing. Prediction markets never had terminal value to begin with — every contract has an expiry date. They're the native pricing instrument for a short-duration world.
Kalshi API: From Data to Decisions (Not Just Another Wrapper)
Every Kalshi API article stops at "here's how to call the endpoint." This one starts there.
How to Build a Thesis-Driven Prediction Market Strategy
Not how to build a bot. How to structure your thinking about a prediction market bet — from causal tree to executable edge.