Virulent Word of Mouse

September 22, 2010

The broadband price index puzzle

Does a consumer price index for broadband differ much from a producer price index for broadband? Though this question sounds like the final exam question for a real boring graduate class in economic measurement, I urge you to stay with me. There is something puzzling here about measuring innovation. Resolving that puzzle reveals something fundamental about broadband in particular, and about measuring innovation in general.

First, the answer. Why, yes, as a matter of fact, the two indexes can differ, but rarely do. Broadband’s experience in the last decade is an example where they do.

How can that be? How can the consumer and producer price index measure a different rate of improvement for the same phenomenon? They are based on different concepts. While those different concepts do not have to lead to different numerical answers, in broadband’s case they do. That is because a consumer price index ignores much technical improvement (more below), while a producer price index does not necessarily.  The details and reasons expose some puzzling principles for measuring technical change — arguably, explaining why our society undervalues economic improvements in broadband.

(There. Now you know how to get an A on this exam question.)

Indeed, I confess that this post is motivated by all the email I have received recently about a price index I helped make for broadband.  The email asks the same question: “How can a consumer price index for broadband possibly find little movement? Hasn’t there been a lot of innovation?” Well, let’s dive in and find out.

Different type of Price indexes?

I have written about the price of broadband in other posts, describing research done by Ryan McDevitt and myself. We continue to get interesting questions from readers that suggest we have not fully explained ourselves.

Our research did not find much price movement in consumer price indexes for broadband in the the last three to five years. Now, frankly, Ryan and I were a bit taken aback that anyone found it surprising. This finding should not surprise anyone with any familiarity with government statistics (So I guess most readers do not read government statistics regularly, which, come to think of it, is not so surprising….)

More to the point, the Consumer price index for Internet access has not changed much since late 2006, after a bunch of dial-up providers dropped their prices. Since then there has been almost no movement in prices. (See for yourself: go here, then ask for city average, and “Internet services and electronic information providers.”) Here is a graph.

If anything, Ryan and I found more price decline than the official US numbers. The graph above shows only increases. We actually found some evidence of modest price decline.

Several people emailed me. These emails said, in effect, that we were getting results that did not reflect their experience.

For example, one person wrote me and said (and I paraphrase): “In 2001 the price for the fastest service — then 768 k DSL – was $50. Today, the price for 15 meg FiOS is $44.95. Based on performance, that seems like a huge price drop.”

Yes, this writer has a point. The facts seem about right too. There have been some tremendous changes and improvements in access technology in the last decade. Yet, here is the confusing and surprising part: the consumer and producer price index would not necessary show a big improvement.

The consumer and producer price index answer related but different questions. Consumer price indexes show how much prices changed to keep users (essentially) equally well off. In contrast, producer price indexes show how much it cost producers to supply (essentially) the same good.

In both cases, these questions are harder to answer than they look. In both cases prices/cost should be adjusted for quality. In both cases, the index will stress what happens in the market. Producer and consumer price indexes are transactional indexes. They only count changes in prices for economic transactions that take place. That means a consumer price index will be a weighted average of users’ expenditure, while a producer price index would be a weighted average of a producers’ expenditure. (In most markets, user and producer expenditure are the same, so the weights will not differ).

But each uses fundamentally different data, which reflects the different question each asks. The producer price index uses producer data reflecting a producer’s experience, while the consumer price index uses data reflecting a consumer’s experience.

These do not have to give the same answer. Changes in the cost of providing goods may or may not be reflected in the changes in the prices users paid for goods. Indeed, when there is dramatic technical change, as there has been in broadband, producers and user can show very different experiences.

In the case of broadband, the producer price index changed a lot, but not necessarily the consumer price index. And there is one good reason for that.

Let’s get very specific about why. The example above about DSL and FiOS illustrates something that must be true: the cost of supplying broadband has fallen. Indeed, the cost of supplying 15 meg of access in 2001 would have been staggering, so that is quite a lot of improvement in less than a decade.

Ergo, a producer price index would seem to record a very large amount of change. If it were compared to dial-up, these changes would seem especially dramatic.

However, let’s not jump to conclusions too hastily. First, none of the improvement after 2001 counts unless somebody uses very fast broadband. In other words, because there were very few users of FiOS until two or three years ago, all the improvements prior to that received little weight, and (essentially) did not matter for either a producer or consumer price index.

Let me say it again. Producer and consumer price indexes are transactional indexes. They only count changes in prices for economic transactions that take place.

All the improvement prior to a transaction does not count in a transactional index. That identifies one of the first and most common errors made in public discussion about technical improvement. Technical improvement has little importance in either a producer or consumer price index until that technical improvement plays an important role in the value of many economic transactions.

More specifically, most of the adoption of broadband in this country happened in the middle of the decade. Only a little happened earlier in the decade (see graph below). Therefore, all the improvements in broadband at the beginning of the decade might impress an engineer or technologist, but do little for an economist. The improvements do not receive much weight because these happened prior to its adoption by many households.

As it turns out, this also is where a standard consumer price index treats this same historical event a little differently than a producer price index.

Many economists think the transaction bias of a consumer price index is too stringent. That is, a consumer price index does not start counting a new service, such as FiOS, until somebody starts to use it regularly. By design, it only compares improvements to the base year, only against the time in which FiOS first gets introduced. That is because the gains that arise in the base year do not count. That ignores the benefits to users from adoption of new technology.

Ryan and I were aware of this concern, so in prior work Ryan and I figured out what gains would have gone into a consumer price index had the improvements from the adoption of broadband been taken into account.

Specifically, we looked at data between 1999 and 2006 and estimated how much prices would have had to decrease in order to give uses the same benefit they gained from upgrading to broadband (from dial-up).

While the gains were large for each adopter (especially an early adopter), only a small percentage of households upgraded in any given year, so we could not get an enormous number. Nonetheless, we figured it had to add up to something, since many households were making the switch and willingly.

We came to the estimate that the upgrade alone would have accounted for 1.6% to 2.2% of price decline per year. That adds up to quite a price drop over such a short time — 12% to 18% over eight years — and it is above and beyond what a normal consumer price index would have shown.

(I would guess that we would get roughly similar estimates for the period between 2007 and 2010. However, we have not figured that out yet.)

Here is the bigger point: That is potentially a large mismeasurement in an industry generating thirty to forty billion dollars of economic activity. It is also the type of gain that a consumer price index will miss, but a producer price index will pick up. After all, all that technical improvement is precisely what induced adoption.


What does that mean altogether? In one paper Ryan and I found that consumer price indexes ought to include more movement to reflect the gains from adoption, and in another paper we found little price movement after users adopted.

More to the point, our findings in these two papers contrast with what a producer price index would say. A producer price index would have to show that — due to some impressive technical improvements — producers can now do some dramatic things today that were impossible a decade ago.

I wonder how big that number would be? It has to be a large number.

In other words, much innovative energy and improvement has gone into improving broadband technology enough to induce someone to adopt it. So far, not much innovation has translated into lower prices or qualitative improvement after someone has adopted it.

That illustrates a very general principle about measuring innovation.  While innovation can open up new potential — sometimes dramatically — that does not mean all that potential is valuable. In a consumer oriented society, not all innovation creates new value. Only the innovation that consumers use gets counted.

Price indexes can be very valuable tools in policy discussions. However, it is essential to understand how those tools work and what biases come with them.

More broadly, there is no such thing as one price index. Different types of indexes answer different types of questions. In the case of broadband, a producer price index would show that producers have made enormous gains.

Leave a Comment »

No comments yet.

RSS feed for comments on this post.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

Blog at

%d bloggers like this: