Introduction - Here’s the Problem
Bottom Line: Alibaba’s Shopping Prices Indexes look like something you’re familiar with. In fact, they are little short of revolutionary, and could change the way we understand and measure economic activity.
On the same day that China published CPI data for December, at 2.1% yoy, Alibaba also reported its own online shopping prices indexes. The two headline indexes were the Alibaba Shopping Price Index (aSPI), which showed prices up 3.9% mom and 9.3% yoy; and the Alibaba Shopping Price Index-core (aSPI-core), which showed prices up 0.9% mom and up 0.2% yoy.
That’s a huge difference in result, and in both cases a very different result from China’s official CPI. What’s going on?
In fact, understanding the difference between the aSPI and the aSPI-core takes us into profound and important questions about how we calculate inflation.
Alibaba generates a massive source of verifiable data available for immediate analysis, and this is put to use in generating its indexes. In doing so, it has the potential to provide a more accurate, or ‘realistic’ tracking of price movements than any mere sample-based survey of prices can hope to do. But once you edge away from the idea of a survey-based inflation index and towards tracking massive amounts fulfilled interactions between consumers and suppliers, the assumptions and distortions embedded in the way we calculate inflation come leaping out at you.
Shortcomings in inflation indexes which we have lived with quietly for decades are no longer theoretical and of limited impact. Rather, they have a major impact on the published results, with the ‘errors’ generated by what turns out to be the very stuff of actual consumer behaviour. Generalizations that were plausible and acceptable 40-50 years ago and which were thought to produce only minor ‘errors’ turn out, in the age of online marketing and shopping, to be not merely implausible but demonstrably false, capable of generating major errors and, finally, unnecessary.
Alibaba’s two indexes represent a quite fundamental challenge to our ideas about inflation, and about pricing.
So How is Alibaba Calculating its Indexes?
As the starting point, keep in mind that the aSPI is reporting China’s inflation at +9.3% yoy; the aSPI-core shows it at just 0.2%, and these two indexes are drawn from the same underlying big data about Chinese online shopping transactions.
First, the difference between aSPI and aSPI-core bears no similarity with the differences typically seen in CPI and Core CPI. Elsewhere, the ‘Core-CPI’ is usually simply the CPI index with the more volatile pricing elements taken out - usually food and energy. Ostensibly, this is because those volatile elements can mislead about the ‘underlying’ inflationary trend. You may be sceptical about the purpose and relevance of producing these ‘core’ CPIs, but they tend to be watched closely by policymakers.
The aSPI-core is nothing like that: rather, it is constructed along familiar lines, tracking monthly price movements of a fixed basket of nearly 100,000 commodities and services under 500 basic categories, broken down ultimately into 10 different sectors (including food and energy, and other ‘volatiles’). The methodology is directly comparable to those usually used to construct CPI indexes, and is easily understood.
But faced with Alibaba’s mass of real-time transaction data, so are the shortcomings. Here’s what Alibaba says about it: “Due to the frequent product replacement in modern society, the sale and prices of many products have a life-cycle characteristic, that is, when one product is first released, it can enjoy a relatively high premiums because there are little similar products, thus little alternatives on the market; however, as time goes on, more and more alternative products as well as technical updating products will show up and compete in the market, and the premium will be lower and lower. This characteristic brought by technological progress and market competition will bring a natural trend of falling prices, and consequently the fixed basket index in the long term are likely to underestimate the rising trend of the cost of consumption spending. As new products emerge and spread faster in the online retailing, the problem may be even more obvious.”
To put it bluntly, if you track a fixed basket, you miss precisely those goods and services which suppliers introduce to drive both demand and pricing. Instead, you just record how prices fall as products and services drop out of favour or fashion. This year’s colour is Greenery (or so Pantone tells me) but you are still tracking the price of last year’s colour (Rose-quartz). Moreover, as technology and distribution systems accelerate globally, the errors introduced become greater and greater because the lifespan of any particular new product at the top becomes shorter, and its decline more dramatic. You end up producing a price index which reflects only one aspect of consumer & supplier behaviour, and systematically the most lagging and deflationary aspect.
Constructing the aSPI - A Radical Alternative
The aSPI (the one which shows inflation running at 9.3% yoy) is constructed to acknowledge these problems and deal with them directly. Instead of tracking a fixed basket of goods, it makes two changes:
1. For each category, it calculates the Volume Weighted Average Prices (VWAP) of goods at the lowest level of subdivision it can manage. By taking the VWAP it acknowledges that a consumer may choose to buy a more expensive product in the category simply because he/she wants it more. It has nothing to say about how or why that choice was made.
2. It then weights these underlying indexes according to the transaction shares in the latest month.
In other words, the index illustrates the result of the the numerous negotiations between suppliers (who seek premium prices, perhaps by innovation) and customers (who may seek lower prices, but also know what they want). As a result, the aSPI includes not just information about general price movements, but also information about how consumption choices vary, driven by variations of the relative price of cheap and expensive goods, but also by seasonal factors, by structural variations in income and income expectations.
Consider, for example, someone wanting to buy a computer.
A Capable Computer
Here we have to tangle with ‘hedonics’, which have been an increasingly significant input into price index calculations, particularly for technology goods and services. Typically, ‘hedonics’ will observe that, if a new computer is more powerful than the old one it replaces, and yet the price remains the same, then the real price of the technology has fallen, because you are getting more computer for your money. If you assess what kind of computer you have by checking its specifications, then this is certainly true.
But Alibaba’s data shows the problem in this accounting: consumers are buying not a bundle of specs, but rather what they perceive as ‘a capable computer’, or a ‘half-way decent phone.’ What constitutes ‘a capable computer’ now is quite a different creature from what it would have been five years ago. But in the most important sense of the word, it has hardly changed. And if the price of ‘a capable computer’ has gone up, it has gone up, regardless of the change in the specifications. Taking the volume weighted average price will simultaneously discover what a capable computer is, and what’s happened to its price.
Who is right? The hedonics proposition or the ‘capable computer’ proposition? Alibaba’s data answers this question: consumer behaviour leaves no room for doubt, the consumer seeks a ‘capable computer’, Alibaba has the data to know what at any one time constitutes a ‘capable computer’, and it also knows what’s happening to the price. Once you’ve got that information, the argument is simply over because there are no interesting questions left for hedonics to answer. The consumer is doing the only relevant defining.
What is the aSPI, Then?
Does the aSPI, then, measure ‘consumer inflation’? Not exactly, but perhaps it measures something more important: it measures changes in the price of goods and services which a consumer decides to pay in order to achieve a currently acceptable level of living. It records changes in the state of the deal actually struck between suppliers wanting to sell and consumers wanting to purchase, using their understanding of the products involved.
It strikes me that this is potentially extremely valuable information, if you want to know about the state of the market, the state of the business cycle, the state of profits growth, and, ultimately the state and ‘real’ size of the economy. (Also, if you look at it from the point of view of how consumers actually behave, and how companies actually seek to price/add value, perhaps we have been underestimating inflation for years.)
Currently it is difficult to interpret because:
a) there is no easily available language to describe it;
b) there is no easily available of economic theory within which to place it; and
c) we currently have only Alibaba’s short-run time series (since 2011) to examine, and are unable to make cross-border comparisons, or have the data to show how it alters within a business cycle.
But it requires no great stretch of the imagination to see how useful such an indicator could be. At the very least, changes in the pricing of ‘the deal’ could be expected to expose the state of the business cycle early and in detail. More widely, such an indicator is well positioned to produce a fundamentally more accurate description of economies, and accounting of economies, than is currently possible under existing data-definitions. (Anyone who has spent any time trying to understand the difficulties involved in constructing plausible deflators will surely agree.)
The repeated failures of economics is linked to the fact that it deals in trying to understand the wrong data, which itself is produced in ever-more sophisticated ways in order to finesse the original errors. Alibaba’s indexes show that the arrival of global online markets has made these historic indexes wronger, quicker, and there is no longer a need not to challenge these such errors. I’m not saying that Alibaba’s indexes get it right, but they are, perhaps, getting it ‘less wrong.’