Outlet Substitution Bias Estimates for Ride Sharing and Taxi Rides in New York City

The arrival of new merchants poses problems for measuring inflation, and many think the resulting biases in the official statistics are nontrivial. The BLS methods treat identical commodities sold by different merchants as distinct, different goods but to the extent the goods are close substitutes then the CPI will be biased upward by an estimated 0.08 percentage point per year (Moulton 2017).

There have not been many empirical studies to inform these estimates, owing to the paucity of the highly granular merchant-level data required. Studies based on external non-BLS sources have typically used a unit value index that essentially treats goods sold at different merchants as perfect substitutes, a controversial assumption. We also use a unit value index but with a different interpretation: We view a quality adjusted price index as the target and demonstrate that, in our context, the unit value index we calculate may be viewed as an upper bound to this unobserved target.

Using detailed data from email receipts, we find that the arrival and growth of ride-sharing services in New York City likely imparted a nontrivial bias in the official price indexes for that city: a lower bound of 0.5 percentage point per year over the period 2015–2017. We attribute the magnitude of the bias to the sustained growth of ride sharing over this period, from 40 percent of the market in 2015 to 70 percent by 2017.