Hudson River Trading shifted from handcrafted features based on human intuition to training models on raw, internet-scale market data. This emergent approach, similar to how ChatGPT is trained, has entirely overtaken traditional quant methods that relied on simpler techniques like linear regression.
Top trading firms like HRT actively keep the memory of catastrophic failures, such as the Knight Capital algorithm malfunction, alive. They hold regular talks on "The Knightmare" to instill a culture of deep paranoia and defense-in-layers, prioritizing operational risk management over a "move fast and break things" mentality.
Demanding interpretability from AI trading models is a fallacy because they operate at a superhuman level. An AI predicting a stock's price in one minute is processing data in a way no human can. Expecting a simple, human-like explanation for its decision is unreasonable, much like asking a chess engine to explain its moves in prose.
Contrary to the hype around alternative data, the most crucial input for intraday trading AI is standard market data feeds from exchanges. This raw, high-volume data on quotes and trades is the truest expression of market intent, far outweighing the predictive value of news or social media feeds.
AI models can predict short-term stock prices, defying the efficient market hypothesis. However, the predictions are only marginally better than random, with an accuracy akin to "50.1%". The profitability comes not from magic, but from executing this tiny statistical edge millions of times across the market.
The primary constraint for scaling high-frequency trading operations has shifted from minimizing latency (e.g., shorter wires) to securing electricity. Even for a firm like Hudson River Trading, which is smaller than tech giants, negotiating for power grid access is the main bottleneck for building new GPU data centers.
The "golden era" of big tech AI labs publishing open research is over. As firms realize the immense value of their proprietary models and talent, they are becoming as secretive as trading firms. The culture is shifting toward protecting IP, with top AI researchers even discussing non-competes, once a hallmark of finance.
High-frequency trading firms are expanding into medium-frequency horizons (days to weeks). They use their sophisticated short-term AI models, which can predict optimal prices within the next hour, to inform the execution strategy for their longer-term positions, creating a cascading effect where intraday precision enhances multi-day trading performance.