writing — essays on derivatives, RWAs & market structure

The KOSPI roughly doubled between January and late June, then gave a quarter of it back in three weeks. The bubble argument is a valuation argument, and it misses the machine: 97 trillion won of retail buying against the largest foreign exit on record (much of it benchmark formulas, not views), margin loans at highs, and single stock leveraged ETFs on Samsung and SK Hynix that gathered 7 trillion won in a month and must trade A·r·L(L−1) into every close, in the direction of the move. Why the crash runs on the same flywheel as the rally, the L(L−1)σ²/2 decay arithmetic retail holders are ignoring, why SKHY's Nasdaq listing means the duopoly now gets force-rebalanced in two time zones, and the three observables that will show how it resolves.

This month DTC settles its first tokenized Russell 1000 trades; the SEC's crypto-wrapper exemption is delayed indefinitely after the incumbent exchanges pushed back; and equity perps on Hyperliquid cleared billions without waiting for either. By October the same Apple cash-flow claim prints three prices with three rights bundles and three arbitrage pipes. Why the law of one price is really a statement about pipe capacity — ADR premia, the March 2020 ETF basis, GBTC's one-way pipe — why weekend funding on the S&P perp is the market's price for gap risk to Monday's open, and the three observables that will tell you how the unbundling resolves.

AI just made the easy half of quant research free — twenty years of data, the query, the chart, all from a plain-English question. The half that actually loses money, fooling yourself with an overfit backtest, is untouched, and cheap search makes it worse. The math of self-deception: the expected maximum Sharpe of N random strategies (√(2 ln N)/√T), the Deflated Sharpe haircut, survivorship and look-ahead bias, walk-forward and the probability of backtest overfitting, and the net-of-cost reality. Why a research engine's most valuable output is the sentence "there is nothing here" — and the tool I'm building to say it.

Kalshi and Polymarket volume went from $5bn to $24bn a month, the CFTC just proposed rules, and traders keep getting caught on contracts about Maduro, Iran, and takeovers. The fix everyone reaches for is legal. The real problem is structural: a contract that resolves on a single discrete event is the purest adverse-selection machine in markets. The Glosten-Milgrom math on a binary payoff, why equities survive informed flow and event markets don't, why LMSR and order books both fail, why the CFTC's price-discovery line is the wrong cut, and the accuracy–liquidity–integrity trilemma you only ever get two-thirds of.

Momentum, low-volatility, and quality are documented across dozens of markets — then a large fraction of the premium evaporates, or inverts, on the NSE. A math-first account of why: net-of-cost alpha and breakeven turnover, the STT floor that taxes turnover not profit, the square-root impact law under thin promoter-locked float, momentum crash risk, betting-against-beta, and the constraints (price bands, F&O ban periods, borrow) the backtest leaves out. The rule that falls out: in a high-cost market, slow factors win.

The SEC is days from publishing an innovation exemption that will let crypto platforms wrap US public equities into on-chain tokens without the underlying company's consent. The headline reads as deregulation. The mechanism is the depository-receipt model from 1927, on a new ledger. Why the rights-passthrough condition is the load-bearing part, what this changes for DeFi and 24/7 equity trading, and what it doesn't change for private-company synthetics.

Anthropic cleared at $1.4T in secondary. The Anthropic prestock on Jupiter trades at roughly a third of that with 200%+ APR funding, and Hyperliquid pre-launch perps regularly hit four-digit APR. Why these prices price first-day float clearing and forced sellers, not project value — and what the same misreading is already costing on-chain.

SEBI alleged Jane Street made ~₹36,500 crore manipulating Indian weekly index options. The trade is mechanically simple, the math is freshman options, and the reason it worked at this size is a structural quirk of the Indian market.

Tokenized stocks have existed for over five years. They mostly don't work. The structural fixes — primary issuance, real institutional liquidity, dividend pass-through — and what it takes to actually put real assets on-chain.

Dating apps are markets in attention, but they've been designed as if the spread doesn't exist. Adverse selection, toxic flow, and the case for an AI matchmaker — Aria — as a market maker.

On-chain options remain a rounding error in crypto derivatives volume. The reason isn't lack of demand — it's that AMM designs leak capital to informed flow during oracle lag. A look at LVR, endogenous implied volatility, and what to do about it.