Broken Windows, Part I: Measured GDP Versus Welfare

Shamubeel posted here on Monday on whether natural disasters can be beneficial for an economy. In the comments, Miguel Sanchez and I discussed a bit whether economists are too quick to shout “broken windows fallacy” in such cases. There are a couple of interesting issues here, each of which is worth a separate post. 

Miguel points to this paper from the BIS (with a great title up to the colon, pity they felt obliged to add the post-colon clarification). The paper makes the claim that insured events, while not necessarily beneficial are “inconsequential in terms of foregone (sic) output”. A quick skim of the paper suggests that there are two separate aspects to this result. There can be a degree of over-insurance when a natural disaster destroys productive capital, since replacing that capital will typically result in newer and possibly more advanced capital. If the insurance liability falls outside of the region (for instance, as a result of reinsurance), then this improvement to the capital stock will have been financed from outside the region, and it is easy to see that this can generate a situation where a disaster leads to greater output (naturally, to be weighed against any direct human costs of the disaster). A fair amount of the insurance liability in New Zealand, however, fell inside New Zealand. In this case, the resulting improvement in the capital stock can still lead to an increase in the discounted flow of current and future GDP, but only because of a flaw in the way GDP is measured, and not because of any actual benefit.

To explain, consider how intermediate goods are treated in the measurement of GDP. If a household buys foodstuffs to make meals, the expenditure on that food is considered a final good and measured in GDP. If a restaurant buys those same ingredients, however, to prepare meals for customers, the sale of the meals is measured in GDP, but the expenditure on ingredients is not, as their value is already included in the price of the meal. To do otherwise would be double counting. Let’s imagine that, contrary to this normal practice, we were to change the definition of GDP and count both the food sold to restaurants and the meals sold to customers in GDP. In that world, if there was a preference shift and people chose to eat out more, we would see a big increase in measured GDP, but not one that reflected a comparable increase in welfare. Even worse, imagine that the government, under pressure to improve the data on GDP growth were to pass a law requiring people to eat in restaurants rather than at home. Measured GDP would have grown, but welfare would have fallen as people were forced to spend their income in ways different from what they would like.

This is obviously silly, and we would never make such a change to the way GDP is measured. It is, however, exactly analogous to the way we treat investment in GDP. Just as people can choose whether to spend their income on ingredients or eating out, based on relative costs and their own preferences, people can choose whether to consume to today, or save and consume in the future, with the interest they earn from their saving derived in large part from the return that can be obtained from the saving when used to invest. In other words, investment today is just an intermediate good that generates consumption in the future. By including investment in the measure of GDP today and then the flow of output from that investment in the measure of GDP in the future, we are double counting, just we would be if we counted both food sold to restaurants and the meals produced from it. And if a natural disaster leads to an increase in investment, funded not from outside, and not from a decrease in investment elsewhere, but from reduced consumption by those holding the insurance liability, then the flow of measured GDP will rise, but only because of the increased double counting not because of any increase in welfare, just as in the fanciful case where the government required eating out. As best I can see, the result presented by the authors of the BIS paper  rests on this double counting convention. 

I will follow up on Miguel's second point tomorrow.