TradeExpert, a trading framework that employs Mixture of Expert LLMs

Market efficiency & individual edge

• Many see markets as “very but not perfectly” information-efficient: easy arbitrage is rare, but short-term prices are driven by sentiment, PR, and flows, while long-term trends revert toward fundamentals.
• Several argue it’s nearly impossible to tell if an individual outperformer is skilled or just lucky, and that most people should behave as if EMH is true and use index funds.
• Others emphasize structural disadvantages for individuals (no insider info, worse execution, lower capital, higher stress), suggesting the opportunity cost usually makes active trading irrational.
• Some describe the market as irrational or even Ponzi-like, yet still feel compelled to participate because not doing so risks falling behind others who do.

Information, domain expertise & insider edge

• Domain knowledge (e.g., gaming industry) is seen as a potential edge for short-term trades, but multiple comments argue this knowledge is “table stakes,” not a durable advantage—true edge mostly comes from material nonpublic information.
• There’s debate over whether individual specialists can exploit narrow niches; some say yes (especially in small caps or illiquid names), others say most experts cannot systematically monetize their knowledge.
• Insider information is repeatedly described as the only truly durable advantage.

HFT, market structure & scale

• Several commenters stress that modern equity markets are dominated by machines; retail “alpha” is seen as either ignorance or insider trading.
• HFT/market makers mostly seek to earn the spread and manage order flow, not value companies. They dislike “toxic” informed flow.
• Big quant firms’ enduring profits are attributed to infrastructure, cleaner data, privileged order flow, and deep market-microstructure expertise, not simple predictive models.
• Some note profitable edges often exist only at small scale and are not shared publicly.

Valuation, P/E ratios & bubbles

• Long debate over high current P/E ratios: some argue they show prices “unhinged from reality”; others say P/E is a crude snapshot, poor for growth or cashflow-intensive models, and dangerous as a timing tool.
• Several stress that missing a few extreme winners (e.g., high-growth names) can doom active stock-pickers relative to simple indexing.

AI/LLMs, MoE & the TradeExpert paper

• Multiple commenters suspect the framework is a gimmick: one-year backtest in 2023, likely data leakage (model training up to mid‑2023, test set in 2023), unrealistically high Sharpe (~5), and no clear comparison to simple baselines or “boglehead” portfolios.
• Some note the strong contribution from the OHLCV “Market Expert” suggests traditional signals, not LLM “intelligence,” drive results.
• Practitioners say they’ve found AI/LLM approaches at best on par with classic quant/stat-arb methods, but with far higher cost and complexity, and no evidence of a robust, scalable edge.
• MoE terminology is seen as somewhat abused; here it’s closer to the older multi-model meaning than modern load-balanced MoE architectures.