Quiz 1 Question 13 of 20

An HFT firm trains an ML-based execution model on two years of historical equity data sourced from the Securities Information Processor (SIP) consolidated feed, which aggregates best-bid/best-offer quotes across all exchanges with approximately 2–5 ms latency. The firm then deploys this model in production against a direct proprietary exchange feed with sub-100 microsecond latency. After deployment, the quant team observes that the model's input feature distributions at inference time diverge substantially from those observed during training, and realized alpha is 60% lower than backtest projections. What is the PRIMARY cause of this performance degradation?

Select an answer to reveal the explanation.

Motivation