The UK’s February inflation stats were a pleasant surprise for a couple of reasons.
Firstly, the year-on-year growth rate fell to 2.8 per cent, closer to target than economists had forecast. Secondly, and perhaps more importantly, the Office for National Statistics updated its price quote data sets with a new glossary (direct download link) and data structure.
We’ve already spent a lot of time dwelling on the price quotes tables, which all the prices ONS agents observed for certain items in the inflationary ‘basket’. So why not spend some more?
Until February 2020, the ONS tracked an item called “Computer game top 20 chart”. At that point, it was into three different categories, imaginatively named Computer Game 1, Computer Game 2 and Computer Game 3.
An ONS spokesperson told us:
Splitting into three categories allows us to obtain more price quotes on what is one of the most volatile areas of the consumer basket. This improves the overall estimate, reduces volatility and aids users to more easily interpret these data.
According to the 2019 CPI manual, computer games — like certain other items, such as books and DVDs — are subject to special rules when it comes to price collection. For bestselling books, agents would get three price quotes, and computer games worked similarly.
Here’s how the price monitoring theoretically works: first up, a computer game is selected from a retailer’s bestsellers list. The price of that title is then tracked month-by-month based on shelf price. If the game falls out of the bestsellers list, an alternative game is collected.
So, say a new Call of Duty title is selected. An agent observes its price at £44.99 in a given shop. Eleven months later, the same game is still in the top 20, but is now £24.99. The next month, the Call of Duty title has dropped out of the top 20. The agent now picks a different game, and observes its price instead.
From an “inflation data should capture inflation” perspective, the ideal way for this to work is that perhaps:
1) The first game doesn’t drop in price as rapidly as in that example, and:
2) The game they replace it with is just a newer Call of Duty game.
In that way, you’re hopefully, roughly, measuring the rate of inflation in the price of an AAA game. And, because of smooth year-on-year replacement, there aren’t sudden jerky movements in the year-on-year growth rate (even if the actual cost resets to its high point each year).
There are several problems with this approach:
1) Month-on-month inflation with regards to this item reflects the price decay of a particular game relative to its price at launch, rather than offering good indication of the overall change in the price of video games as a whole.
2) Say the first Call of Duty game stays in the top 20 for seventeen months, and then drops out. If it is then replaced with a newer title that came out in the meantime, you would see lumpy shifts in the year-on-year rate.
3) Different games may experience price decay at significantly different rates — particularly ones like the Fifa titles, which are bought as much for the latest kits and player data as for actual game features.
Per the ONS’s detailed guidance, which we recently acquired, here’s what agents are being told to price:
The redactions are frustrating, and reflect the ONS’s general refusal to name specific brands and products. But permit us to go a little bit through the looking glass, and to reobserve the original file we based that data upon:
Those redacted sections, as you can clearly see…
…are of different lengths, because, duh, they’re telling agents to look for different things.
What those bars hide is indicated by the guidance: agents are told to “[c]heck correct platform”, because they’re being asked to observe games for three different platforms.
And, thanks to the new data structure mentioned above, it’s a bit clearer what those platforms are. In the new version of the price quotes, the item description (eg “Computer game 1”) is joined with a “consumption category” description:
So, we can surmise agents are asked to observe three games in the physical retailers they visit: one from the Nintendo Switch top 20, and one each from the PlayStation and Xbox top 20s.
And while the guidance suggests agents “[t]ry to price different titles for the 3 games” — an attempt to diversify the pool — it implies one could (doing their job badly) observe Fifa for all three consoles.
It might make sense for there to be some kind of central guidance in place, saying eg: “We’re currently trying to observe Call of Duty for Xbox, Fifa for PlayStation and Animal Crossing for Nintendo Switch”, but because bestseller charts vary by shop, so do the game choices. An ONS spokesperson told us:
The replacement is at the discretion of the collector and the guidance is to select a replacement that is representative of the stock in the shop. They will consider the stock on the shelves, and potentially use the type of game or the type of customer it’s aimed at to select a similar replacement.
This means an agent in, say, Reading could track their Computer game 1 (Nintendo) in Currys as Fifa, while another in observes Super Mario Party Jamboree in Currys Skegness. In a classic case of smallcagedmammalness, the average price could then end up reflecting this arbitrary compositional mix.
In terms of how inflation is calculated (based on “elementary aggregates” derived from the prices collected from one stratum, typically delineated by region and shop type), these things shouldn’t matter much as long as the Reading agent keeps checking Fifa during their Currys run, and the Skegness one keeps checking Super Mario Party Jamboree during theirs. Which, as we’ve already seen, they won’t.
Time for a chart:
Xbox, u OK hun? The apparent seismic drop in the price of a game for Microsoft’s console pretty much single-handedly drove a 37 per cent month-on-month drop in the (admittedly small) “Games, toys & hobbies” subcomponent of CPI during February, according to Robert Wood from Pantheon Macroeconomics. The PlayStation game has also had some pretty wild moments, particularly in early 2021.
Wood told us:
The CPI is a bit like a sausage; the more you think about what goes into it the less appetising it looks. Computer games are one of many data collection problems that mean the CPI is more volatile than it should be, and in some months probably does not fully reflect the reality on the ground.
Notoriously volatile computer game prices drive unnecessarily erratic movements in inflation from month to month. By collecting the prices of only a limited number of computer games, and seemingly substituting game titles with hugely different prices when a game becomes unavailable, the ONS delivers wildly volatile price readings that almost certainly fail to reflect the underlying reality of game prices but move headline CPI.
In fairness to the statisticians, they are trying to improve the use of industry scanner data in the CPI, which will massively increase sample sizes. That is complex work. But it’s still a huge failing that the ONS currently has such a limited volatile sample of computer games and some other items.
What, fundamentally, is going wrong here? As mentioned previously, the model of vibes-based title selection and sporadic replacement seems bound to create volatility, but this volatility seems alarmingly chronic. An ONS spokesperson told us:
Computer game prices are particularly volatile and do show wide price range fluctuations over both long and short-term time periods. The degree of volatility across the three games would depend on the speed with which products enter and leave the charts, and the price distribution of products within that particular item.
Nintendo games are, apparently, a relative beacon of stability. The ONS’s refusal to share specific information on item choices means we can only speculate why, but there’s a pretty obvious hypothesis: unlike its more powerful console cousins, with their constant game edition updates, the Nintendo Switch has a pool of popular titles that have been stable for the whole period captured here. For example, this could just be an index mainly of the price of Mario Kart 8, which launched at about £45 in 2017, and still retails at about £38.
It doesn’t seem that simple — though it’s hard to tell why. One possible clue is in the sheer number of agents who weren’t able to observe a price for the game they were seeking, and instead gave it an “M” code for “missing”. Here are those figures for February:
Yuck. So whatever it was they were looking for, fewer than a third of the ONS’s agents managed to find it.
But something more weird is happening here. Let’s repeat that exercise, but divide things up into simply accepted/rejected and take in the long view since 2019 (when the general top 20 item was on its last legs):
¯\_(ツ)_/¯
This might be a product of some underlying change in the basket, or possibly a conscious drive to make more observations of these items. We asked the ONS for more details, and will update if we get them.
And what were they observing? As might be expected at this point, it seems inconsistent:
It’s not quite RROD version two, but something’s definitely going wrong with the Xbox — look at that spread, and the heavy reliance on comparable (plus symbol) rather than identical (circle symbol) item observations. Effectively, there were only two observations — out of 98 attempts — where an ONS’s agent found the product they were looking for. No wonder the series is being weird.
Let’s go shop-by-shop, focusing on Xbox. We know from previous coverage that once you filter down by region, shop code and shop type you can create a continuous price series for a shop.
Intuitively you’d expect the price series for a set game at a set shop to be fairly smooth, with occasional adjustments. What we absolutely don’t expect to see are a load of wild and wiggly lines:
Womp.
Source link
