Another in our series exploring the particular challenges of doing econometric modelling in certain sectors. Guess what sector we’re covering this time?

“Our market’s a bit different…”

Ask any econometrician of any experience, and the chances are they will have heard this, and more than once: our market is different. Of any category from bread to booze. And, whilst every category does, indeed, have its unique characteristics and quirks, what separates them is less important than what unites them, at least from a modelling point of view.

…Some really ARE ‘a bit different’…

In the case of the automotive sector (‘cars’, to the masses) it really is true. Modelling cars IS a bit different, and those differences are due to the peculiar nature of the purchase.

Consider the following. For most of us, a car is:

  • The biggest purchase we make after a house
  • Often bought or leased with some form of credit
  • Consciously, or unconsciously, the most public brand statement we will ever make
  • An infrequent purchase…
  • …influenced by many stakeholders (aka the family)
  • Very physical – we want to touch it before we buy
  • Pretty much the only thing we own with a significant residual value i.e. we usually sell it on when we’re done with it

From the supply side, models have longevity but change significantly over time – a Ford Fiesta really isn’t what it used to be. Also, fulfilment can be long and complicated.

…with important implications for econometric modelling

All of which presents us with significant challenges when building econometric models for car sales.

Here are some we’ve met, and have had to overcome, in order to successfully build econometric models in this tricky sector.

What cars are we interested in?

This seems simple, but isn’t. Suppose we decide to focus on new car sales to private customers, i.e. excluding commercial and fleet sales. There is a finer line than you might think between a new car and a used car. Where does a demonstrator vehicle fit in the picture? What about a car that has been pre-registered to a dealership – it’s ‘new’ but, on paper the person who buys it will be the second registered owner. This may seem pedantic; it’s anything but. Demonstrators typically make up a significant percentage of car sales. Someone I know went to a well-known German manufacturer of cars to purchase a new mid-level saloon, and came back with an ex-demo high-level executive model. To him, they were valid substitutes for each other.

We have just assumed away the effects, if not the existence of second-hand cars, but they are a valid choice, and therefore another potential substitute for a new car. What’s happening in the second-hand car market may affect what’s happening in the new car market.

The 1965 Coupe de Rust and an Audi A-something – substitutes for each other?

When did we sell them?

Volume sales data by date are usually simple for clients to provide. However, for cars, you first have to define the point of sale. Was it when a deposit was taken? Or when the balance was paid (or contract signed)? It isn’t usually when it was registered, which often comes long after (or before – I kid you not) the actual sale.

How much does it cost?

I’ve just checked the Audi website and an Audi A4 saloon costs from £27,810 today. Notice the ‘from’. What does it cost when I’ve added the must-have 13-speaker surround-sound system, or the obligatory head-up display? And run-flat tyres? And pearlescent paint in exactly the shade I want.

Of course, for each sale I should be able to ascertain the final purchase price, but this is the ex post ‘realised’ price, rather than an ex ante ‘offered’ price, so it’s reflects the decisions of people who bought, rather than the people who decided not to.

For that matter, what exactly is ‘price’?

In any case, does the final, total price matter? If I’m buying on finance, I any be more interested in the monthly repayment, the size of the upfront deposit required or even the annual rate of interest.

How often do you buy a car?

If you’re like most of us, it’s a relatively infrequent purchase. The chances are the car ads you see today will have no immediate effect whatsoever on your purchase behaviour – you’re not in the market for a new car. Looked at the other way, this must mean that much of the advertising for cars must be aimed at securing our future purchases. How do we factor that into our models?

How do you choose a car?

A car a big, complex thing with many ‘stakeholders’ (i.e. other people with opinions). You might fancy the sporty little number, but where do you put the shopping, or the kids, for that matter? Perhaps you value performance and efficiency, but one of your fellow stakeholders prefers appearance. It’s can be a difficult decision with many things to trade-off against each other. All of which makes for a complex path to purchase. Maybe you start online, or with a brochure, or perhaps you drop in to a few local dealerships to see the metal. The point is there are – potentially – many intermediate steps before a purchase, all of which may be modelled econometrically.

Dealers – you matter to us

Talking of dealerships, cars are stubbornly physical, meaning that traditional sales channels have retained their importance, notwithstanding all the ‘disruption’ of the digital age. Dealers matter. And, by extension, so does what they do, giving us – would-be econometricians – a further source of marketing activity to measure in our models. That activity is primarily tactical and local, and is seldom well recorded, giving us an almighty data headache. We shouldn’t underestimate the importance, too, of sales commission, and the potential that has to screw up our results.

“I’d like to see the one 4 levels down and 17 from the left” – a traditional UK car dealership

The sunk costs involved in developing new car models, means that once launched, models tend to persist for many decades. They have longevity. The aforementioned Ford Fiesta first appeared when I did, in 1976, and shows no signs of stopping soon, also like me. Yet, it isn’t the same car. Here, courtesy of Wikipedia, is the original Mark 1 Fiesta, in the shade of brown in which everything in the 1970s came, head-to-head with the latest flivver (Mark 7.5). You’d hardly credit that they are the same vehicle.

Things have changed, and not just tastes in colour

Of course, when we build marketing mix models we are usually interested in a much shorter period of time, usually the last 3 years, but in that time we’re highly likely to encounter a facelift or two, of varying degrees of severity, for which we have to control in our models.

Some tips for modelling cars

We don’t want to give the whole game away (it’s what we do for a living, after all) but there are a few implications for modelling cars that spring out of the above.

  • We have to control for facelifts – which, of course, are usually supported by increased media
  • Price needs careful treatment – test a range of proxies for price
  • Consider modelling intermediate KPIs, as well as final sales – orders, leads, web visits are all possibilities
  • Make sure that brand and tactical activity is adequately captured in the models

For a conversation about cars, feel free to contact us.

Philip Gaudoin

Sign Up