Spatial poverty traps?

wes lz; POLIcy RESEARCH WORKING PAPER 1862 Spatial Poverty Traps? loakethe difference between growth and contraction in living Jyotsna Jalan ,standards for otherwise Mlartin Ravallion :1denUical households? Apparenty so. Evidence of spatal poverty traps strengthens the case for investtn in the geographic capital of poor people. The World Bank Development Research Group December 1997  POLICY RESEARCH WORKING PAPER 1862 Summary findings Can place of residence make the difference between with latent heterogeneity (whereby hidden factors entail growth and contraction in living standards for otherwise that seemingly identical households see different identical households? consumption gains over time), yet identify the effects of Jalan and Ravallion test for the existence of spatial stationary geographic variables. poverty traps, using a micro model of consumption They estimate the model using farm-household panel growth incorporating geographic externalities, whereby data from post-reform rural China. neighborhood endowments of physical and human They find strong evidence of spatial poverty traps. capital influence the productivity of a household's own Their results strengthen the case - both for efficiency capital. By allowing for nonstationary but unobserved and equity - for investing in the geographic capital of individual effects on growth rates, they are able to deal poor people. This paper - a product of the Development Research Group - is part of a larger effort in the group to understand the geographic determinants of poverty and the implications for policy. The study was funded by the Bank's Research Support Budget under the research project "Policies for Poor Areas" (RPO 681-39). Copies of the paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Patricia Sader, room MC3-632, telephone 202- 473-3902, fax 202-522-1153, Internet address psader@worldbank.org. December1997. (32 pages) The Policy Research Working .Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Produced by the Policy Research Dissemination Center  Spatial Poverty Traps? Jyotsna Jalan and Martin Ravallion' World Bank Development Research Group, World Bank, 1818 H Street NW, Washington DC, 20433. The assistance and advice provided by staff of China's State Statistical Bureau-in Beijing and at various provincial and county offices-are gratefully acknowledged. We thank Jaesun Noh of EVIEWS for technical support. Useful comments on the paper were received from seminar participants at the World Bank, University of Maryland, College Park, and at the MacArthur Foundation/World Bank Workshop on Emerging Issues in Development Economics, Washington DC, July 1997. The financial support of the World Bank's Research Committee (under RPO 678-69) is also gratefully acknowledged.  i I  1 Introduction Consider two households living in different areas but identical otherwise. Suppose that one of the areas is less well endowed with physical, human and social capital-in short geographic capital-than the other. A spatial poverty trap can be said to exist if the household living in the better endowed area sees its standard of living rising over time, while the other does not. Various theoretical models have helped understand how such poverty traps can arise.2 If borne out by empirical evidence, spatial poverty traps suggest both efficiency and equity arguments for investing in poor areas, such as by developing local infrastructure or by assisting labor export to better endowed areas. Is it possible to test for spatial poverty traps? There are a number of problems.3 Aggregate growth empirics can test for divergence, whereby initially poorer areas grow at lower rates (following Barro and Sala-i-Martin, 1992). However, this is neither necessary nor sufficient for a spatial poverty trap, since geographic aggregates do not allow one to separate effects which are external to individuals from purely internal effects (Ravallion and Jalan, 1996). Suppose that one finds lower growth rates in areas with lower average wealth. This may reflect increasing returns to individual wealth, or geographic externalities, whereby living in a poor area lowers retums to individual investments. Aggregate geographic data cannot tell us which it is. Instead, cross-sectional micro data might be used to test for geographic effects on living 2 On the theoretical possibilities for a poverty trap (with and without externalities) in a neoclassical one-sector growth model see Azariadis (1996) and references therein. 3 For a review of the empirical literature on processes creating poor areas see Ravallion (1997). 2  standards at one point in time.4 However, to test for spatial poverty traps we need to identify dynamic effects, and to control for latent heterogeneity; that calls for longitudinal observations. Both household-level panel data and geographic data are clearly called for to have any hope of identifying spatial externalities in the growth process. The problems do not end there. The geographic effects that one might find in household panel data may well be spurious in that they arise solely because geographic variables proxy for omitted non-geographic, but spatially autocorrelated, household characteristics. For example, we might find that the average wealth of an area is positively correlated with growth rates at household level, controlling for individual wealth. But this may be because some household attribute relevant to growth, and positively correlated with average wealth, has been omitted. (Better own education may yield higher growth rates, be correlated with wealth, and be spatially autocorrelated. Then average wealth in the area of residence could just be proxying for individual education.) One might attempt to deal with this by adding variables. But one might reasonably expect considemble latent heterogeneity in any micro data. Nor is it sufficient to allow for latent fixed effects in the levels of consumption (as is common practice). We need an econometric model which allows for individual effects in consumption growth rates. At this point one might turn to the standard practice in panel data models of treating the latent heterogeneity as a time-invariant fixed effect (albeit a fixed effect in the growth rates, rather than the levels of consumption). However, this immediately wipes out any hope of identifying impacts of the time-invariant geographic variables of interest-of which there are 4 See, for example, Borjas (1995) on neighborhood effects on schooling and wages in the U.S., and Jalan and Ravallion (1997a) on geographic effects on chronic poverty in rural China. 3  likely to be many. In that case, the cure to the problem of latent heterogeneity leaves an econometric model which is unable to answer many of the questions we started out with. Nor, for that matter, is it obviously plausible that the heterogeneity in individual effects on growth rates would in fact be time invariant; common macroeconomic and geo-climatic conditions might well entail that the individual effects vary from year to year. This paper proposes an estimable microeconometric model of consumption growth which can identify underlying (including time-invariant) geographic effects while at the same time allowing for latent heterogeneity in household-level growth rates. We are able to test whether consumption growth rates at the farm-household level vary spatially after controlling for both observed and unobserved heterogeneity at the household level. Our theoretical model extends the Cass-Koopmans-Ramsey model of optimal consumption growth in a straightforward way to allow geographic effects on the marginal product of own capital, analogous to the role of knowledge externalities in the models of Romer (1986) and Lucas (1988). Our econometric model uses longitudinal observations of growth rates at the micro level collated with other micro and geographic data. The model allows individual effects with nonstationary impacts, following a specification proposed by Holtz-Eakin, Newey and Rosen (1988). Our model allows us to simultaneously deal with latent heterogeneity in growth rates (correlated with both the geographic and non-geographic variables), while still being able to retrieve estimates of the effects of time-invariant geographic capital on subsequent consumption growth at the household level. We believe that the methodology proposed here for micro-growth empirics has potentially wide applications in understanding the processes whereby 4  some individuals do so much better than others over time. We implement the approach using data for rural areas of southern China over 1985-90. There is a widely held view amongst China scholars and observers that many of the poorer rural areas-typically in more remote inland provinces-have shared rather little in the country's overall economic growth since reforms began,5 and there is supportive evidence of rising inter-regional inequality and divergence.6 Anti-poverty policies in China since the mid-1980s have relied heavily on public investment in lagging "poor areas" (Leading Group, 1988; World Bank, 1992; Jalan and Ravallion, 1997b). This is also a setting in which there appears to be very little migration of entire households from one rural area to another; the limited migration that is observed is the export of labor surpluses, primarily to urban areas, and would only rarely entail that the whole household moves.7 Thus we can abstract from the complications that arise in identifying geographic effects when location is endogenous. The following section outlines our model of consumption growth. Section 3 describes our data while section 4 presents our results. Section 5 summarizes our conclusions. 5 For example: "As China's economic miracle continues to leave millions behind, more and more Chinese are expressing anger over the economic disparities between the flourishing provinces of China's coastal plain and the impoverished inland" (New York Times, Dec. 27, 1995, p.1.) 6 Ravallion and Jalan (1996) provide evidence that counties with lower initial average wealth saw lower subsequent rates of consumption growth. We also find evidence of spatial externalities. However, our estimation method (following standard methods in the literature on cross-country growth empirics) did not exploit the panel nature of our data, did not allow for latent heterogeneity, and did not identify the specific aspects of geographic capital that matter to the divergence. ' There are various administrative and other restrictions on migration in China, including registration and residency requirements. For example, it appears to be rare for a rural worker who moves to an urban area to be allowed to enrol his or her children in the urban schools. 5  2 The micro model of consumption growth To motivate our empirical work we extend the standard Cass-Koopmans-Ramsey model in a natural way to include production by the farm-household and allow geographic externalities in the production process. We then outline our econometric model and the estimation method. 2.1 Theoretical model Analogously to the role of firm-specific knowledge and external (economy-wide) knowledge in the Romer (1986) model, we hypothesize that output of the farm household is a concave function of various privately-provided inputs, but that output also depends positively and non-separably on the level of geographic capital, as described by a vector of geographic variables representing physical and social characteristics of the area of residence!8 We make the standard assumption that the household maximizes the utility integral: 0o , 1-a tl~6 Pd (1) where a is the intertemporal elasticity of substitution, C is consumption (the logarithm of which is denoted c), and p is the subjective rate of time preference. The household operates a farm which produces output by combining labor and own capital (which can be interpreted as a composite of land, physical capital and human capital) under constant returns to scale. However, the household's farm output also depends on a vector of geographic variables, G, reflecting s The model outlined here can be extended to allow (inter alia) depreciation of capital and exogenous rates of technological progress and population growth, but it will preserve this feature. 6  external effects on own-production. Output per worker or person is F(K, G) where K denotes capital per worker. Output can either be consumed or invested: F[K(t), G(t)] = C(t) + K'(t) (2) The derivation of the optimal rate of consumption growth then follows standard methods for dynamic optimization, as outlined in an Addendum available from the authors. It can be shown that the optimal rate of consumption growth satisfies: C /(t) = [FK(K, G) - p]/a (3) The key feature of this model for our purpose is that geographic externalities influence consumption growth rates at the farm-household level, through effects on the marginal product of own capital. The model permits values of G such that the optimal consumption growth rate is negative; given G, output gains from individually optimal investments are not sufficient to cover the discount rate and so consumption falls. Whether that is anything more than a theoretical possibility will be tested in the following sections. There are other ways in which geographic effects on consumption growth might arise, not captured by the above model. For example, we could also allow geographic variables to influence utility at a given level of consumption, by making the substitution parameter and the discount rate functions of G. Or one might introduce borrowing constraints which differ from one area to another. While our empirical model will allow us to test for geographic effects on consumption growth at the micro level it will not allow us to identify the precise mechanism linking area characteristics to growth. 7  2.2 Econometric model The theoretical model above motivates an empirical model in which the growth rate of household consumption depends on both its own capital and on geographic capital. To allow for differences in the quality and quantity of family labor (given that labor markets are thin in this setting) we let education and demographics influence the marginal product of own capital; these may also influence the rates of intertemporal substitution and/or time preference. However, we also allow for aspects of own capital, and other shift parameters in utility and production functions, which one cannot hope to fully capture in the data available. So there is latent heterogeneity in consumption growth rates. Furthermore, it is possible that these omitted variables will be correlated with the geographic variables, leading to biases in OLS estimates of the parameters of interest. We have a random sample of N households observed over T dates, where T is small and N is large. Our empirical specification is interpretable as a linearization of equation (3) giving: \cit = a + PXft + 4z, + &i (i=1,2,..,N; t=3,..,T) (4) where Ac,t is the growth-rate of consumption of household i in time period t, xi, is a ( kx 1) vector of time-varying explanatory (geographic and household) variables, zi is a (p x 1) vector of exogenous time-invariant explanatory (geographic and household) variables, and