The International Comparison Program (ICP) collects the survey data on prices across countries that are used to estimate Purchasing Power Parity exchange rates (PPPs for short). These are then used to make international comparisons of real incomes and for other purposes, including measuring global poverty and inequality.
The results of the latest ICP round in 2005, led to downward revisions of real GDP for many developing countries (World Bank 2008a and 2008b). In a dramatic and much publicised example, the World Bank’s estimate of China’s real GDP per capita for 2005 fell by 40%, from $6,760 to $4,091 per annum. Figure 1 plots the Bank’s revised estimates of real GDP per capita for 2005, using the 2005 ICP, against those it made for the same year just prior to the release of the 2005 ICP – it shows that there were other large revisions. There is a clear sign of bias, with greater underestimates for countries with a lower GDP per capita. These revisions come entirely from the new PPPs.
Figure 1. Revisions to GDP per capita for 2005 implied by 2005 ICP
Source: World Bank (2008b)
Note: The 45-degree line indicates no revision
The interpretation of these revisions is clouded by the fact that many changes in data and methods were introduced in the 2005 ICP round. These changes have cast doubt over whether the new PPPs can be compared over time. Some observers have questioned the new PPPs. Maddison and Wu (2008), for example, assert that the new PPP for China is “weird” and “implausible.” The large data revisions have also led some observers to doubt the accuracy of the whole exercise of making international real income and poverty comparisons.
This raises two key questions:
- Can we make sense of the changes in Purchasing Power Parity exchange rates?
- What accounts for the large revisions to estimates of real GDP for the same date?
The dynamic Penn effect
I have tried to answer these questions by comparing the latest PPPs with those from the prior round for 1993 (Ravallion 2010).
To motivate the empirics we need to step back a bit to recall why we use PPPs in the first place. Market exchange rates can only be expected to (eventually) equate purchasing power over internationally traded goods. There are also non-traded goods, such as most services, which are naturally cheaper in countries where real wage rates are lower. This is the essence of the well-known Balassa-Samuelson effect. By considering both traded and non-traded goods, PPPs allow for more valid real income comparisons. Because real wages tend to be lower in poorer countries, we also observe lower price-level indices – the PPP deflated by the market exchange rate, or the inverse of the real exchange rate – in those countries (see for example, Summers and Heston 1991 and Rogoff 1996). This is often called the Penn effect, after the Penn World Tables.
The motivation for the new empirical work summarised here is the observation that the Penn effect should also hold over time if the data are in reasonably accord with the assumptions of the Balassa-Samuelson model. As poor countries grow, the labour productivity of their traded-goods sector will tend to rise, spilling over to wages in producing non-traded goods, and so their price structures should become more like those of developed countries. The PPP will move closer to the market exchange rate. We can call this the dynamic Penn effect.
To test for this, I have regressed the proportionate changes in price levels between ICP rounds on corresponding growth rates of GDP at market exchange rates. The results confirm the expected dynamic Penn effect, as can be seen in Figure 2 which also gives the least-squares regression line. I also find that this effect is even stronger in initially poorer countries. Thus the widely-observed static Penn effect has been attenuated over time.
Figure 2. The dynamic Penn effect, 1993-2005
Source: Ravallion (2010).
On its own, the dynamic Penn effect accounts for about one fifth of the variance in the proportionate changes in the log price levels over 1993-2005. An augmented version of my basic model, allowing for measurement error in the PPPs not based on ICP price surveys, can explain almost half the variance in the proportionate changes in price levels.
For the purpose of predicting changes in PPPs, I find that 99% of their variance between the 1993 and 2005 ICP rounds can be explained by just two variables: GDP growth rates and changes in nominal exchange rates.
Returning to the example of China, it is not too surprising that China’s price level index has been rising appreciably, given the country’s high growth rate. My calculations suggest that about two thirds of that increase is accountable to the dynamic Penn effect. The bulk of the remainder probably reflects an upward bias in China’s PPP due to the 2005 ICP’s poor coverage of China’s rural areas, where the cost of living is lower (Chen and Ravallion 2010).
Implications for PPP revisions
The methods currently used for updating PPPs in lieu of new ICP price surveys (which are many years apart) ignore the economic changes described above. The current method – as used by the World Bank’s World Development Indicators (and others) – simply adjusts for inflation since the last ICP round. For example, all that growth and structural change in China between 1993 and 2005 was essentially ignored when updating the country’s PPP prior to the 2005 ICP.
My results suggest that a better approach would be to bring the dynamic Penn effect explicitly into the model for updating the price-level index for non-benchmark years, using market exchange rates to back out the implied PPPs. Along these lines, in recent research (Ravallion 2010) I propose a simple and feasible way of exploiting the dynamic Penn effect in updating PPPs prior to each ICP round. It is shown that this alternative method reduces the mean (absolute) error in predicting the 2005 PPP, ahead of the release of the 2005 ICP, from 7% underestimation (using the inflation-adjustment method) to less than 2% overestimation, and the incidence of extreme errors (in both directions) is reduced noticeably.
In short, many of the large revisions to real GDP data in Figure 1 could have been avoided by employing better economic models of how PPPs evolve between ICP rounds.
Note: These are the views of the author and should not be attributed to the World Bank or any affiliated organisation.
Chen, Shaohua, and Martin Ravallion (2010), “China is Poorer than we Thought, but no Less Successful in the Fight Against Poverty”, in Debates on the Measurement of Poverty, Sudhir Anand, Paul Segal, and Joseph Stiglitz (eds), Oxford University Press.
Maddison, Angus and Harry Wu (2008), “Measuring China’s Economic Performance,” World Economics 9(2):13-44.
Ravallion, Martin (2010), “Price Levels and Economic Growth: Making Sense of the PPP Changes Between ICP Rounds”, Policy Research Working Paper 5229, World Bank.
Rogoff, Kenneth (1996), “The Purchasing Power Parity Puzzle”, Journal of Economic Literature, 34(2):647-668.
Summers, Robert, and Alan Heston (1991), “The Penn World Table (Mark 5): An Extended Set of International Comparisons, 1950-1988”, Quarterly Journal of Economics, 106: 327-368.
World Bank (2008a), Global Purchasing Power Parities and Real Expenditures. 2005 International Comparison Program, World Bank, Washington DC.
World Bank (2008b), Comparisons of New 2005 PPPs with Previous Estimates. (Revised Appendix G to World Bank 2008a), World Bank, Washington DC.
World Bank (2009), World Development Indicators, World Bank, Washington DC.