🤓 Latest Submissions
Midpoint - the options bill nobody shows you
Retail options traders lose more to the spread than to being wrong. Bryzgalova, Pavlova and Sikorskaya (Journal of Finance, 2023) put aggregate retail options losses at $2.10bn against $6.4bn of trading costs, three times more lost to the toll than to bad trades. If you trade shares, SEC Rule 605 forces your broker to publish an execution-quality receipt. If you trade options, it does not: 17 CFR 242.600 defines an NMS stock as "any NMS security other than an option." One word, and every option is exempt. No options execution data exists to compare yourself against. Midpoint is that missing receipt. I measured 65 real contracts on a live Alpaca paper account by paying the cost myself, one SPY put cost $1 to get in and out of, another $306, and turned it into a tool anyone can use with two numbers off their own broker screen. The second half is an autonomous agent trading under the discipline those measurements imply. It runs a perception-decision-action loop every ~70 seconds, takes permission from the slope of the VIX term structure rather than its level, and puts every proposal through 14 risk gates mapped clause by clause onto SEC Rule 15c3-5. It has judged 651 proposals and refused 623. The original contribution: the agent prices the trades it REFUSED. Every refusal is journalled with the market snapshot behind it, then settled at expiry against what actually happened. Across 138 refusals, saying no was worth +$2,491. Conventional transaction-cost analysis cannot do this, it is built from executed orders and is blind to every trade never made. P&L is $100,013 on the judged account: small, and deliberately so. Over five sessions a skilled strategy and a lucky one are statistically indistinguishable, so I optimised what five days genuinely can measure - cost, and report profit honestly, attributed by the order that caused it. Two findings came out null; one result was published and then withdrawn. Both remain on the site.
4 Sep 2026
