WPS7056 Policy Research Working Paper 7056 Agricultural Productivity, Hired Labor, Wages and Poverty Evidence from Bangladesh Shahe Emran Forhad Shilpi Development Research Group Agriculture and Rural Development Team October 2014 Policy Research Working Paper 7056 Abstract This paper provides evidence on the effects of agricultural surplus and deficit households. Taking rainfall variations as productivity on wage rates, labor supply to market oriented a measure of shock to agricultural productivity, and using activities, and labor allocation between own farming and subdistrict level panel data from Bangladesh, this paper wage labor in agriculture. To guide the empirical work, this finds significant positive effects of a favorable rainfall shock paper develops a general equilibrium model that underscores on agricultural wages, labor supply to market work, and the role of reallocation of family labor engaged in the produc- per capita household expenditure. The share of hired labor tion of non-marketed services at home (‘home production’). in contrast declines substantially in response to a favorable The model predicts positive effects of a favorable agricultural productivity shock, which is consistent with a case where productivity shock on wages and income, but the effect on labor-deficit households respond more than the labor- hired labor is ambiguous; it depends on the strength of real- surplus ones in reallocating labor from home production. location of labor from home to market production by labor This paper is a product of the Agriculture and Rural Development Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at fshilpi@worldbank.org. 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 views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Agricultural Productivity, Hired Labor, Wages and Poverty: Evidence from Bangladesh 1 Shahe Emran IPD, Columbia University Forhad Shilpi World Bank Key Words: Agricultural Productivity, Home Production, Market Work, Wage, Hired Labor, Labor Supply Response, Poverty JEL Classification: O13, J22, J43, Q10 1 We would like to thank Will Martin, Luc Christiaensen, Claudia Berg, Fenohasina Maret for comments on earlier versions of the paper. We acknowledge Yan Sun for excellent research assistance. (1) Introduction A common approach to analyzing the e¤ects of agricultural growth on rural poverty has been to estimate its impacts on rural wages (see, among others, Foster and Rosenzweig (2004), Lanjouw and Murgai(2009)). This is motivated by the observation that a signi…cant proportion of poor people – endowed mostly with labor but few other productive assets (e.g. landless) –participate in the rural wage labor market and wage income is the main source of income for them. The available evidence on rural labor markets in developing countries, however, shows that the extent of wage employment is limited (Rosenzweig surplus labor’ in the form of under- and (1988)), and there is a substantial amount of ‘ unemployed family labor.1 In many developing countries, poor households are poor because most of their labor endowment is employed in home-based non-marketed services activities with very low returns, not because they are (openly) unemployed. In the presence of low productivity home production, the poverty impact of agricultural productivity may depend on how allocation of labor from home production to own farming and wage labor changes in response to agricultural productivity growth.2 The reallocation of labor from low productivity home production to other market oriented activities has been emphasized as a hallmark of long-run structural change of an economy in the literature (Laitner (2000), Buera and Kaboski (2012)). We provide evidence on the impacts of agricultural productivity changes on wage, labor supply (hours devoted to market oriented activities), and labor allocation between own farming, and wage labor in agriculture. To derive testable hypotheses about the impact of 1 For instance, Rosenzweig and Foster (2010) …nds that 20 percent of rural labor force in India is ’surplus’. 2 This is more so in African countries where wage employment in agriculture and non-agriculture in rural areas is very limited (see Davis, Guiseppe and Zezza (2014)). The reallocation of labor in response to agricultural productivity growth in African case is between home production and own farming. 1 agricultural productivity growth on labor allocation and wages, we develop a simple model where household members can be employed in three types of activities: own farming, wage labor in agriculture and household production (non-traded and non-marketed services).3 There are two types of households in the farm economy who di¤er in two dimensions: some households are endowed with more labor, and the technology for home production (degree of diminishing returns to labor) is also di¤erent between the two groups. The model yields the standard predictions of a positive e¤ect on agricultural wages, and a reallocation of labor from home production to market work in response to a positive agricultural productivity shock. There are two interesting implications of the model particularly relevant for our empirical analysis: (i) while wages respond positively to an agricultural productivity shock, the extent of the wage increase is lower the higher is the reallocation of labor from home production to market work (own farming and wage labor), (ii) the response of the quantity of hired labor depends critically on the di¤erences in the strength of diminishing returns to labor in home production across labor-surplus and labor-de…cit households. In particular, it is possible that the amount of labor hired through the labor market may go down in response to an increase in agricultural productivity. Thus a focus on the response of wage and hired labor alone may lead to misleading conclusions regarding the impacts of agricultural growth. We test the predictions of the model using an upazila (subdistrict) level panel data set from Bangladesh. To understand the implications of agricultural productivity, we exploit variations in rainfall across upazilas and over time, and implement a procedure that focuses 3 Given the focus on the interactions between home production and agriculture, we abstract away from non-farm production in a village. We, however, emphasize that the predictions regarding agricultural wage and labor allocation are robust; they remain intact in a more general model including non-farm production. 2 on the e¤ects of rainfall shocks in reduced form regressions on the outcome variables (wage, employment in own farming and hired labor, hours worked for market oriented activities, and per capita consumption) and also on the measure of agricultural productivity (crop yield). The evidence from the reduced form regressions is su¢ cient to test the predictions of the theoretical analysis which relies on the fact that rainfall variations can be interpreted as shifts in the production function, because rainfall is a major determinant of crop yield in Bangladesh (Sarkar et. al. (2012), Bhowmik and Costa (2012)). We also provide an in- strumental variables interpretation of our estimates, using rainfall variations across upazila and over time (relative to mean) as an instrument for crop yield (rice yield). The regres- sions include upazila …xed e¤ects to remove the in‡uences of time invariant unobserved area characteristics, and year …xed e¤ects to wipe out the common price (international) and other macroeconomic shocks. To be as clinical as possible, we allow for time varying direct impacts of these factors by including interaction of a ‡ood-prone area dummy and travel time to the two metropolitan cities (Dhaka and Chittagong) with the time trend. We included an extensive set of control variables to account for time varying direct e¤ects of in- frastructure and other area characteristics. Empirical estimation issues and strategy to deal with them are discussed in detail in Section 3. It is worth emphasizing that while rainfall shocks have been used for identi…cation in a variety of contexts, agricultural productivity is probably among the most natural contexts where rainfall can provide reasonable identifying variations ((Foster and Rosenzweig (2004), Adhvaryu, Chari and Sharma (2013), Bruckner and Ciccone (2011)). The regression estimates reported later show that a positive rainfall shock has a sig- ni…cant positive e¤ect on wages; a one percent increase in rainfall (relative to the mean) 3 increases wages by about 0.46 percent. The e¤ect on hired labor is, in contrast, negative and statistically signi…cant; a one percent increase in rainfall reduces hired labor by 0.73 percent. The negative response of hired labor is consistent with the case where the la- bor reallocation from home production by labor de…cit households is stronger than that of labor-surplus households. Our results also indicate that households increase hours sup- plied to the market-related activities in response to a positive rainfall shock, thus providing additional evidence of reallocation of labor from home production. When interpreted as instrumental variables estimates of the e¤ects of productivity increase, the estimates show substantial impact of an increase in rice yield on wage, hired labor, and labor supply to the market activities. The rest of the paper is organized as follows. Section (2) develops a model of the farm economy with a focus on the role of home production and derives three propositions on the e¤ects of agricultural productivity on the labor market. The following two sections present the empirical strategy and data, respectively. Empirical results are discussed in section 5. The paper is concluded in the …nal section. (2) Agricultural Productivity in a Farm Economy We construct a simple model of a farm economy consisting of two (types of) households (h and k ). Each household owns A units of agricultural land, but they di¤er in terms of 0 4 the endowment of labor, household h (L0 0 h ) with more labor than household k (Lk < Lh ): The households produce two goods: food (agriculture) and a home good. The households also di¤er in a second dimension, they have access to di¤erent technologies for home good production. 4 A richer model where households di¤er in land endowment and skilled labor also generate the same st of qualitative conclusions. 4 Households consume three goods/services: a home good (d), and two market goods (food (f ), a non-farm good (m)). Both food and non-farm goods are assumed to be internationally traded, and we take the food commodity as the numeraire. The assumption that both food and non-farm goods are tradable implies that their prices are pinned down at the international market, which is useful for abstracting away from the demand side factors, and focusing only on the supply side responses. Assuming identical preferences, the utility function for households in the village is the following: U = u(cf ; cm ; cd ) where cf is consumption of food, cm is consumption of the non-farm good. We as- sume that utility function u(cf ; cm ; cd ) takes the Cobb-Douglas form and 'i is the share of good/service i: The budget constraint for the market goods can be stated as: Y = c f + Pm c m where Pm is the price of the non-farm good, the agricultural good is the numeraire, i.e., Pf = 1 and Y is total market income in the village. The equilibrium in the farm economy is characterized by labor market clearing and an external balance condition (export food and import non-farm good at world prices). By Walras law we can ignore the external balance condition. Wages are thus determined by the labor market clearing condition. (2.1) Labor Demand for Market Work in Agriculture For workers in the farm economy, there are three employment options: (i) home pro- 5 duction, (i) family owned farm, (iii) other farms.5 Note that we include both own farming and working as hired labor for others as market work, because the focus is on reallocation from home production that includes all non-market economic activities. Households pro- duce food using land and labor with the same CRS technology. The food output by each household can be described as: 1 Qf i = F (A; lf i ) = A lf i for i = h; k where represents total factor productivity in food production, A is the endowment of land which is assumed to be …xed, and lf i is the labor used in food production by household i. The demand for labor in agriculture does not vary across households given that they face the same technology and prices, and labor demand for each household can be derived as: 1 (1 ) 1 lf h = lf k = A[ ] (1) w Total labor demand in farming is then: 1 1 1 lf = lf h + lf k = 2(1 )41 ( ) (2) w 5 Note that we do not include production of the non-farm good in the model, as the focus of this paper is on the e¤ects of agricultural productivity increase on wage and hired labor in agriculture when there is potentially signi…cant labor supply response through reallocation of labor from home production . A richer model that includes non-farm produced in the village (possibly with skilled labor) yield similar qualitative conclusions. We, however, will discuss the implications of the nonfarm production when discussing the em- pirical strategy, as general equilibrium responses in the non-farm sector have bearings on the identi…cation and interpretation of the results. For an extended model that focuses on production of non-farm goods and allows for heterogeneity in the skill of labor and nontradability of the non-farm good, please see Foster and Rosenzweig (2004), and Emran and Shilpi (2014). 6 1 Where 41 = A[1 ] . (2.2) Labor Supply for Market Work in Agriculture In addition to using labor in agriculture, each household also engages in the production and consumption of a home good (d). The production function for the home good is of the following form: i Qdi = ldi for i = h; k and 0 < i <1 Thus the curvature of the home good production function di¤ers across households. The relevant opportunity cost of home production for both types of households is the market wage rate w. The marginal condition determining the optimal use of labor in home production can be expressed as (assuming = 1; for simplicity): i 1 i ldi = w for i = h; k (3) Since labor allocated to home production varies inversely with the wage (equation 3), the supply of labor for market work can be written as: 1 (1 i) i @Li Li (w) = L0 i ldi = L0 i ; with Lw i = > 0 for i = h; k (4) w @w The model set up above generates an upward sloping labor supply function for market work. The model is general enough that the home good can also be interpreted as leisure, but it avoids the awkward possibility of a backward bending supply curve of labor in a low income village economy.6 An alternative model is where there is (open) unemployment, and 6 It is not realistic to expect that people would like to consume more ‘leisure’ when managing three 7 labor supply responses occur primarily at the extensive margin. The formulation adopted here is attractive, because explicit unemployment is not high in rural areas of developing countries, and poor people are poor not because they are unemployed (consuming leisure), but because they work long hours in extremely low productivity activities such as foraging. Those low productivity non-market economic activities are modeled as home production in our model.7 Note also that our de…nition of labor supply to market work corresponds to the traditional de…nition of total labor supply that includes self-employment on own farm as well. The distinction between market work and home production in our case is that market work consists of all work whose output can be and is usually transacted in the market. For instance, labor spent on producing rice that is consumed at home is considered as market work since rice is widely traded in the market. Home production on the other hand consists of services (e.g. meal preparation, child care or simply leisure) which is consumed at home and is not usually sold in the market. The span of home production is much wider in villages in developing countries, because of limited development of markets, for example, the markets for child care and prepared meals are missing in most of villages in developing countries. The supply of labor to market oriented activities (as opposed to home production) by each household depends on agricultural productivity indirectly through its e¤ects on wage. The larger is the value of i, the larger is the magnitude of supply response of labor for market work. A rise in wage draws labor out of home production and into market work. meals a day is a challenge. 7 The model outlined here does not allow the possibility of labor market rigidity that may arise from socio-cultural practices (e.g purdah restriction) which in turn lock household workers (esp. female) to low productive home production activities. Welfare gain from moving workers from home production to market work would be much higher in the presence of such rigidities. However, overall conclusions of the model regarding home versus market work are valid under that alternative scenario as well. 8 Given the assumption of Cobb-Douglas form for the utility function and labor used in home production being determined by the marginal condition in equation (3), demand for market goods can be expressed as: 'i Y ci = , i = f; m (1 'd )Pi Note that although the home good enters into the utility function in a non-separable way, production and consumption of the home good is determined by the marginal condition in equation (3). This is due to the fact that the production function for the home good displays diminishing returns. This in turn ensures an unambiguous upward sloping labor supply function for market work. In the standard labor-leisure models, the production function for leisure (home production) is assumed to have constant returns; one unit of produces’one unit of leisure. In that case, the supply side does not pin down the labor ‘ production (and consumption) of the home goods, the demand side plays a critical role. An increase in wage increases demand for the home good assuming it is a normal good, and thus the curvature of the labor supply function for market work depends on the relative size of income (negative) and substitution (positive) e¤ects. The assumption of decreasing returns to scale in the home (leisure) production is however more plausible in the context of developing countries. Our empirical analysis later con…rms that labor allocated to market activities does increase with an increase in wages and/or agricultural productivity. (2.3) Market Clearing for Labor Setting labor demand equal to labor supply, the equilibrium condition in the labor market can be expressed as: 9 1 1 1 lf = lf h + lf k = 242 ( ) = Lh (w) + Lk (w) (5) w where 2 = 41 (1 ): Proposition 1:Given the assumptions that food is produced under CRS technology using land and labor, and the home good is produced under decreasing returns to scale (DRS) technology using labor alone, a positive productivity shock in agriculture (i.e., a higher ) results in an increase in the wage rate; the higher the response of labor supply to the wage, the lower is the change in the equilibrium wage rate. Proof: The market clearing condition in equation (5) can be used to derive the following result: 1 1 1 @w 2 2 (w ) = 1 1+ >0 (6) @ (Lw w 1 h + Lk ) + 2 2 (w ) The …rst part of the result above follows from the fact that the denominator in equation (6) is positive as long as i < 1. The second part derives from the fact that the denominator is a positive function of the supply response of households for market work to wage change (captured by Lw w h + Lk ). (2.4) Response of Hired Labor The increase in the wage rate following an increase in agricultural productivity discussed in proposition (1) above has been a focus of empirical work on the e¤ects of agricultural growth on poverty. A central point of this paper is that the e¤ects on wages are only half of the story, as the incomes of the poor rural households also depend on whether they are largely dependent on low productivity home-based activities. If the supply of labor 10 to market work (including own farming) is sensitive to productivity and wages, then the primary margin of adjustment may be reallocation of labor from home production to market production. A related important implication of high labor supply elasticity is that the wage response observed in the data will be smaller the higher is the labor supply response, but total income may in fact go up much more because of the labor supply response. Thus if one focuses on the wage response alone, it might give us a partial estimate of the impact of an agricultural productivity increase on household welfare which may be signi…cantly biased downward. An immediate solution to this seems to be to supplement the wage analysis with evidence on the response of labor supplied to the market, i.e., hired labor. If agricultural productivity increase shifts the demand curve for labor, then we would expect the use of hired labor to go up. However, this simple intuition turns out to be misleading in a model with home production, because the response of hired labor depends on the di¤erences in the technology of home goods production across labor-rich and labor-de…cit households. How does hired labor in farming respond to a productivity shock? Since the labor endowment of household k is smaller than that of household h, household h is a net seller of labor and household k is a net buyer of labor. Let lw be the labor hired for farming work (by household k ) which can be written as: 1 1 1 lw = 2 [ ] Lk (w) w " # 1 (1 k) 1 1 1 k = 2 [ ] L0 k (7) w w 11 The response of hired labor by household k with respect to an increase in productivity is derived as follows: ( 1 ) w (1 k) @l 1 1 1 1 1 1 @w 1 1 1 1 1 k = 2 [ ] 2 [ ] + (8) @ w w@ w 1 k w The …rst term in the right hand side of equation (8) above is the direct productivity e¤ect that increases demand for labor in agriculture, and the last term combines the general equilibrium e¤ects through higher wages on the demand for labor in both home production and agricultural production. Proposition 2: In a rural economy where there is heterogeneity in households’endow- ments of labor, the e¤ects of agricultural productivity on hired labor depend on the labor supply responses of the labor-surplus and labor-de…cit households. Assuming constant re- turns to scale in agriculture and decreasing returns to labor in home production, we have the following results: (i) When labor supply response of labor-de…cit households with respect to a change in the wage is larger than that of labor surplus households ( k > h) then an increase in agri- cultural productivity leads to a decrease in hired labor. (ii) When the labor supply response of labor-de…cit households with respect to the wage is smaller than that of labor surplus households ( k < h) ; then an increase in agricultural productivity leads to an increase in hired labor. @w dlw Proof: Substituting for from equation (6) above into the equation for (i.e., @ d equation (8)) and rearranging terms we get the following: 12 1 1 1 1 dlw 2 ( w ) 1 (Lw h Lw k) = 1 1 (9) d w w [ w(Lh + Lk ) + 2 2 1 (w ) ] Now note that " 1 1 # (1 h) (1 k) 1 1 h 1 k Lw h Lw k = w 1 h w 1 k w The proof then follows from the fact that Lw h Lw k > 0 if h > k and Lw h Lw k < 0 if h < k. The intuition behind the results in proposition (2) re‡ects the fact that an increase in agricultural productivity increases returns to own-farming, and induces both types of households to substitute away from home production which is subject to decreasing returns to labor. Part (i) of proposition (2) shows that when Lw k is quite large, the induced supply response of de…cit household could displace hired labor. It is interesting to note that even though the amount of hired labor can go down in response to a productivity increase in agriculture, the wage response is always positive. For low value of supply response of the de…cit household, substitution between home production and own farming is smaller, leading to a higher wage and more hired labor. Note that in the above framework, if there are no adjustments at the margin of home production (which includes leisure), and thus Lw w h = Lk = 0, the impact of agricultural productivity increase on hired labor is zero. While a zero response of hired labor can also result from a coincidence where h = k , admittedly this is not a very realistic case. The response of hired labor to agricultural productivity can thus be a fruitful metric for gauging the importance of home production and labor supply 13 response. This is especially important for empirical analysis because the household surveys in developing countries usually lack reliable information on home production activities. Although the response of hired labor to agricultural productivity shock is ambiguous a priori and thus can lead to misleading conclusions about the poverty impact of agricultural productivity changes, note that the response of total labor devoted to market work is positive under the plausible assumption that h > 0; k > 0. In the empirical analysis, we thus look at both hired labor and total labor devoted to market production as opposed to home production. Proposition 3: Regardless of its impact on hired labor, an increase in agricultural productivity increases total income in a village. The increase in village income is higher, the higher is the labor supply response with respect to wage (i.e, larger values of k and h ). Proof: The net income of each household is the sum of land and labor income from market activities. The total net income of the two households is thus: 1 1 1 Y =2 41 f g w 1 Where 41 = A[1 ] and Y is the total village income. The change in village income in response to an agricultural productivity shock can be derived as: dY @Y @Y @w = + d @ @w @ 14 @w Substituting from equation (6) for @ and rearranging terms, we have the following: h 1 i 1 1 dY Y 242 (w ) + Lw h + Lw k = >0 (10) d 1 1 1 242 ( w ) + [Lw w h + Lk ] It is easy to check from equation (10) above, the increase in income is higher when the labor supply responses of the households are higher. In this model there are thus two sources of income gains following an increase in agricultural productivity: a reallocation of labor from home production to agriculture, and a higher productivity of agricultural activity. In other words, the income and poverty impacts of agricultural productivity will be larger when households can increase their labor supply to market work, which does not necessarily imply an increase in hired labor through the market as shown in proposition (2) earlier. This result is important because it underscores the importance of looking at the e¤ects on both price (wage) and total labor supplied to the market (not only hired labor) to understand the e¤ects of an agricultural productivity increase on the poor. (3) Empirical Framework To estimate the e¤ects of agricultural productivity growth on wages, labor allocation across own farming and hired labor, and household consumption, we construct a subdistrict (upazila) level panel data set using three rounds of Household Income and Expenditure Surveys (HIESs). To test the theoretical predictions in propositions 1-3, a natural regression speci…cation is: Oijt = j + t + jt + 1 Zjt + "ijt (11) where i indexes the outcome variables (e.g. share of employment in an activity, wage, 15 per capita household consumption expenditure etc), j denotes upazila, Oijt is the outcome variable i, j and t denote the e¤ects of upazila and year speci…c factors respectively. Our focus variable is jt which measures agricultural productivity, Zjt is a vector of upazila characteristics and "ijt is the error term. Estimation of the impact of agricultural produc- tivity on employment and wages however presents some di¢ culties. Unobserved upazila characteristics when correlated with both wage/employment and agricultural productivity may create spurious correlations, and provide biased estimates of the e¤ects of agricultural productivity change. For example, consider the heterogeneity in access to markets due to geographic location; an upazila which is closer to the metropolitan cities (Dhaka and Chit- tagong) will have higher agricultural productivity (higher demand, and cheaper and more reliable supply of inputs such as fertilizer and pesticide) and higher wages (because of em- ployment opportunities in the cities). Thus when we regress wages on crop yield, we might …nd a positive "e¤ect", both driven primarily by di¤erences in access to markets across di¤erent upazila. It is, in general, not possible to control for all such potential confounding factors in a regression speci…cation, and thus OLS results may be misleading. An important advantage in our application is that we construct a panel data set, which allows us to use upazila …xed e¤ects ( j in the regression equation (11) above) to remove the e¤ects of all time invariant but unobserved upazila characteristics. The year …xed e¤ects ( t ) control for any macro economic and international shocks (including commodity price shocks) that may have a¤ected both agricultural productivity and outcomes of our interest.8 In the empirical analysis, we follow a two step procedure: …rst, a reduced form regres- sion of an outcome variable (for example, wage) on the instrument, and second, a reduced 8 The year …xed e¤ects will control for any general equilibrium e¤ect common to all households (e.g. prices). 16 form regression of the productivity measure (yield per acre) on rainfall. This two-step pro- cedure has some important advantages in our application. First, the reduced form estimates of the e¤ects of rainfall on the outcome variables such as wages and employment in agri- culture are of interest on their own; for example, they provide us evidence on the potential bene…ts of increased irrigation investment on the rural economy. Second, when the focus is on the e¤ects of productivity increase in agriculture, one can interpret the variations in rainfall as variations in the parameter in the model. Finally, with a focus on the standard measures of agricultural productivity such as crop yield, and rainfall as an instrument, the reduced form estimates of rainfall on wages, employment and consumption are still useful. Because they provide evidence on the existence of a causal e¤ect of higher crop yield which is not subject to weak instrument bias (Chernozhukov and Hansen(2008)).9 We estimate the following reduced form regressions: Oijt = j + t + 1 Rjt + 1 Zjt + "ijt (12) Vjt = j + t + 2 Rjt + 2 Zjt + jt (13) where Rjt is the annual rainfall in upazila j and Vjt is the measure of productivity. In empirical estimation, rainfall variable is expressed in logarithm. Thus our empirical model with upazila and year …xed e¤ects provides estimates of the impact of rainfall shock on the growth of outcome variables. Rainfall shock for a upazila is de…ned as the deviation of rainfall in any year from its mean over all the years. A positive coe¢ cient of rainfall ( 2 > 0) for instance in the yield regression means that an increase in rainfall over its mean 9 We, however, emphasize that the main results of this paper do not depend on the exclusion restriction on rainfall; what we need is that rainfall a¤ects productivity signi…cantly. 17 level (a positive rain shock) increases rice yield. We implement a …xed e¤ects estimation procedure that removes the upazila level un- observed …xed factors by de-meaning all variables in the regression. Such demeaning may, however, exacerbate the attenuation bias as it is likely to magnify any measurement error in the measure of agricultural productivity (Griliches (1963)). We use an instrumental variable approach to remedy the attenuation bias and also the possible biases due to any other omitted variables not taken care of by the upazila and year …xed e¤ects. Following a large and mature literature, agricultural productivity is measured by crop yield (Foster and Rosenzweig (2004), Adhvaryu, Chari and Sharma (2013)). More speci…cally, we use rice yield per acre, as rice is the predominant subsistence and cash crop in Bangladesh. We exploit rainfall variations as an instrument for rice yield. Rainfall is found to a¤ect agricul- tural yields in both developed and developing countries, and hence is considered widely a credible instrument for agricultural yields (Foster and Rosenzweig (2004), Adhvaryu, Chari and Sharma (2013), Rajan and Ramcharan (2010), Bruckner (2012)).10 To ensure that rainfall primarily captures variation in agricultural productivity, we include an appropriate set of controls in Zjt : One may argue that the agricultural labor market will be in‡uenced by the e¤ects of rainfall on non-farm activities. For example, construction employment may rise with higher rainfall if rainfall leads to ‡ooding and destruction of the infrastructure which results in higher repair and reconstruction work. In so far as agricultural labor is also employable in construction work, this will have an impact on the agricultural wages. Flooding and destruction may, on the other hand, lead to a 10 Rainfall has been used as an instrument in a variety of contexts ranging from civil war to foreign aid ‡ ow in the recent economic literature. While there are limitations to relying on rainfall variations for identifying information in many applications, rainfall variations are probably the most natural candidate for exogeneous variations in agricultural productivity, especially in developing countries. 18 negative correlation between rainfall and non-farm labor demand if it disrupts production activities. The positive and negative e¤ects of nonfarm sector on the agricultural labor market may, in some cases, largely o¤set each other. One can include a ‡ood-plain dummy to control for such e¤ects. Note that the negative e¤ects of ‡ood caused by heavy rainfall, especially, on prices and wages will depend on the location of an upazila, because access to urban markets provides a cushion against such shocks. Travel time to the urban markets can be used to account for such heterogeneity. Travel time would also capture the spatial variations in the prices of tradables, because it is a reliable proxy for the transport and other marketing costs. Since both travel time and ‡oodplain dummy are time-invariant (or can change only very slowly over years), they are subsumed by upazila …xed e¤ects. As a conservative strategy, we allow for time varying e¤ects of these two variables, and include their interactions with time trend in the regressions. We also include the proportion of 11 households in a upazila with electricity as a control. Availability of electricity may foster non-farm activities that are less susceptible to the weather, and may have di¤erential e¤ect on part of the agricultural labor market through substitutability of labor. To capture changes in labor endowment, we control for upazila population, and proportion of active labor force with secondary or above education (human capital). We also control for upazila population in 1991 (initial condition) interacted with the time trend. Note, however, that it is a conservative speci…cation, because by controlling for variations in labor endowment across upazilas and over time, we also deny the possibility that agricultural productivity changes a¤ect the population in a village. A …nal issue for the empirical speci…cation is that rainfall is expected to have signi…cant 11 Our empirical results are robust to controls for agglomeration economies such as area share in total industry employment in 1991. 19 e¤ect on rice yield only if the upazila is predominantly rural in its economic activity. We thus exclude upazilas located in two main metropolitan areas from our sample. In addition, we control for the share of "urban" households in total households in the upazila. (4) Data For the empirical analysis, we combine di¤erent data sources to de…ne a upazila (sub- district) level panel data set covering the period 2000 to 2009/2010. Our main data source for the outcome variables (wage, employment in di¤erent activities and household con- sumption expenditure) is three rounds of the Household Income and Expenditure Surveys (HIESs), available for 2000, 2005, and 2010. The HIESs are based on a nationally represen- tative sample of households.12 These surveys are conducted primarily for the estimation of poverty incidence and thus provide reliable information on household economic activities, per capita household expenditure and regional price de‡ators. Upazila level data on out- come and explanatory variables are generated from the HIESs using appropriate population weights. As noted earlier, productivity growth in agriculture is measured by growth in crop yield. The predominant crop in Bangladesh is rice/paddy, of which three di¤erent types (Boro, Aman and Aus) are grown.13 The o¢ cial source of agricultural statistics provides yield data at district level, and unfortunately there are no estimates at the upazila level.14 The source of the yield data used in this paper is the community part of the HIES. We 12 The sample sizes for HIES are 7,440 in 2000 10,080 in 2005 and 12,240 in 2010. 13 High yielding variety of Boro rice now accounts for more than half of rice production (56%). Aman is the next important crop accounting for 44% of rice production. Yields of both of these varieties are much higher than Aus. 14 These data are actually reported at old (and much larger) district level – there are about 20 old districts. With newly created districts, there are now 64 districts in Bangladesh. The source of these data are Statistical Yearbooks published by the Bangladesh Bureau of Statistics. 20 de…ne rice yield per acre as the average of yields of Boro, Aman and Aus rice. The upazila (subdistrict) level yields are the average over villages surveyed within a upazila. Since the number of villages within a upazila are limited, the estimated yield may involve signi…cant measurement error. We compared yield growth estimates from HIES aggregated to the district level to the corresponding estimates from the o¢ cial agricultural statistics, and they show comparable growth during the decade of 2000. The yield estimates at the upazila (subdistrict) level used in this paper thus are useful as a measure of productivity. Rainfall data are drawn from Bandyopadhyay and Skou…as (2012). The original data on rainfall come from the Climate Research Unit (CRU) of the University of East Anglia. The CRU reported estimates of monthly rainfall for most of the world at the half degree resolution from 1902 to 2009. The CRU estimation combines weather station data with other information to arrive at the estimates.15 To estimate the sub-district (upazila/thana) level rainfall from the CRU data, Bandyopadhyay and Skou…as (2012) use area weighted averages.16 Travel times to the metropolitan cities are computed using GIS software and the road network from mid-1990s. Data on ‡ood prone areas are drawn from the Bangladesh Water Board database. All population variables are drawn from the population censuses. Over the years, a number of larger upazilas were split to form new upazilas, thus in- creasing the total number of upazilas from 486 in 1990 to 507 in 2000 to 543 in 2010. We use upazila maps to identify the borders of upazilas over time and matched all upazilas 15 Previous versions of the CRU data were homogenized to reduce variability and provide more accurate estimation of mean rain at the cost of variability estimation. The version 3.1 data is not homogenized and thus allows for better variability estimates. The estimates of rainfall near international boundaries are not less reliable as compared with those in the interior of the country, as the CRU estimation utilizes data from all the weather stations in the region. 16 For example if an Upazila/thana covers two half degree grid cells for which CRU has rainfall estimates, then upzila/thana rainfall is estimated as the average rainfall of the two grid-cells, where the weights are the proportion of the area of the upazila/thana in each grid-cell. For details, please see Bandyopadhyay and Skou…as(2012). 21 in 2000 and 2010 to 1990 upazilas. The upazila level panel is de…ned using 1990 upazila boundaries. The number of upazilas in the sample used for econometric analysis is, how- ever, smaller (355 upazilas with data for more than one year), as we drop upazilas located in the two largest metropolitan areas (Dhaka and Chittagong). Table 1 provides the summary statistics for upazilas over the years. Consistent with the secular decline in agricultural employment in developing countries discussed in the literature on structural change, agricultural employment declined from 46 percent in 2000 to 41 percent in 2010. Within the farming sector, employment in agricultural daily labor registered a sharper decline than self-employment. A large proportion of the decline in agricultural employment has been absorbed in daily (unskilled) labor in the non-agricultural sector, and a smaller proportion in self-employment. Wages for agricultural labor increased substantially over time, with the growth of nominal wage equal to 8.9 percent per annum between 2000 and 2010. The annual average growth in real wage (de‡ated by CPI) is about 2.1 percent between 2000 and 2010. The summary statistics in Table 1 also indicate substantial growth in rice yield between 2000 and 2010. Average rice yield per acre grew by an annual rate of 3.8 percent. This rate is consistent with about 3.7 percent growth in agricultural GDP during the same time.17 There has been considerable expansion of irrigation during the decade as well - from 60 percent in 2000 to 68 percent in 2010. The estimated standard deviation (Table 1) shows that there are considerable variations in rice yields across upazilas. Per capita household expenditure also exhibited considerable growth about 3.5 percent per annum. Strong growth in per capita household expenditure is re‡ected in the substantial decline 17 Crop agriculture accounts for 56 percent of agricultural GDP and rice is the single most important crop in Bangladesh not only as a subsistence but also as a cash crop. 22 in poverty during this time: the incidence of poverty declined from 48.9 percent in 2000 to 31.5 percent in 2010 (World Bank, 2013). Among the other variables, access to electricity by households improved considerably during the decade (6.3 percent annual growth rate). There is a decline in the proportion of urban households in our sample over the years which re‡ects higher growth of population in metropolitan cities compared with the other urban areas (rural towns). (5) Empirical Results In this section, we present the main empirical results along with some robustness checks. The main variables: wages, per capita expenditure, yield, and rainfall are expressed in logarithms, while the hired agricultural labor is measured as share of total employment. All regressions include upazila and year …xed e¤ects. All standard errors are corrected for correlation in the error term within upazila. (5.1) Rainfall and Agricultural Productivity We begin with the evidence on the e¤ects of rainfall variations on agricultural produc- tivity. Table 2 reports the results from regressions where log of crop yield is regressed on log of rainfall after controlling for upazila and year …xed e¤ects. Column (1) shows the results when no other explanatory variable is included in the regression. The speci…cation in column (2) includes the full set of upazila level time-varying controls as discussed in em- pirical strategy section (section 3). All of the regressions show statistically and numerically signi…cant impact of rainfall on rice yield which is consistent with a priori expectations. The estimated coe¢ cients imply an increase in yield growth when there is a positive shock in rainfall over its mean level. This result is consistent with …ndings from a rich body of evidence accumulated by the agronomists and crop scientists that shows that rainfall is a 23 major determinant of yield growth in rice in Bangladesh in the recent decades (see, for example, Sarkar et. al. (2012)). While positive rainfall above the mean increases rice yield, for appropriate interpretation of the results, it is useful to understand whether this re‡ects only the impact of transitory weather shock on farming. While the rainfall variations across upazilas and over time are expected to a¤ect the yield directly, they are also likely to a¤ect long-term productivity di¤erences by in‡uencing investment in irrigation. The third column reports estimated e¤ect of rainfall variations on the area irrigated in a speci…cation with upazila …xed e¤ects and other controls used in our main regressions. Thus the estimated coe¢ cient shows the determinant of irrigation expansion over time. A positive and statistically signi…cant coe¢ cient on the rainfall variable in this regression indicates that irrigation expansion over our sample period has happened increasingly in areas with relatively higher rainfall.18 Thus rainfall variable in our panel regressions captures not only transitory shock in agriculture but also the di¤usion of modern technology in farming over time.19 Note also that irrigation may also reduce risk by decreasing variability of yield, but may not a¤ect the average yield if it does not lead to the adoption of modern rice varieties. The expansion of irrigation in Bangladesh allowed adoption of Boro rice whose yields are signi…cantly higher than other rice types (Aman and Aus). This is con…rmed in the results in Table 2 which shows that higher rainfall does increase yield signi…cantly. 18 Historically, irrigation is adopted …rst in the drier regions in Bangladesh resulting in a negative corre- lation between area irrigated and rainfall in the cross-section data. However, expansion of irrigated areas happened increasingly in high rainfall areas –as con…rmed by our panel regression result. 19 Note also that modern farming technology such as irrigation may also reduce risk by decreasing variability of yield even without increasing yields. That is not the case in Bangladesh. The expansion of irrigation in Bangladesh allowed adoption of Boro rice whose yields are signi…cantly higher than other rice types (Aman and Aus). This is con…rmed in the results in Table 2 which shows that higher rainfall does increase yield signi…cantly. 24 Another issue in the IV interpretation of rainfall is that it may be capturing not only agricultural productivity shocks but also resulting price changes. In a completely seg- mented rice market at the upazila level, a rainfall shock would a¤ect the equilibrium rice price through income e¤ect. However rice market is the most developed and spatially inte- grated market in Bangladesh (see, for example, Golleti, Ahmed and Farid (1995), Hossain and Verbeke (2010)). In addition, we control for the distances to the main city markets, which would capture spatial price dispersion due to transport costs. The theoretical model assumes that rice price is pinned down by the international market, and available evidence on rice markets in Bangladesh clearly supports this assumption. (5.2) Rainfall, Agricultural Productivity and Labor Market Outcomes We start by presenting the regression results for wages for hired daily laborers employed in farming.20 Consistent with proposition (1) of the theoretical model, the results in column (1) of Table 3 show a statistically signi…cant and positive impact of rainfall on agricultural wages. This result suggests that the income of unskilled workers employed in agriculture, who are mostly landless poor people, gets a boost from higher agricultural productivity. The estimated coe¢ cient implies that a one percent increase in rainfall increases agricul- tural wages by 0.46 percent. Our theoretical results imply that even when agricultural productivity increase has a signi…cant positive e¤ect on the wages of unskilled farm labor, hired labor in farming may increase or decrease depending on the relative shifts in the labor demand (due to productivity growth) and supply (due to reallocation of labor from home to market work). 20 The number of observations for wage regression is slightly less (341 upazilas). This is because the wage data is trimmed by dropping 2.5% of observations in both upper and lower tails of the distribution. While such trimming is done to correct for coding mistakes. However, such trimming does not a¤ect our result: if anything the estimated coe¢ cient of rainfall is larger in the untrimmed data. 25 Column 2 of Table 3 reports the regression results for hired labor expressed as a pro- portion of total employment.21 Before presenting the estimates, we note that according to the theoretical analysis presented above, if there is no home production and thus the total supply of labor is …xed, then agricultural productivity increase has a strong positive impact on wages, but does not have any impact on hired labor (see proof of proposition 2). In a more complete model with production of non-farm goods, there can be a positive response of hired agricultural labor even if there is no home production, because labor is reallocated from the non-farm sector through the labor market in response to a productivity shock in agriculture. The evidence in column 2 of Table 3, however, contradicts both of these predictions, because we …nd a signi…cant decline in hired labor in agriculture in response to a positive rainfall shock. The evidence also indicates that rainfall has a statistically and numerically signi…cant positive e¤ect on the share of own-farming in total employment (column 3 in Table 3). This con…rms that in response to a positive shock in agricultural productivity, households reallocate more labor to own farming. As noted above, a negative e¤ect on the hired agricultural labor is not consistent with a model if there is no home production and/or labor supply response. However, it is consistent with the case where labor de…cit households respond more than labor surplus households to a wage change through adjustments in home production (in terms of parameters of the theoretical model, k > h ). However, one can argue that an alternative explanation for the observed e¤ects on hired labor and wages can also be o¤ered in terms of heterogeneity in technology of market goods instead of heterogeneity in home production. Consider the case where there is no labor 21 The regression results are una¤ected if shares are de…ned in terms of hours worked. 26 supply response (e.g. labor supply to market work is …xed) but there is heterogeneity in the technology of the locally produced market good (food) between the households. Suppose that the labor de…cit households use mechanized technology for rice cultivation so that productivity shock has no impact on its labor demand. On the other hand, labor surplus household uses more labor intensive technology and a rise in agricultural productivity increases its labor demand. Such heterogeneity can lead to higher wages and lower hired labor as surplus households reallocate labor to own farming in response to agricultural productivity increase, and thus the supply of labor to the market goes down. It is however important to appreciate that if the labor market response is driven by reallocation within the market goods sector, then we should not observe any change in the total labor supply to market work. We present a formal test of this prediction later in the paper. We emphasize here that there are good reasons to suspect that such technological heterogeneity in food production cannot drive the behavior of wages and hired labor in the context of Bangladesh. Because the level of mechanization is low in Bangladesh and there is very little heterogeneity in the technology used in production of any particular variety of rice across the country. We provide direct evidence below that a positive rainfall shock induces the households to increase total labor supply to market activities as measured by hours worked in column 4 of Table 3.22 The dependent variable here is log of per capita hours spent on market work. The results indicate a statistically signi…cant and positive impact of rainfall on hours spent on market work. This provides strong support to the conclusion that households allocate 22 As in most households surveys in developing countries, HIESs have no data on time spent on home production over the time period considered here (2000-2010). We, however, have data on hours worked on market activities (including own farming). If households indeed reallocated labor from home activities to market work, then one would expect a rise in hours devoted to market work in response to agricultural productivity shock. From the information on employment and hours spent on each activity provided by the HIESs, we compute the per capita hours spent on market work. 27 more labor to market activities (as opposed to home production and/or leisure) in response to agricultural productivity shock and that observed e¤ects of agricultural productivity increase on hired labor and wages are not simply due to reallocation of labor within market oriented activities due to heterogeneity in production technology utilized by households in food (rice) production. The results presented so far con…rms that agricultural daily wage increases with a posi- tive rainfall shock. But there is a decline in hired labor in response to positive productivity shock. From the theoretical analysis a higher wage and a positive response in labor supply to market activities imply that we would expect a signi…cant positive e¤ect on household income. While data on wages are available in the HIES, these surveys unfortunately do not provide any reliable information on income from self-employment in agriculture and non-farm activities. Thus we cannot directly estimate the impact of positive agricultural productivity shock on village income. The surveys, however, provide good information on household consumption expenditure. In the absence of information on income from self-employment, we take per capita household expenditure as an indicator of household income and welfare. The regression result with log of per capita expenditure as the depen- dent variable is reported in column 5 of Table 3. The results show statistically signi…cant and numerically large positive impact of rainfall shock on per capita expenditure. The coe¢ cient estimate implies that a one percent increase in rainfall increases per capita ex- penditure by 0.25 percent. The regression results suggest a decline in hired labor in response to rainfall shock. Since the poor are more likely to be dependent on hired labor, we check how rainfall shock a¤ects their consumption poverty. The average per capita expenditure level of households 28 belonging to the bottom 40 percent of expenditure distribution in a upazila is taken as a proxy for the welfare of the poorer section of population. The regression result with dependent variable as the log of per capita expenditure of households at the 40 percent of expenditure distribution is presented in column 6 of Table 3. The result indicates statisti- cally signi…cant positive e¤ect of rainfall shock on per capita expenditure of these poorer households as well. The magnitude of the estimated coe¢ cient (0.265) is slightly larger than that for the full sample (0.252) but they are statistically indistinguishable from each other. Though the amount of hired labor declines in response to rainfall shock, the result- ing decline in income and expenditure of the poorer households are more than o¤set by an increase in wage, and hours worked in market activities and self employment.23 As a result, agricultural productivity increase has a signi…cant and positive impact on per capita expenditure by poorer households. The evidence in Table 5 suggests that the poor bene…t equally from the positive rainfall shock as the non-poor. (5.3) Robustness Check The sample of upazilas used for estimation in Table 3 includes all upazilas outside the two metropolitan cities, and thus some upazilas which are fairly urbanized (signi…cant proportion of urban households) are also included. A reader may wonder whether our results are a¤ected if we restrict the sample to predominantly rural upazilas (less than half the population in urban municipalities). The results shown in Table 4 indicate that the overall results regarding rainfall shocks impact on wage, hired labor, market work and per capita expenditure hold true in this restricted sample as well. It is especially reassuring 23 A reduction in home production however leads to a decrease in utility from this particular source. Yet our results show that agricultural productivity increase leads to overall welfare gain due to shift of the agricultural production possibility curve outward. 29 that the results from this robustness check are remarkably close to the results reported in Table 3. (5.4) Economic Signi…cance The estimated e¤ects of rainfall variations on wage, and labor allocation between em- ployment in own farming and hired labor in agriculture as reported in Table 3 provide strong evidence in favor of a model with signi…cant labor supply response in market activi- ties as households reallocate labor from home production. The evidence presented in Table 2 also establishes clear and strong links between agricultural productivity as measured by rice yield per acre and irrigation on the one hand and rainfall variations on the other. The rainfall shocks thus can be interpreted as shocks to the productivity parameter in our model. It is thus reasonable to interpret the results on the e¤ects of rainfall variations as capturing largely the e¤ects of agricultural productivity. As we noted in the empiri- cal strategy section, our result can also be given an instrumental variables interpretation, which allows us to provide point estimates of the causal e¤ects of agricultural productivity increase on the agricultural labor market. Using the coe¢ cient estimates in Tables 2 and 3, we compute the IV estimates of the impact of agricultural productivity on outcomes of our interest. The IV estimate for the e¤ects of agricultural productivity on the wages of daily labor implies that a one percent increase in agricultural productivity (crop yield per acre) increases the wage for unskilled agricultural labor by about 0.93 percent, which is a substantial impact. A one percent increase in the productivity increases share of people engaged in own farming by about 1.4 percent, while it reduces the hired daily labor in agriculture by 1.5 percent.24 The estimate 24 While household may also draw labor out of non-farm activities and into own farming in response to agricultural productivity shock, our results show that bulk of the reallocation happens within agricultural 30 for per capita expenditure implies that a one percent increase in agricultural yield increases per capita expenditure of all households by 0.5 percent and of poorer households by 0.52 percent. The implied causal e¤ects of agricultural productivity growth on agricultural wages, labor reallocation to own farming and poverty are thus numerically substantial. (6) Conclusions This paper provides a theoretical and empirical analysis of the e¤ects of agricultural productivity on the rural labor market and poverty using an upazila (subdistrict) level panel data set from Bangladesh. The focus is on the response of agricultural wage, and labor allocation between home production and agriculture (own farming and hired labor), and their implications for poverty measured in terms of per capita household consumption. The theoretical model emphasizes the role of sensitivity of labor allocation between home production and market activities and the importance of self-employment in agriculture. The predictions from the theoretical analysis highlights the limitations of focusing on the e¤ects of agricultural productivity change on the standard labor market indicators: wage and hired labor. The positive e¤ects of a productivity increase on the wage rate may be muted if the labor supply response from home production is high enough. The e¤ect on hired labor is not unambiguous, a higher agricultural productivity may even reduce the amount of hired labor in agriculture. Even when the impacts on both wage and hired labor are positive, the main margin of adjustment for many households may be between home production and own farming. The important point here is that one can have signi…cant positive e¤ect of agricultural productivity on household consumption through a positive response of labor hours devoted to market activities as opposed to home production, even activities–between own farming and hired labor (see columns 2 and 3 in Table 3). 31 if we see little or “perverse”e¤ects on wage and employment through labor market (hired labor). Following a large literature on the importance of rainfall shocks in agricultural produc- tivity variations in Bangladesh, we exploit rainfall shocks (relative to the mean level) across upazila and over time to understand the e¤ects of productivity increase on the wage, hired labor, total hours devoted to market activities, and household consumption (per capita). We use a two step empirical approach that focuses on the reduced form regressions of rain- fall on the measure of productivity (rice yield per acre) and on the set of outcome variables noted above. The advantage of this approach is that we can test the theoretical predictions without imposing the exclusion restriction on rainfall required in an instrumental variables approach. The reduced form approach is credible given the evidence from a large liter- ature that …nds rainfall variations to have substantial e¤ects on productivity (rice yield) in Bangladesh (see, for example, Sarkar et. al. (2012), Bhowmik and Costa (2012) and the references cited there in), which is also con…rmed in our data. The evidence from the reduced form regressions show that a higher rainfall shock increases agricultural wages sig- ni…cantly, but reduces the amount of agricultural hired labor. The negative e¤ect on hired labor is not consistent with our model if there is no home production from which house- holds can reallocate labor in response to agricultural productivity increase. But existence of home production is only necessary, the negative response in hired labor also requires heterogeneity in marginal returns to labor in home production. The results on total labor supply to market activities show a statistically signi…cant and numerically substantial pos- itive impact of rainfall shocks. The evidence also indicates that a positive rainfall shock increases per capita consumption signi…cantly, thus implying that agricultural productivity 32 increase played an important role in poverty reduction achieved in the last two decades in rural Bangladesh (World Bank, 2013). We also provide point estimates of the causal e¤ect of an increase in crop yield (rice) on the outcome variables by imposing exclusion restriction on rainfall in the outcomes regressions. We use upazila and year …xed e¤ects, and include a set of carefully chosen variables to control for time varying heterogeneity to enhance the plausibility of the exclusion restriction. The IV estimates suggest that a one percent increase in crop yield increases agricultural wage by about one percent, and reduces the share of hired labor by more than one percent. References Adhvaryu, A., A. V. Chari and S. Sharma. 2013, “Firing Costs and Flexibility: Evi- dence from Firms’Employment Responses to Shocks in India,” Review of Economics and Statistics, vol. 95, no.3. Bandyopadhyay, Sushenjit & Skou…as, Emmanuel, 2012. "Rainfall variability, occupa- tional choice, and welfare in rural Bangladesh,"Policy Research Working Paper Series 6134, The World Bank. Bhowmik, A, K, and Ana Cristina Costa (2012). A Geostatistical Approach to the Seasonal Precipitation E¤ect on Boro Rice Production in Bangladesh, International Journal of Geosciences, 2012, 3, 443-462 Brückner, Markus, 2012. "Economic growth, size of the agricultural sector, and urban- ization in Africa," Journal of Urban Economics, vol. 71(1), pages 26-36. Brückner, Markus and Antonio Ciccone, 2011."Rain and the Democratic Window of Opportunity," Econometrica, vol. 79(3), pages 923-947, 05. Buera , Francisco J. and Joseph P. Kaboski, 2012. "The Rise of the Service Economy," 33 American Economic Review, vol. 102(6), pages 2540-69, October. Chernozhukov, V and C, Hansen (2008), “The Reduced Form: A Simple Approach to Inference with Weak Instruments”, Economics Letters, 2008, 100(1), 68-71. Davis, Benjamin, Stephania Di Giuseppe and Alberto Zezza, 2014. "Income Diversi- …cation Patterns in rural Sub-Saharan Africa: Reassessing the evidence," mimeo, World Bank. Emran, M Shahe, and Forhad Shilpi (2014), “Agricultural Productivity and Structural Change in Rural Non-farm Economy: Evidence from Bangladesh”, Working Paper, World Bank and IPD, Columbia University. Foster, A. and M. Rosenzweig, 2004, “Agricultural development, industrialization and rural inequality," mimeographed, Harvard University. Goletti, F., Ahmed, R., and Farid, N. (1995). Structural Determinants of Market Integration: The Case of Rice Markets in Bangladesh. The Developing Economics, 33(2), 185-202. Griliches, Z. 1963. The sources of measured productivity growth: U.S. agriculture,1940– 1960. Journal of Political Economy, 81 (4): 331–46. Hossain, Mohammad Ismail, and Wim Verbeke. 2010. “Evaluation of Rice Markets Integration in Bangladesh.”Lahore Journal of Economics 15 (2): 77–96. Laitner, J. (2000) "Structural Change and Economic Growth," Review of Economic Studies, 2000, 67, pp. 545-561. Lanjouw, Peter and Rinku Murgai, 2009."Poverty decline, agricultural wages, and non- farm employment in rural India: 1983-2004," Agricultural Economics, vol. 40(2), pages 243-263, 03. 34 Lanjouw J. and P. Lanjouw, 2001, “Rural Non-farm Employment: Issues and Evidence from Developing Countries,” Agricultural Economics, vol. 26, no. 1, pp.1 –24. Rajan, Raghuram G. and Rodney Ramcharan, 2011. "Land and Credit: A Study of the Political Economy of Banking in the United States in the Early 20th Century," Journal of Finance, vol. 66(6), pages 1895-1931, December. Sarker, M. A. R., Alam, K., and Gow, J. (2012). Exploring the relationship between climate change and rice yield in Bangladesh: An analysis of time series data. Agricultural System, pp. 11-16. Rosenzweig, Mark R., 1988. "Labor markets in low-income countries," in: Hollis Ch- enery & T.N. Srinivasan (ed.), Handbook of Development Economics, edition 1, volume 1, chapter 15, pages 713-762. Rosenzweig, Mark R. and Andrew D. Foster, 2010."Is There Surplus Labor in Rural India?," Working Papers 991, Economic Growth Center, Yale University. World Bank, 2013. Bangladesh: Assessing a Decade of Progress in Poverty Reduction, World Bank, Washington DC. 35 Table 1: Summary Statistics 2000 2005 2010 Mean SD Mean SD Mean SD Share of Employment in Self-farming 26.9% 18.2% 23.9% 17.0% 25.7% 17.5% Agricultural Daily Labor 19.4% 16.5% 15.9% 14.0% 15.5% 13.2% Self-Non-agriculture 19.9% 13.8% 22.9% 14.2% 20.9% 12.3% Per Capita Real consumption (Taka) 812.51 349.96 889.83 317.14 1142.57 360.35 Agricultural Wage (taka) 66.90 49.16 56.70 24.01 158.05 68.25 Rice Yield (ton/acre) 0.95 0.11 1.01 0.13 1.35 0.46 Population in 1991 314630 148304 297031 147249 295322 144862 Proportion Urban 0.24 0.30 0.23 0.00 0.20 0.22 Proportion of household with electricity 0.32 0.32 0.40 0.31 0.51 0.31 Percent with irrigation 61.47 29.97 59.79 31.39 67.20 23.79 Proportion of workers with secondary or above education 0.14 0.08 0.14 0.07 0.15 0.07 Rainfall (mm) 1390.9 423.4 1635.8 382.2 1457.2 361.6 Table 2: Rice Yields, Irrigation and Rainfall Log(Rice Yield) % of Area Irrigated (1) (2) (3) Log(Rainfall) 0.376*** 0.493*** 24.17** (6.382) (9.098) (2.118) Travel time to nearest large city 0.0005*** 0.0199*** (10.80) (3.672) Proportion of households with electricity 0.00718 8.296 (0.269) (1.616) Share of urban population -0.130* 4.502 (-1.854) (0.420) Log(Population) 0.00754 0.237 (0.368) (0.0796) Proportion with secondary or above education 1.291** -212.4** (1.970) (-2.492) Log(population91)*trend -0.0108 1.551 (-0.728) (0.710) Floodplain*trend 0.00360 -3.066 (0.256) (-1.296) Year and Upazila Fixed Effects Yes Yes Yes R-squared 0.509 0.623 0.084 Number of Upazilas 355 355 355 Robust t-statistics in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 3: Agricultural Productivity, Rainfall, Wages, Employment and Poverty Log(agricultural Employment in Log (per Log (per Log (per wage) Agriculture capita capita capita Hired Own- hours consumption Labor farming worked) consumption) of the poor) Log(Rainfall) 0.458*** -0.124** 0.175*** 0.209** 0.252** 0.265*** (3.108) (-2.373) (3.510) (2.442) (2.503) (2.973) Travel time to nearest large city -8.66e-05 -2.67e-05 2.23e-05 -0.000104** 0.000139** 8.76e-05 (-1.150) (-0.751) (0.747) (-2.278) (2.199) (1.535) Proportion of households with electricity 0.0903 -0.180*** -0.0881** 0.196*** 0.390*** 0.300*** (1.258) (-6.647) (-2.623) (4.952) (9.137) (7.918) Share of urban population 0.118 -0.00654 -0.0285 -0.0264 -0.0737 -0.0871 (0.659) (-0.158) (-0.596) (-0.381) (-0.828) (-1.002) Log(Population) 0.156*** 0.00429 0.0260* 0.0544** 0.0479* 0.0393 (3.102) (0.328) (1.902) (2.183) (1.771) (1.644) Proportion with secondary or above education -1.643 0.263 -0.440 -0.343 0.749 0.368 (-1.019) (0.640) (-0.811) (-0.454) (0.788) (0.462) Log(population91)*trend -0.0182 0.0259** -0.0399*** 0.00737 0.00298 0.0113 (-0.658) (2.210) (-3.570) (0.434) (0.152) (0.633) Floodplain*trend 0.0362 0.00669 -0.0104 -0.0245 -0.0652*** -0.0611*** (1.294) (0.516) (-0.860) (-1.110) (-2.992) (-3.127) Year and Upazila Fixed Effects Yes Yes Yes Yes Yes Yes R-squared 0.814 0.150 0.094 0.175 0.507 0.570 Number of Upazilas 341 355 355 355 355 355 Robust t-statistics in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 4: Robustness Checks: Estimates from Predominantly Rural Sample (less than 50 percent population in urban municipalities) Log(agricultural Employment in Log (per Log (per Log (per wage) Agriculture capita hours capita capita Hired Own- consumption Labor farming worked) consumption) of the poor) Log(Rainfall) 0.397*** -0.115** 0.209*** 0.239** 0.296*** 0.329*** (2.684) (-1.974) (3.502) (2.472) (2.923) (3.828) Year and Upazila Fixed Effects Yes Yes Yes Yes Yes Yes Full set of controls Yes Yes Yes Yes Yes Yes R-squared 0.819 0.156 0.091 0.189 0.522 0.589 Number of Upazilas 316 320 320 320 320 320 Robust t-statistics in parentheses *** p<0.01, ** p<0.05, * p<0.1