Code creates a dashboard that showcases the power of Conjoint Analysis for the Electric Vehicle Lease Market (as at January 2020) in San Francisco. The video below gives a walkthrough of the dashboard.
In January 2020, the following lease offers were available on Electric Vehicles in San Francisco (source: https://electrek.co/best-electric-vehicle-leases/).
The lease details are contained in the fields monthly_cost, upfront_cost and term. The remaining fields concern specifications of the electic vehicle (e.g. range, Sedan or SUV). The relative popularity of the vehicles can be found here: https://insideevs.com/news/343998/monthly-plug-in-ev-sales-scorecard/
The central piece of data required for the analysis is a conjoint study. Such a study would collect a group of individuals and present them with a series of products. The respondents would either be asked to choose a single product from the group or rank them in order of preference. By so doing, we seek to understand individual preferences and the relative importance of different attributes using a series of experiments.
By performing statistical inference, one can infer the priorities of each individual (which attributes are particularly important to them) as well as the utility they gain from each level of each attribute.
Given a person's responses, one can subsequently predict which product a person will buy. If the person is rational and the conjoint accurately reflects her preferences, she will choose the product that maximizes her utility. Given the utility scores for three products, U1, U2 and U3, a traditional conjoint predicts a person will choose product 1 with probability exp(U1) / [exp(U1) + exp(U2) + exp(U3)].
As there is no publicly available conjoint data on electric vehicles, user preferences were simulated for the various features in the above table. Here are the simulated preferences on monthly cost for three individuals:
All three individuals have monotonically decreasing utility (i.e. they prefer paying less each month). Person B is the most price sensitive as she has the largest range of utility scores for monthly_cost. For all three people, the sum of the utility scores will be zero. This will be true for any attribute.
We examine the same user's utility for the Electric Vehicle range.
We see all three prefer EVs with larger ranges with Person A ascribing the most importance to vehicle range. We can also measure utilities on categorical variables. A popualar measure would look at the utility people get from various brands.
Persons A and B strongly prefer Tesla to other brands. Person C does not place much importance on the brand of the Electric Vehicle, but weakly prefers Nissan over Tesla.
Utility Scores are simulated for the 10 attributes above for 560 individuals. The individuals are partioned into three segments: Millenials, Gen X and Boomers. The utility scores are drawn from slightly different distrbutions for each segment.
The dashboard has two tabs:
- the first shows the relative attractiveness of the thirteen EVs to the 560 respondents
- the second enables one to run "what-if" scenarios examining the potential financial and market profile implications of adjusting the terms of the lease
We can use conjoint analysis to predict the EV model each respondent will purchase.
We see that Tesla is the most popular brand, chosen by ~60% of the respondents. We can edit lease details using the table on the left, and all the figures in this tab will update to reflect the new lease conditions.
The heatmap below shows the strength of attraction of each model's customers. Diagonal elements reflect how strongly the customer prefered their choice to anything else on the market. Tesla, Model X customers are particularly happy with their choice. The off-diagonal values show the relative attractiveness of other EVs to each of the various EV customers. For example, some of the VW, e-tron customers could be persuaded to switch to a Tesla, Model X if the deal was sweetened.
The right scatter plot shows those respondedents that purchased the EV corresponding to the row which was clicked on. The x-axis shows the probability they purchase the row EV whilst the y-axis shows the relative utility of the EV corresponding to the column that was clicked on compared to the row EV.
The distributions below are for the four groups of users predicted to purchase Chevrolet, Nissan, Tesla and Toyota EVs.
For each group, the distribution is of the probability that they purchase the vehicle in question. We see those predicted to purchase Tesla EVs are the most likely to indeed do so with Nissan's predicted customers being the next most likely to purchase Nissan EVs.
For each EV brand, we can focus on the respondents predicted to buy their vehicles and see how attractive they find their predicted EV-of-choice versus the other seven brands on offer.
We see that the Nissan brand order of threat from other brands is Tesla and Chevrolet as these distributions are the most to the right after the Nissan distribution.
The dashboard provides a flexible interface enabling one to modify the values of an attribute for a chosen set of products and predict what impact this has.
Here, we choose to view the impact of increasing the monthly installment of the Model 3 and Model S by upto $100 per month whilst also increasing the the monthly installment of the Model X by upto $200 per month.
Given our predictions of how customer preferences change with the increasing monthly installments, we can calculate the impact on Key Financial Measures such as Number of Customers, Revenue, Average Revenue per Customer and EBIT.
We can also see how customers move between brands due to the increase of Tesla's monthly installments.
We can zoom still further by looking at the net position by Segment for a particular monthly installment.
We can choose to modify two attributes simultaneously to see what may be the optimum outcome for a particular EV maker.
