1 Productivity, Tradability, and The Great Divergence* Paul Bergin University of California, Davis, and NBER Reuven Glick Federal Reserve Bank of San Francisco Alan M. Taylor University of California, Davis, NBER, and CEPR Draft: March 2004 Abstract Long-run cross-country price data exhibit a puzzle. Today, richer countries exhibit higher price levels than poorer countries, a stylized fact usually attributed to the so-called the Balassa-Samuelson effect. But looking back fifty years, or more, we find that this effect virtually disappears from the international macroeconomic data. What is often assumed to be a universal property is, in fact, quite specific to recent times. What might explain this historical pattern? We argue for the usefulness of an alternative framework, where goods are differentiated by tradability and productivity. We employ a Ricardian continuum-of-goods model to explain this fact and find that endogenous tradability allows for theory and history to be consistent under a wide range of underlying productivity shocks. Moreover, this theory could illuminate the theory and empirics of economic growth in both the past and present. * Taylor gratefully acknowledges the support of the Chancellor’s Fellowship at the University of California, Davis. We thank David Jacks and Janine Wilson for excellent research assistance. For helpful comments we thank Stephen Broadberry, Nicholas Crafts, Robert Feenstra, Maurice Obstfeld, Mark Taylor, the NBER ITI group, and seminar participants at New York University and Columbia. All errors are ours.2 To repeat, no one scenario can alone explain such a fact. Many different, and mutually exclusive, sufficiency conditions could lead to it. The Penn [Balassa-Samuelson] effect is an important phenomenon of actual history but not an inevitable fact of life. It can quantitatively vary and, in different times and places, trace to quite different process, as we shall see. (Samuelson 1994, 206) 1 Introduction: Conventional Wisdom It is conventional wisdom today that richer countries have higher price levels than poorer countries. Figure 1a illustrates this idea, displaying the association of 1995 log price levels and of per capita incomes based on Penn World Table (PWT) data. That this is the consensus view is clear since similar charts appear in most textbook discussions of these phenomena (see, e.g., Krugman and Obstfeld, 2003, Figure 15.4). And although many rival theories exist to explain such an effect, there is also a broad consensus on how to explain this “stylized fact”: the standard story appeals to the Balassa-Samuelson theory, based on the divergence of productivity levels in a world of traded and nontraded goods. Having languished from time to time, these ideas are now enjoying a renaissance and are being incorporated into many new open-economy macroeconomic models. Of course, the apparent robustness of this story has proved to be of considerable relevance for many derivative conclusions in the theoretical and empirical literature. From work on sophisticated mathematical models of real exchange rates to the serious applied problem of judging differences in international living standards, the presumed correlation has had important economic and political ramifications. Since many of our PPP-based real income estimates for LDCs, past and present, often rely on extrapolations from the PWT based on this kind of relationship, and since such estimates are then used for such diverse tasks as evaluating long-run growth performance or allocating foreign aid, it is important that the patterns in the data be judged stable and predictable. This paper raises some challenges to this comfortable consensus on the sources of covariance in international prices and incomes. The first challenge is empirical. Whilst correlations such as those seen in Figure 1a are indisputably present in today’s data, one need only look back into the past to find evidence of weak or even negative correlations between national price levels and incomes per capita. After examining postwar data in great detail (and finding suggestive evidence going back several centuries) we conclude that the price-income correlation was not really very strong until the last three or four decades. This result is new and disturbing. What can explain it? This poses a second challenge to the prevailing view, and it is a theoretical challenge. We propose a new model of real exchange rates that builds on some key intuition in the Balassa-Samuelson theory, but which, by allowing for endogenous tradability, can also deliver time-varying correlations between incomes and prices as seen in the historical data. 1 A self-effacing Samuelson (1994) actually refers to the “Penn effect” in the original. This honors the laborious empirical work by the ICP/PWT team to collect the vast quantities of data that supplied the all-important evidence, and also the robustness of the empirical finding as compared to the fragile theoretical underpinnings he exposes. With apologies to the author, we adopt the more common usage here.3 Stylized Facts? In theories of the real exchange rate built around tradable and nontradable goods the central stylized fact to be “explained” is the effect noted by many scholars over the years, but highlighted by Bela Balassa and Paul Samuelson in their seminal papers from the year 1964: the tendency for poorer countries to have lower overall price levels than rich countries.2 Figure 1a shows a scatter plot of log relative price levels versus log relative income per capita for a cross section of 142 countries the year 1995. We performed an OLS regression on these data, (1) ln(Pi/PUS)=α+βln(yi/yUS)+εi. and the fitted values are shown as a straight line in the figure.3 We follow standard terminology and use the term “Balassa-Samuelson (BS) effect” as shorthand for a positive (and statistically significant) slope estimate β.4 In our example, the slope is 0.41 with a standard error of 0.04. The theory that Balassa and Samuelson constructed to explain this phenomenon is also now textbook material, and the simplest version runs as follows.5 Consider two countries, home and foreign, where foreign variables are denoted with an asterisk (*). Let there be two goods, traded (T) and nontraded (N), produced competitively in each country using only homogeneous labor as an input, with wages W and W* in each country. Let the labor productivity in each sector be AT