Good Luck to Anyone Building an Internet Inflation Index
(Bloomberg) -- When official inflation numbers go up, people complain about rising inflation.
And when the inflation numbers go down, those same people complain that official measures are riddled with corruption and poor modeling and all kinds of interventions that lead to artificially depressing the true pace of price increases throughout the economy.
So of course, private researchers (and newsletter-mongers) have been hawking their own measures for years.
Entrepreneur and crypto investor Balaji Srinivasan has launched his own project, which aims to scrape all the different price data across the internet, for a range of goods and services, in order to produce an independent, decentralized price index that obviates the need for official data.
The theory sounds nice. Collect all the data every day. Add up all the numbers. Divide. Voila: the real rate of inflation. No need for anything further or fancy. Just look at true prices.
Blogger JP Koning has raised some concerns over this approach here. Allow me to add a few more.
There’s a great paper that was put out by the BLS in 2008 titled “Addressing misconceptions about the Consumer Price Index.” And although it’s not its goal, it does a great job of explaining the challenge of using a naive approach, where you just take all the prices, add up all the numbers and then divide to get a true inflation rate. For one thing, any good index is going to have to use some kind of modeling and human decision making.
Starting on page 3 of the report, you get to the core problem. Here’s the heart of it:
A simple, if extreme, example suffices to get the point across. Suppose that a person buys four candy bars each week: two chocolate bars and two peanut bars. The bars cost $1 each, so her total spending per week on candy bars is $4. Now suppose that, for some reason, the price of chocolate bars quadruples to $4, while peanut bars remain at $1. The goal of the CPI is to measure how much the consumer needs to spend each week to consider herself just as well off as she was before the price increase.
The line I’ve bolded is key. Prices of goods are always going to fluctuate in idiosyncratic matters for various reasons. And because individual prices fluctuate, consumers are always going to be making tweaks to their take-home grocery basket. It doesn’t make sense to assume that all consumers will just maintain an identical consumption basket, causing their candy bar bill to jump to $10. But by the same token, it also doesn’t make sense to assume that the shopper will just buy four peanut bars instead, keeping their candy bar bill at $4 but taking home a subjectively less optimal candy bar mix for the week.
The game (or the goal) is to figure out how much more the consumer has to pay to receive the same candy-bar end utility. Maybe the answer is that she buys one chocolate bar and then three peanut bars, bringing the bill to $7, in order to achieve the amount of candy-eating pleasure that only cost $4 the week before. That’s definitely candy bar inflation, but at $7, it’s more realistic than $10.
Again, this is all extremely crude, and the paper gives other examples of relevant choices (EG: buy more hamburger meat when steak is expensive? How much more do you have to buy?). Anyway, given that there is no preference uniformity among consumers and there’s no objective exchange ratio between peanut bars and chocolate bars, some economists have to make assumptions or calculations in order to address these issues. You can see why a naive “add-them-all-up-an-divide” approach won’t suffice.
Another tricky area, that’s filled with misconceptions, is quality adjustments (also known as hedonic adjustments). Cynics and skeptics claim that inflation is being depressed by comparing, say, the price of a cell phone today, imputing it with near infinite quality improvements, and then comparing that to an old rotary phone (which you can hardly even buy anymore). Starting on page 5 of the report, they get into the fact that not a lot of the total inflation basket is subject to much modeling, but even beyond that the approach is far more modest. And, contra to critics, the BLS does make reverse quality adjustments.
The report gives some examples:
To take the most straightforward example of quality adjustment, which the CPI handles automatically, suppose the maker of a 1.5-ounce candy bar selling for 75 cents replaces it by the same brand of candy bar, still selling for 75 cents, but weighing only 1.0 ounce. If the shrunken size is ignored, it looks like the price hasn’t changed. The CPI, however, prices candy and most other food items on a per-ounce basis and would automatically record a 50 percent increase in the quality-adjusted price of the item, from 50 cents per ounce to 75 cents per ounce.
Another example of how the need for quality adjustment arises is a hypothetical (but plausible) situation in which the CPI has been tracking the price of a specific model of 32-inch standard-definition color television at a certain store. If the store no longer sells that model, the CPI data collector will find a replacement model to price each period thereafter. In the event that the store has decided to sell only high-definition televisions (HDTVs), one of those will necessarily be selected as the replacement. In that case, the replacement television may cost 4 times the price of the previous standard-definition model. It would be unreasonable to treat this rise in price as a sudden fourfold increase in cost, given that the HDTV model has a larger screen size, a higher resolution picture, and other enhanced features. The BLS must make some estimate of how much of the price difference is due to the improved quality associated with the HDTV model.
Obviously, some of this is going to be more art than science. But the alternative is to use no art at all, which also seems pretty absurd. Many people would agree that advances in technology are good.
Like JP Koning, I’m pretty excited to see where Balaji’s price project goes, how it tracks with (and where it deviates from) the official numbers. There’s certainly a lot that can be done from a data-collection standpoint, just by looking at the price of goods and services on the internet. However, there’s good reason to think that a pure, naive collection of prices, without individuals making decisions about how they should be adjusted, will not do a better job than official sources at producing an inflation index that better tracks the actual consumption experience of the public. You could solve this by having decentralized teams apply their own adjustments to the raw data, but then you’re back to square one of having to trust individuals, which in theory, a crypto/decentralized approach is supposed to solve.
But seriously, good luck to the project!
©2021 Bloomberg L.P.