Moontower has released this issue, mentioning the Metagame event held at Lighthaven in Berkeley from November 6 to 8. Moontower readers can enjoy a $50 discount on regular tickets. This issue also shares AI-related thoughts, TIPS investment analysis, crude oil market learning resources, and pre-sale information for "How I Invest Volume 2", which will be shipped on October 20.
Read originalAuthor Kris Longmore shares content from his 72-minute interview on The Algorithmic Advantage podcast, looking back on an early experience at a Sydney traders' meetup where he was unable to answer a proprietary trader's question about his own trading edge. He later joined that proprietary firm as a researcher and partner, and summarized relevant insights after leaving to trade independently.
Read originalThe article points out that data mining and AI-driven 'quant by feel' are identical in essence and philosophy, both attempting to skip the core step of understanding how markets work and profit directly from the market. A genuine trading edge requires identifying the counterparty being profited from, the reason they continue to pay, and the rationale for one's own ability to participate in the trade; such structural edge cannot be identified through backtesting optimization alone.
Read originalDrawing on two academic papers, the author identifies three structural limitations of using LLMs in trading research: First, training data is saturated with trading common sense and lacks mechanism-driven edge thinking. Second, across 35+ models, accuracy in time-series value tracking declines logarithmically as the number of historical updates increases, and neither prompt engineering nor expanding the context window can fix this. Third, outputs from 25+ models on open-ended questions show 71%–82% similarity, and models with different architectures converge to the same common answers.
Read originalThe author notes that AI has dramatically lowered the barrier to developing quantitative strategies, enabling fully coded and backtested strategies to be generated within minutes. However, cheap experimentation amplifies the multiple-testing problem, producing a flood of false backtest results that appear effective but are actually overfit to noise.
Read original
