In today’s age of customization, a good or service’s price highly depends on the options or features a customer is looking for.
The price of a new automobile or airline ticket, for example, can rise significantly depending on the add-ons or upgrades selected.
While it is clear a product’s attributes collectively dictate the final price, it is not always evident how influential each individual feature is. To isolate these impacts, we built a statistical model that breaks down the price contributions by product attributes.
We chose to show wine in this example due to its diverse characteristics and the large amount of data available through Systembolaget, a top 3 wine retailer in the world operated by the Swedish government. However, the same approach can be applied to any good or service.
The interactive tool below demonstrates the results from the model. Adjust the inputs (light gray boxes) for any attribute for up to 3 wines to observe how price changes as you switch between styles, grapes, countries of origin, etc. (refreshing the page will restore all inputs to their defaults). Start at the top with style and work down the list – the tool is setup to not allow unlikely style/grape or package type/size combinations, but review all selected inputs to avoid picking nonsensical combinations.
An explanation of each attribute can be found below the tool.
Style: Whether the wine is a red, rosé, or white
Grape: The type of grape used. Only grapes with enough observations to establish a genuine price impact are included. The rest are called “Other”
Country of origin: The country in which the wine was produced. Only countries with enough observations to establish a genuine price impact are included. The rest are called “Other”
Blend: Whether the wine is a blend or not
Age: How long the wine has been aged. Selecting zero indicates no aging.
Organic: Whether the wine is organic or not
Alcohol percentage: How alcoholic the wine is
Package type: Whether the wine comes in a glass bottle or box
Package size: The size in mL of the wine pack. Note that the tool displays price per liter and not unit price, meaning a smaller pack size will lead to a higher price than a larger pack (as convenience packs are more expensive than bulk packs on a per liter basis)
Cork, cap, or tap: Whether the pack type uses a type of cork, a screw cap, or a tap
Brand size: How big the brand is, based on sales. The observations were sorted into four classifications based on reasonable cutoffs.
To build the model, we:
- Collected 2017 pricing and attribute data from Systembolaget on 459 individual wines
- Tagged each observation with as many defining characteristics as possible – those seen in the tool above were the attributes that had information available for enough observations to run a meaningful statistical analysis
- Ran a regression (R-squared: 0.88), controlling for heteroscedasticity and collinearity
With the results, we were also able to identify the most over and under-priced wines by percentage difference, which are displayed graphically and listed below.
This approach is generalizable. If we were to analyze car prices, we would, e.g., tag engine size, number of cylinders, suspension, leather seats, SUV/van/sedan. If we looked at airline ticket prices we would tag distance traveled, time of day, airport location, seat pitch, baggage allowance, and many more.
For another example using this same approach, see our other post on price premiums of fortified foods.