Estimating Market Shares w/o Sales Data

A Novel Approach using Online Listings and Reviews

Any market analysis starts with understanding the competitive landscape as well as the role and power of main players in that market, whether it is for an antitrust analysis or a regulatory counterfactual analysis.

In the absence of publicly available sales data, calculating market shares becomes nearly impossible.¹ In one such instance, our team produced a novel approach for a federal investigation (by the International Trade Commission) to estimate market shares of home appliance manufacturers in the U.S., using product listings (prices and reviews) on the publicly available websites for Home Depot, Lowe’s, and Best Buy. In this white paper, we discuss the methodology and its strengths and weaknesses.

Online Shopping and User Reviews
There are many reasons why customers increasingly prefer shopping online, including convenience, ability to compare prices and consumer reviews, avoiding time and travel costs, and even avoiding crowds during the pandemic.² According to a 2016 survey conducted by Pew Research Center, “82% of U.S. adults say they at least sometimes read online customer ratings or reviews before purchasing items for the first time, including 40% who say they always or almost always do so.”³

Inspired by these observations, our team explored a seemingly simple but intuitive approach to approximating market shares using online product listings.  For product markets with a significant online presence and a relatively low number of seller platforms, is it possible to use product list prices and the number of customer reviews to estimate sales volumes?

In an ideal world, market shares are calculated either using units sold (quantity) or revenues (price times quantity sold). An illustration of this approach is shown below (per firm per product):⁴

Publicly available online listings contain information, such as list price and the number of reviews, on products sold by competitors in a specific product market. For major retail outlets (e.g., Sears, Best Buy, Home Depot, and Lowe’s, who collectively cover more than 80% of the U.S. home appliance market) we can aggregate such online data by utilizing [Price] x [Reviews] as “estimated sales”, in order to calculate the market shares. The accuracy of this estimation relies on multiple factors, which are discussed at a high level below.

i. How sensitive are the estimation results to the date of data scrape? 
Timing of the data collection can affect results due to price fluctuations and product variety and life-cycles. This can be a significant concern if the market shares rely exclusively on a few products instead of a variety of products. For example, if one product is at the beginning of its sales cycle and another at its end, this approach would bias the older product.⁵ In our case, home appliances cover hundreds of products across multiple manufacturers, and not any single manufacturer relies heavily on a single line item. Therefore, independent of when the price-review data are collected, there would be products in their mature stage and others in their early stage, preventing the methodology from biasing one brand over another due to timing.

CR Consumer Reports Inc., a non-profit research organization, reports that an average home appliance price typically falls slightly but steadily over time during its life cycle: about ~10% off its initial launch price within 12 months of its launch date.⁶ However, this applies to U.S. markets only, where we observe very stable and minimal inflation. For markets experiencing high inflation, this requires careful treatment. Even still, high inflation should not create a bias since the price is only a nominal and relative adjustment factor in this estimation. 

ii. How well the number of customer reviews represents the number of units sold?
To examine how well online user reviews represent quantity sold, we tested the underlying relationship by scraping data from the German online retailer Mind Factory,⁷ which discloses units sold for all of its online listed products along with the price information. Below we display the relationship between the number of online reviews and the total quantity sold for the best selling products.⁸ The hypothesized relationship is evident in the left graph. The graph on the right shows that this relationship becomes linear with higher number of sales (i.e. review percentage of units sold converges to a constant). These strongly suggest that reviews can be an appropriate proxy of the sales amount.

We observe, as suspected, that the relationship between the number of reviews and units sold is highly linear for larger samples, suggesting the number of reviews can be an asymptotically consistent estimator for quantity sold, a much desired statistical property.

iii. Number of outlets selling the products?
Lastly, and very importantly, the majority of the shopping platforms must be covered in the data collection process in order to ensure that a brand that is exclusively sold on a single platform is not biased against. In our example, over 80% of the home appliance market is represented by the top four players (Best Buy, Lowe’s Home Depot and Sears), enabling our approach to cover a majority of the market.

Results
We collected data from the top four retailer giants as described above, and compiled Price x Review based market share estimates.⁹ Then, we compared our estimates to (i) a publicly available estimate, and (ii) a Google Trends based estimation (another useful method commonly used to estimate market shares in product markets) as shown below:

The results indicate that, in the absence of other options, researchers can turn to online product listings to estimate market shares, given the limitations considered in this white paper.

We are in the process of fine tuning this approach for a more rigorous application. Stay tuned for more!