CPI Bias and Stagnation

Tyler has been defending his stagnation hypothesis with an intuitive argument about when the CPI is most skewed. I’m not 100% persuaded of Tyler’s intuition.

There is a severe conceptual problem that plagues any measure of inflation. It seems easy enough to measure the change in price of a basket of goods, but real-life consumers do not buy the same basket of goods from year to year. Entirely new goods get introduced, and even goods that seem nominally the same tend to improve in quality. These both introduce biases.

Suppose you’re starting with a basket of goods that contains x and y. Then new good z gets introduced. You don’t have the period-ago price of z, so you might continue for a period assuming that consumers are just buying x and y. You add z to the bundle in the following period when you can make a price comparison for z. This is going to cause CPI to overstate inflation, because in the period where z was available on the market but not included in the basket, consumers had greater choice with a given amount of money than the index suggests.

Now suppose you’re starting over, again with a basket of goods that contains x and y. Instead of adding a good z, assume that in some period x and y are replaced on the market by x’ and y’, which are improved versions of x and y. Consumers buy x’ and y’ in roughly the same quantities that they bought x and y, respectively. Since x and y are no longer on the market, there is no price comparison that can be made across periods. One way around this is to assume naïvely that x’ = x and y’ = y. Once again this is going to cause CPI to overstate inflation, because a given amount of money can buy more quality than the index suggests.

Tyler claims that the first problem is more severe in practice than the second. I am not so sure. One reason that we can never be sure is that statisticians at the BLS attempt to correct for both kinds of problems. This is a confounding factor that takes simple intuition out of the picture. How can we know whether net of statistical correction one problem is worse than the other?

But even assuming naïvely that the statistical corrections are equally effective or ineffective, it’s not obvious that the first problem is the most severe. In the periods when truly new goods first get introduced, they typically do not make up a large fraction of the real-world consumption bundle. This is an inherent limit on how much damage they can do to the index. In contrast, secular improvements in goods tend to affect the whole bundle. There is much bigger scope for damage to the index from quality improvements than from the introduction of new goods.

Tyler has been advancing the differential CPI bias argument to amplify, not make, his core argument, but if he has it backwards, that CPI bias has been worse in the post-1973 period than in the pre-1973 period, then his whole stagnation hypothesis crumbles. I’d like to see more discussion of his CPI intuition.

Update: Bryan has more.

7 replies to “CPI Bias and Stagnation

  1. Eli Post author

    I think that would affect whether consumption of z comes at the expense of x or y, but it’s not clear to me how that affects the CPI.

    But maybe you’re after this point: in eras where the bundle contains more substitutable goods, CPI bias will be more severe. If x and y are substitutes, then when the price of x goes up, people shift to y and vice versa. The composition of the bundle changes endogenously. I think it’s fair to say that the post-1973 era has more goods and therefore more substitutability, which means CPI bias is higher than in the pre-1973 era.

  2. Adam

    I am also uncertain of how some things are even counted in CPI.

    For instance, I used to have to pay money to get anything close to the quantity of reading material, video, and images that I consume on a regular basis. Now, other than the initial purchase of a computer and the monthly internet fees, I don’t have to pay any money at the margin for most of my content consumption. I realize this falls into the “infovore” category which you said in a previous post may not have as big an impact on the average American, but still–I think the average American does consume some amount of digital content for no monetary cost. How is this counted in CPI?

    I suppose you could argue that these unpriced digital goods are like “quality” improvements to priced hardware and internet service, which would put it in the second category of CPI biases. What do you think?

  3. Eli Post author

    Adam, those goods are not counted at all in CPI, but they’re also not counted in GDP, so it’s not clear that we’d want them in the CPI. But as you can see, it’s really hard to come up with a good, single metric for quality of life.

  4. Adam

    Which is why debates like this are so difficult.

    By the way, have you ever read any Cox and Alm? They make a good case that consumption is a much better measure of quality of life than income. This is an article of theirs from a couple of years ago.

    I always felt their was something to their argument, though I think the common response would be that if the consumption was achieved through unsustainable debt, it wouldn’t make a very good measure. But pointing out specifics things, such as that the average poor household has more air conditioners, cable TVs, etc etc, than they did in your prior period of choice, is always a good indicator of quality of life improvements.

  5. Indy

    Here’s something related I posted at Scott Sumner’s blog not too long ago:


    I always like to refer back to this page about hard drives when inflation issues like this come up.

    The best place to start on our trend is about 30 years ago, when a single megabyte of hard drive cost about $200 (in 1981 dollars!). By mid 1994 – that had dropped to $1. By 2000, to 1 penny. Today, last I checked, you can get a 2 TeraByte WD “Green Caviar” drive shipped to your house for $70 – that’s 3.5 cents per gigabyte in 2011 pennies.

    The US government, if it wanted to, and for only seventy bucks, could keep a 600K file on every citizen on one single hard drive. The collected works of mankind and entire libraries can now be stored in one server and comprehensively searched almost instantaneously for a few thousand dollars. Makes you wonder what a multi-billion-per-year NSA can do.

    Now, according to the CPI, a dollar thirty years ago would be worth about $2.60 today. By my calculations, the real price of digital storage has declined by a factor of around 14 million!

    But of course, we can’t just consider space alone. You’ve got to think of other real quality improvements like speed (over 1000 times faster than a 1981 hard drive), reliability, compactness (1000 times more compact), noise, power consumption, ease of use and installation (“plug and play”) resistance to physical shocks (my laptop could tell some stories, but keeps plugging along, but really old drives would crash if you looked at them funny) etc. I’m going to say that, per megabyte, current hard drives are also at least 70,000 times *better* than they were in 1981.

    That adds up to a total deflation factor of something almost everybody uses of of about One Trillion (give or take an order of magnitude) in a single generation! Think of how many things in our life would not have been possible without this kind of price:capability ratio improvement in digital storage.

    The same story could be told for computing power in terms of transistors and operations-per-second as well, though that’s been decelerating quite a bit lately.

    I don’t think there’s any historical precedent for something like this. The fact is we’ve got two parallel worlds coexisting in the general marketplace.

    One – where most Americans spend most of their money (80% IIRC), food, clothing, shelter, transportation, education, health care – is defined by relatively slow and steady changes in price and quality and contributions to “utility”. Using the chain-method to compare one year with the previous year can safely assume that price differences are not mostly quality-influenced.

    The other world almost needs to chain months or weeks together. And their ability to deliver life-altering levels of utility has exploded (especially for a certain minority of uses, who Cowen calls “infovores”).


    I’ll add the assertion that I don’t think we’ll see another “Trillioning” (I should trademark that term) like this again.

  6. Pietro Poggi-Corradini

    Another possible bias is the fact that prices are really marginal prices. So there’s a difference between a situation where most people are purchasing their first car and a situation where most people already have a car and are thinking to by a second one or a third one. I don’t know how this would affect prices, but certainly looking at consumption patterns would be a much better indication of stagnation vs. non-stagnation.

Leave a Reply