Insights
Notes on market structure, regulation and quantitative research.
The CFTC paused compute futures. The filing shows what it's weighing.
CME's GPU futures won't list on 5 October. The CFTC extended its review to 9 November. The filing shows how the contracts settle and what is still unproven.
Compute gets a price. Which one isn't settled yet.
Silicon Data publishes two H100 prices, 2.8x apart, and the CME compute futures spec does not say which one settles the contract. Why the difference matters.
The AI buildout gets riskier if it works.
Alphabet posted its first negative free cash flow quarter since its IPO. Three forces now act on hyperscaler accounts, and none of them needs weaker demand.
The CLARITY Act is stalling. The rules are arriving anyway.
Kalshi prices a Senate vote on the CLARITY Act at 92% and the bill becoming law by January at 19%. Two other US crypto rules land before either.
A number you can't explain is a number you can't check.
If a backtest returns values you cannot account for, the cause is often a scale error upstream: finite, in range, and invisible to every automated check.
What a trade actually costs.
Fees are the smallest cost of a crypto trade. Half-spread, funding at its real interval, market impact and adverse selection decide whether a strategy works.
Eight strategies, zero edges: an honest statarb post-mortem.
Eight market-neutral crypto strategies, backtested with full trading costs. None survived. Three false edges, one at +604%, and the checks that exposed them.