Not Buying the Lost Sale “Baloney”
Last week my post on piracy estimates generated substantial debate (still going here and here.) By coincidence, the writer of the “Green Hornet” and “Daredevil” comics, Mark Waid, also stirred up controversy on a similar subject that day, mincing no words in taking on the related issue of “lost sales.” Then, this Wednesday, the U.S. Commerce Department issued a long-awaited, must-read ‘green paper’ which briefly touched on the subject as well.
My contention from last week was a relatively narrow one: losses to an industry may simply reallocate within the economy, and thus cannot automatically be considered losses to the economy. (Further explanation of this below.) Waid points to the separate and arguably more fundamental claim that an infringement equals a lost sale. In fact, Waid expressed two views: first, that his industry benefits from piracy, and second, that he would “go to his grave not buying the baloney that every pirated comic was a lost sale.”
Artists actually express this first view that piracy is beneficial with surprising frequency, , , , although I remain skeptical. While infringement may lead to promotion and interest development that would not otherwise occur, we cannot conclude that this promotion outweighs the eroded incentive to create without some serious empirical data. Some gain does not mean net gain. Some research has suggested there may be a net gain, but such conclusions often rely on correlation and indirect evidence. Certainly, it will vary across industries, as business models differ.
Measuring Lost Sales
I entirely share Waid’s latter view, however, regarding the lost sale fallacy. This problem – assuming that every infringement precludes a market transaction that, but for the infringement, would have occurred – has long plagued piracy estimates. The issue hasn’t gone unnoticed. In March, Ali covered a working paper from the European Commission’s Joint Research Centre, titled ‘Digital Music Consumption on the Internet: Evidence from Clickstream Data,’ which concluded that “much of what is consumed illegally would not have been purchased if piracy was not available.” The 2010 GAO study that I mentioned last week noted this, as did this week’s Commerce green paper, referecing GAO’s statement that “there is not likely to be a one-to-one substitution between legitimate and pirated content, although some degree of substitution is generally acknowledged.”
The lost sale fallacy has repeatedly come under attack because it flouts a fundamental precept of microeconomics: consumers will consume less of a good at a higher price, and more of that good as the price gets lower. And, of course, they will consume the most at a price of zero. We can thus deduce that in any market characterized by a standard demand curve, some infringements (i.e., acquisition at price zero) will not occur if the individual must pay prices higher than zero (i.e., purchase).
All that being said, at least some – if not many – infringements are in fact lost sales. There is certainly some substitution. The question is what proportion of sales are lost and which are illusory. But designing a model for answering this reliably is difficult. What is the demand elasticity for the given piece of content? If a digital product is available from two vendors at two prices, should the loss estimate value the infringement at the higher or lower price? What if the product is not available in the market where the infringement occurred? How can we consider the sale “lost” if the infringer had no legal mechanism for executing a lawful sale? Similarly, should we consider sales as lost if the infringer cannot afford the work infringed? In other words, if we know that in the absence of infringement a given sale would not have occurred, how can we conclude that sale was lost? These conundrums are in addition to even more technical questions, such as whether to account for transient displays or performances the same as downloads.
Disentangling Economic Loss from Transfers
Several of these factors also affect the discussion from last week, whether an infringement is an intra-economy transfer or a loss. The chart below represents a summary of the “simplified approach”, espoused by some critics of last’s week’s post, in the top row. The approach that I contend is more accurate (albeit taboo) is described in the bottom row.
A, one U.S. national, infringes a protected work of B, also a U.S. national, sold in the United States for $20. The cost to the U.S. economy is…
Notably, the values in boxes D and E depends largely on whether the infringement was domestic or foreign — a factor that is also likely to affect whether the sale was lost. (Since a non-trivial amount of infringements occur in markets where the infringed work is not lawfully available.)
Even the taboo approach here is necessarily oversimplified; describing f(x) would in fact require stating, as a function, the complexities of incentive and investment that legal scholars have debated for years. An accurate function might build off the creator-centric model laid out by Judge Posner and William Landes in chapter 3 of The Economic Structure of Intellectual Property Law, although the Landes & Posner model focuses firstly on the answer to boxes A and D, rather than boxes C and F).
Observations and Conclusions
Getting useful estimates about the economic cost of piracy requires answering technically challenging questions. Many studies therefore turn out to be a poor proxy for what policymakers ultimately need to measure: the quantitative extent to which creators’ incentive for future investment is diminished.
So what should be done? Some ideas:
(1) Quantify losses in terms of losses to industry/authors instead of losses to the economy, to avoid introducing complexities I mentioned last week, regarding reallocation versus destruction, whether infringements are domestic or foreign, and what else infringers might spend on, etc.
(2) Ensure data and models are transparent. Withholding datasets from publication prevents academic review, and similarly impedes the possibility for improvements to the model in follow-on research.
(3) Pro-rate aggregate lost sales estimates with variables representing the proportion of illusory sales, specifically, demand elasticity and out-of-market infringement.
Another possibility would be to focus away from lost sales entirely, and instead directly research behavior, since that’s what we actually care about. This is what surveys are for (as someone pointed out in the comments last week). In one example of this, surveys conducted by Booz & Co. polled nearly 200 venture capitalists on their expected propensity to invest in the face of increased copyright regulation, and produced useful data about the likely effect of legislative changes. While survey data may not be as bulletproof as actual market events (just as election polls are not as dependable as election results) a rigorously designed survey can nevertheless be highly probative.
A sound research agenda on these questions is entirely feasible. Doing it right is not impossible; it is just harder than doing it wrong.