Your valuations and price indicator questions answered
On last week’s webinar, we held our first Ask the Expert session with resident data guru, Jim Stamp. Jim heads up the development of our valuations, the key component that drives our price indicators. We received a lot of questions during the session and, as promised, here is a round-up of the questions Jim answered in the session, as well as those we didn’t get time for. If you’ve got any more questions you’d like to ask, just leave a comment at the bottom of this page and we’ll get back to you.
"We have seen cases where Standard/Available features are incorrect. Do you see a way to improve this accuracy?"
- Cazoo
There are situations where the options displayed in Portal and on APIs do not line up with those fitted to the vehicle.
Ideally, we would filter the available features using the date from when the first owner built the vehicle order.
Unfortunately, this date, and even the date of manufacture, is rarely available. So, we have to use the date of first registration.
This date can be significantly later. So, the features displayed in some systems can be out of sync. We are trying to find ways to make this a better experience.
The good news is that we do not have the same issue regarding valuations.
We allow any features that were optional on a derivative to affect the valuation, whilst promoting those that fit the timeline of the specific vehicle.
"How does Spec Adjustment account for the fact many retailers are not recording all spec on Auto Trdader?"
- Cazoo
As your question suggests, our valuations are reliant on our retailers understanding their vehicles, any added optional extras, pricing accordingly, and then us extracting the value.
We would love to have a perfect view of the state of all vehicles we assess. But this isn't possible yet.
We have to accept that the information that we receive forms the value perceived in our marketplace by our retailers and, therefore, consumers.
For example, suppose no one identifies a panoramic sunroof, nor prices accordingly. In that case, we can only assume that retailers view it to have no value.
"How much historic data do you use in the valuation? How does this affect the speed/amount prices move"
- Cazoo
When creating our Retail Valuation, we use 28-days worth of advert data.
These observations are weighted such that yesterday's adverts have more impact than the day before.
We also use the most recent price for each advert, rather than the average. We have chosen this method to allow our valuations to track the market as it moves, whilst still maintaining an acceptable level of stability.
Finding the sweet spot between the valuations jumping around each day, or never reflecting the dynamic nature of the market's supply and demand changes, is vital to the success of our system.
"With relevance-based search (the way it is served to customers) do you know what percentage of customers change to price vs stay on relevance-based?"
- Inchcape PLC
As Chris mentioned in the webinar, we don't discuss the finer details of how the relevance search works.
Still, we're happy to say that majority of people stick with the default.
"How accurate is the taxonomy data you guys use within the indicator? Is it in line with CAP spec check?"
- Inchcape PLC
We like to think that the taxonomy that we curate, represents one of the best out there.
We take data from multiple sources and edit it to create a view of the vehicle market best designed for advert creation.
In its current form, we can tell you what spec was on the vehicle as standard, and what was available at the time of manufacture/registration. We are not yet able to say what build a specific vehicle had when it left the factory.
We like to think that we have created our taxonomy to be consumed by humans, rather than drive a valuations engine or back-office ordering system.
"How do you residualise the various options. Do you do it with the car or is there a defined table?"
- Jaguar Land Rover
We take a significant number of adverts and analyse them to detect retailer pricing behaviour and correlate it to the optional extras identified within each advert.
In doing this, we can build what our retailers think the vehicle is worth without any features and then what each optional extra is worth on top of that base valuation.
We run this process regularly and at a derivative level. Unfortunately, I don't have a spreadsheet which specifies the value of each optional extra for each derivative year/plate combination. I think it would have a fair few rows in it!
"How do we know that the car we are selling is unique and has more features than any other of the same range. Plus how many of the same range have been built..."
- MFB Motors
By comparing the two forms of our retail valuation, you can get some idea of how unique your vehicle is.
We adjust the derivative-level valuation for taxonomy, age and mileage and then by optional extras to create the vehicle-level valuation.
If the vehicle valuation is higher than the derivative, then we have detected a higher than average amount of optional extras on that vehicle.
Regarding the count of vehicles in a range, we do include the visible supply (and demand) in our products. You can find these numbers in the Market Insight area of Portal.
For a direct vehicle comparison, they form part of the inputs for our Retail Rating. You can access them in isolation via the insight panel (or) Retail Accelerator.
"Does the initial valuation assume a 50% spec? As when we put the car live, the following day it changes. For example, if I put a car on at 97%, when I check the next day, it has gone to 98% or higher"
- SW CAR SUPERMARKET
I think your question is two fold. When you initially add a vehicle in Retail Check, we will provide you with a valuation based on taxonomy, age and mileage - our derivative-level valuation.
Once you tell us more information, we will then build a more specific valuation at present based on the optional extras (our vehicle-level valuation).
"100% is good???"
- SW CAR SUPERMARKET
I don't claim 100% is good, but it is the average value that your competitors are using to price similar vehicles.
We see in our data that pricing at 100% seems to be an optimal point for both profit per unit and days to sell.
This optimisation is not Auto Trader's view, but an aggregation of retailers' behaviour - pricing to maximise their profit (high margin vs selling fast).
"Why do Auto Trader’s price indicators not take into account: full main dealer history, number of owners, warranty supplied with car, and the fact it is retailed at a main dealer?"
- Victor Wood
We hold the philosophy to value the car, not the deal, very highly.
Ideally, we would take the service history into account, but it is tough to verify this information.
The warranty and the retailer type are a preference of the consumer. This comes down to implicit versus explicit parts of the deal.
"Surely affecting the price indicators by colour is also subjective, as is the deal? You also mentioned "the deal and warranty etc. mean different things to different people” - so will colour?"
- Victor Wood
I agree that colour preference is subjective, but I have been told, on many occasions, that white/black/grey commands a higher price than other colours.
We've found, however, that this is only true at a market-level. Once you look at even the model level, and even more so at a trim level, other colours start to get an uplift. Burnt orange is generally not a colour in high demand on Fords, but for an ST, it adds quite a few pounds.
"Why can't we turn valuations off?"
- Victor Wood
At a personal level, I guess you can ignore them. However, from a market level, many of our retailers use our valuations every day to price adverts and buy vehicles at the optimal price.
For consumers, they reduce the amount of time they have to spend scrolling and down our website, finding out what a fair deal is. Transparency is one of the main factors that our consumers want when they buy a car.
Our valuations, and the price indicators that rely on them, are one of the things we do to help your customers feel more comfortable with the process. Allowing indicators to be turned off, when a retailer wasn't happy with them, would break that.
"You should still value £50k cars on basic values. They are all over AT. As I said, 42% of my stock is over £50k!!!"
- Marshall JLR
Our ability to be confident in the valuation of optional extras is dependent on several factors:
Retailers need to recognise the optional extras on the vehicle
The retailers need to increase the price of the vehicle in line with what they believe consumers will pay for that optional extra
We need to see enough examples of this pricing behaviour to be able to detect the difference between vehicles with and without the optional extras
All of these can lead to the model under-allocating value to an optional extra. With vehicles over £50k, especially JLR, the worth of optional extras makes up a significant proportion of the final price.
We also see fewer examples of these higher-priced adverts, relative to other pricing segments. So, the risk of undervaluing them is even higher.
Because of these compounding issues, we feel that the potential for putting high-price indicators on these adverts would be higher than any other segment of stock.
Using the derivative-level valuation is an option. Unfortunately, however, with the massive range of prices for each derivative, the indicator boundaries would not be wide enough for you to reflect the value of the spec on the vehicle.
We are looking at using separate boundaries for each make/model, but the execution is complicated. It will require much thought to get them to work in our products.
"Have you adjusted the indicators since re-opening? I.e. the percentages for each flag."
We have not adjusted the boundaries during the lockdown. We made a tweak to the fair/high percentage towards the end of February when we observed a tightening of the margins between the trade and retail markets. We have not put this back, as we recognise that the conditions are still very much in flux.
"Why do you consider that you should be judge and jury over flagging car prices?"
Transparency is core to our consumer website. Buying a car is a difficult thing for many people, and easing that journey is something that Auto Trader has always tried to do.
One of the issues that car buyers have is understanding whether they are paying a fair price. Before price indicators, consumers would have to scroll through adverts, comparing what they thought were similar vehicles to find an equivalent deal. We feel that the indicators are an excellent method to remove the need to do this.
With the most extensive view of vehicles in the UK, we believe we are well placed to build the valuations that drive price indicators.
We spend a considerable amount of time making sure that the systems that produce our valuations are working as expected. This goes hand in hand with finding ways to remove the outliers that cause problems for both our customers and consumers.
"What goes into a valuation? And what, if any, human opinion is involved?"
Our valuations are purely data-driven. Other than creating the models that produce them, there is no human input to the process.
Each night we filter through millions of vehicles to reduce them down to around 800k, which we then feed into the engine which produces depreciation curves.
These curves are built using the age and mileage of each vehicle, hierarchically down through our taxonomy. This means to create a valuation for an Audi A3 35 (2020 plate), we extract a depreciation curve starting with effects from all Audi's, then A3s and so on.
We do this until we get to the data that is from adverts that are of the same derivative year/plate. We use data from the past 28 days (weighted so that yesterday's data has more effect than the day before), from a variety of sources.
These sources include adverts from our site, Dealer Profile pages, Dealer Websites, Manufacturer Websites (UVLs, or Used Vehicle Locators) and exports to partner/competitor sites.
Alongside this process, we take two years' worth of advert data and extract the value of optional extras by analysing the pricing behaviour of retailers and comparing the difference between the spec and pricing.
"When I initially retail check a car, it provides a valuation. This valuation then changes when I add it to my forecourt and confirm the spec. Often confirming the spec lowers the valuation - why is this happening?"
- David, Cheshire Cars
This is due to the difference between our derivative and vehicle-level valuations.
Our derivative valuation is based on the taxonomy, age and mileage of a vehicle. This is what you see when you perform a VRM Lookup in Portal.
Once you add the stock to your forecourt or confirm optional extras, we then layer on our adjustment for the full vehicle spec. When calculating the value of the optional extras, we find what we think the vehicle would have been worth, if it had left the factory with no extras ordered.
You can see this value if you confirm no extras in Retail Check. The derivative valuation will always be higher than this standard fit value as it includes the 'average' spec for that derivative. Once you start adding extras to the stock item, you will be able to see the valuation increase, potentially past the derivative valuation.
"When I add what I consider to be a high-value spec item, like a panoramic roof, the valuation doesn't adjust by enough in my opinion. Why?"
- David, Cheshire Cars
Because we extract the value of options from adverts, we require that they are fully described and priced according to the options on that vehicle.
Suppose a significant number of retailers don't recognise that panoramic roof is optional, or they don't think it adds any value. In that case, we can't extract any value for it. Whilst this is frustrating, it does reflect the market's view of this feature.
"On a previous webinar, Chris said he did not want a "race to the bottom", as those retailers who are not as stable slash the price of their cars after lockdown. The current Auto Trader high/low coloured score system will accelerate this. What are your plans to protect the good dealers who value their standards and have protected themselves against this?"
- John
We carefully monitor the pricing behaviour of our customers. We take great care to make sure that the changes we make do not create this effect.
The indicator boundaries have been selected to allow movement in prices, reflecting the changes in the market.
Looking at the data, we can see no behaviours to indicate that adverts are being positioned to avoid fair or high indicators.
We have also replaced the old default sort order (low-to-high) with our relevance order, which has not allowed prices to inflate.
"Is it possible to also factor in the length of dealer warranty offered into the price equation?"
It is tough for us to extract from adverts the value that is attributable to the offered warranty.
The warranty period is often in line with the pricing strategy of the retailer segment and separating the two is nigh on impossible.
It also falls into the implicit/explicit issue. We know consumers care about mileage, doors, space, age, colour, etc., but vehicle warranty is not something that we can objectively allocate value to.
"Why do you insist on price indicators when they are misleading to the consumer? You do not take in account colour, service history, owners or the package they are sold under eg. approved schemes. There is a mention of approved schemes on your website and a banner on the advert, but no full description of what it includes on the advert itself or an adjustment in price for the package! We often have customers asking for discounts because Auto Trader says it is overpriced compared to Fred in the shed down the road, you do not price like for like. What are you prepared to do about it?"
We know that there are still things that we can take into account in our valuations.
Over the years, we have prioritised working on the things that make the most significant difference to explain the gaps. We are planning to include the vehicle's colour and the number of owners soon.
Approved schemes are tough to assign value to. They represent a more subjective element of the deal, meaning more to some people.
We also find it hard to isolate the value of them within the noise created by pricing strategies.
Service history has always felt like a crucial part of a vehicle's providence. We would like to include it, but we don't have a reliable way of validating the data in our adverts. As mentioned elsewhere, we prioritise valuing the vehicle over the deal.
"I am having to pay more for my stock but this is not reflected in the price Auto Trader puts out eg. "Good Price". How is Auto Trader taking account of the higher prices that vehicles are currently commanding at auction?"
- T James Motors
We have no direct control of our valuations. Our valuations are purely driven by the pricing expertise of our retailers.
If, as a community, our customers increase their prices, to maintain margins due to the inflated trade market, our valuations will change to follow.
"Hi, we have an Audi A3 1.4t S Line Sportback, which as of today is the only one advertised in your platform and is showing High price at £5995, and £746 over market average. There are no other cars to make it over market average. This action is costing our business money."
- Badger Automotive
We can build valuations even when it seems that we have no, or very few, equivalent examples for two reasons:
We use data from many sources, one of which is adverts on our site. We also use UVL, Profile Stock, Dealer Website adverts and export feeds (where we send stock to competitors/partners).
Because of the hierarchical nature of our model, we derive depreciation curves as we move down the taxonomy, increasing the accuracy as we move down towards derivative. We only produce valuations when the curve the model generates explains the adverts used to build it. In other words, we only allow valuations that we think are are sufficiently accurate to be distributed.
"Why do private sellers not have price flags against their adverts?"
The vast majority of private adverts are priced below our trade adverts.
This lower pricing behaviour is because we give them a recommended value which is intentionally between retail and trade.
Using this private valuation to drive the flags would cause a two-tier system of flags, which would be confusing to consumers.
We also know that it takes quite a bit of expertise to be able to tell what features of a car are optional. This would cause us to use the derivative-level valuation, which seems unfair to private customers.
"Does the pricing and days to sell work on LCV's?"
We have retail (derivative) valuations for LCVs, but we haven't yet completed the work for the days to sell metric.
"Can you show valuations for older cars as well?"
We made the decision not to provide valuations for vehicles more than 15 years old.
This is due to the effect that condition can have on cars after this age.
This is a good illustration of where we do not distribute valuations where we don't feel we can be sure that the depreciation curve reflects the input data.
"What time of day do valuations update?"
We update all adverts overnight, around 2 am, and then every time they are edited, during the day.
"Is location considered in valuations?"
We have looked at location effects several times, and have seen the variance change over the years.
While it is true that there used to be massive changes across the UK, we now only see pockets of high-priced vehicles in the South West and in Northern Ireland.
Most of our customers now view the market as a national one, and price accordingly. We do include location in our days to sell metrics (part of Retail Rating), as location can make a difference to what sells where.
"How accurate is your spec adjusted when I'm sending my stock via a data-feed so my vehicle information doesn't always match your taxonomy?"
We treat feed adverts in precisely the same way as we do adverts created in other systems.
All adverts are converted to a block of text, which we then separate and split into potential features.
Each of these is then compared to the features listed as optional for that derivative. If a feature matches the taxonomy sufficiently (we use a fuzzy matching algorithm), it will be passed onto the valuation.
As you can probably see, if you use our taxonomy to specify optional extras, you will get a better hit rate on the matching process. Ultimately, this will give you a more comprehensive valuation.
We hope this information has been helpful. If you have any further questions, please leave them in the comments section below and we’ll get back to you. And don’t forget, you can sign up for the next webinar here.