Alternative Data Coverage Bias

Coverage bias is what happens when an alternative dataset doesn't actually see the whole population it claims to describe, and that gap isn't random. A credit-card panel skewed toward one bank's customers might overweight a particular income bracket or region; a satellite feed of retailer parking lots misses stores without visible lots, like urban locations; app-download data only sees people who own smartphones and use that app store. Any signal built on top of that panel inherits its blind spots, and the blind spots rarely cancel out — they usually push the aggregate number in a consistent direction.

The danger is that coverage bias is invisible in the data itself. A vendor's estimated "same-store sales growth" for a retailer can look perfectly clean and stable while consistently running high, simply because the panel skews toward the retailer's more affluent, higher-spending customer segment — and that skew won't show up as noise, it shows up as a persistent, hard-to-detect offset from the true number.

For example, a credit-card panel covering 3% of a retailer's US transactions might be treated as a proxy for total sales, but if that 3% is drawn disproportionately from urban, higher-income cardholders, the panel-implied growth rate can run 2-4 percentage points above the true company-wide figure quarter after quarter — enough to flip a trading signal from correctly bearish to incorrectly bullish ahead of an earnings miss.