How infrastructure and financial institutions affect agricultural output and investment in India

Policy, Planning, and Research WORKING PAPERS Agriculture Latin America and Carbbean Country Department II The World Bank March 1989 WPS 163 How Infrastructure and Financial Institutions Affect Agricultural Output and Investment in India Hans P. Binswanger, Shahidur R. Khandker, and Mark R. Rosenzweig Prices matter - but so do banks, markets, and infrastructure. The Policy, Planning, arid Rescsrh Cronplex distributes PPR Woriing Papers to disseminate the findings of work in progress and to encourage the exchange of ideas among Bank staff and all other intcrested in development issues Tese papers carry the names of the authors, reflect only their views, and should be used and cited accmdingly. lbe findings. interprctations, and conclusions are the authors' own. hey should not be auributed to the World Bank, its Board of Directors, its management. or any of its member counries.  Plc,Planning, end Research Agriculture How do the decisions of farmers, financial Alfter comparing data on these factors, the institutions, and government agencies interact authors conclude that: and affect agricultural investment and output in a region - and to what extent are these "actors" * Agroclimatic factors continue to govern the influenced by a region's location and agrocli- rate at which districts can take advantage of new matic endowments (for example, rainfall or the agricultural opportunities, and govem public, soil's moisture-holding capacity). bank, and private investment decisions. This paper is an attempt to quantify the * The availability of banks (credit) is more relationships of key factors, using district-level important than the real interest rate as a factor in time-series data from India. aggregate crop output and farmers' demand for fertilizer. Rapid bank expansion in an area Agricultural opportunities in a district are increased fertilizer demand by about 23 percent, seen as the joint outcome of the agroclimatic en- rates of investment in pumps 41 percent, in milk dowments of the district and new technology animals 46 percent, and in draft animals about that becomes available to it. Better agroclimatic 38 percent. Despite their impact on investment opportunities improve output (relation 1), but and fertilizer use, the impact of banks on output also increase the economic return for a private appears to be fairly small (nearly 3 percent). farm investment - say, in a tractor (relation 2). Greater private profit in a well-endowed region * Unsurprisingly, commercial banks prefer to induces farmers to press for more investment in locate in well-watered areas where the risk of infrastructure (relation 3). Financial institutions drought or flood is relatively low. Bank expani- find it more profitable to locate where there is sion is facilitated by government investments in more demand for capital and more repayment roads and regulated markets, which improve capacity (relation 4) and where good infrastruc- farmers' liquidity and reduce banks' and farm- ture reduces their costs (relation 5). Private ers' transaction costs. agricultural investment and use of input is more profitable the better the agricultural opportuni- * In the 1970s, expansion of regulated ties (relation 2), the better the infrastructure markets contributed 4 percent to growth of (relation 6), the cheaper the cost of financial agricultural output and 17 percent to demand for services (relation 7), and the more favorable fertilizers. Expansion of electrification in- government's price and interest policies (relation creased output 2 percent in a decade by increas- 8). Exactly the same factors affect the output ing investment levels for pumps and fertilizer. supply (relations 9, 10, 11). Traditional ap- A primary education added a large 8 percent to proaches to production function estimate the crop output over the decade, primarily by direct impact of capital stocks (investment) and increasing fertilizer demand nearly 30 percent. input use on output (relation 12), ignoring many of the factors discussed here. This paper is a product of the Agriculture Operations Division, Latin America and Caribbean Country Department IT. Copies are available free from the World Bank, 1818 H Street NW, Washington DC 20433. Please contact Josefina Arevalo, room 17-100, extension 30745 The PPR Working Paper Series disseminates the findings of work under way in the Bank's Policy, Planning, and Research Complex. An objective of the serics is to get these findings out quickly, even if presentations are less than fully polished. The findings, interpretations, and conclusions in these papers do not necessarily represent official policy of the Bank. Produced at the PPR Dissemination Center  THE IMPACT OF INFRASTRUCTURE AND FINANCIAL INSTITUTIONS ON AGRICULTURAL OUTPUT AND INVESTMENT IN INDIA by Hans P. Binswanger, Shahidur R. Khandker, and Mark R. Rosenzweig Table of Contents I. Introduction ....................................... 1 It. Analytical Framework. 3 III. Data and Variables ................................. 11 IV. Agroclimatic Endowments, Infrastructure, Population and Crop Output ......................... 17 V. Development of Commercial Banks .................... 22 VI. Determinants of Private Investment ................. 24 VII. Determinants of Fertilizer Demand and Aggregate Crop Output ........................................ 31 VIII. Discussion ......................................... 35 References ............................................... 42  I. Introduction Government expenditure on physical infrastructure and human resource development influence the private production and investment decisions in agriculture and thus are essential ingredients of increased agricultural productivity. Government investments (both physical and human) can directly increase agricultural output by shifting the production frontier as in the case of irrigation. This is what might be called the direct effect of government infrastructure. Government investment also increases the rate of return to private agricultural investment and thereby leads to greater investient and output. Moreover, by increasing the viability and profitability of financial intermediaries,.infrastructure can facilitate the emergence and growth of financial institutions that increase access to working and investment capital or reduce the costs of borrowing for long-term investment. Better credit facilities, by enab}ing the smoothing of consumption, may also increase the willingness of farmers to take risk. 11 This paper would not have been feasible without the patient and persistent efforts of a large number of people who assembled, checked and processed the enormous data base required. The ICRISAT economics staff graciously provided their district data base for the semiarid tropics. Robert Evenson and Ann Judd in turn contributed their North India district data base. James Barbieri assembled the data required to update these data bases to 1982, added additional states, and collected the data on banking, on private capital stocks and investment. He was advised in this endeavor by Devendra B. Gupta who provided the liaison with the respective Indian authorities who were gracious to release data which were still in manuscript form. Dr. D.R. Gadgil opened the data base of NABARD and personally organized the Continued on next page  -2- Thus agricultural output and investment respond to the separate actions undertaken by three economic actors-- faxsers, government agencies and banks. All three actors respond positively to agricultural opportunities implied in the agroclimatic endowments of a region and in new technologies when it becomes available. The magnitude of the effects depends on how each of the actors responds to agricultuzal opportunities and how farmers and banks respond to the government investment decisions. This paper seeks to quantify the inter-relationships among the investment decisions of governmeat, financiaI institutions and farmers and their effects on agricultural investment and output. The central problem of estimating these relationships is that, once government agencies and banks are admitted as actors who respond to agricultural opportunities,. one can no longer take the distribution of government infrastructure and banking iastitutions as exogenously given or randomly distributed. The impact of government infrastructure on investment and output is likely to be more pronounced in a better endowed region than in a poor endowed region and governments will therefore invest more where opportunities are greater. The resulting unobserved variable problem may be circumvented by either a precise quantitative characterisation of the agroclimatic potential or by using the fixed effects technique of estimation. For analyzing growth in aggregate output Continued from previous page assembly of the banking data. Apparao Katikineni organized the screening, and further computer processing of the many different and disparate data bases which had to be merged into a single data base. Kathy Graham did all of the cartographic measurements. We were also assisted by Sneylata Gupta and Dan Ghura.  -3- both Binswanger, Yang, Bowers and Mundlak (BYBM, 1986) and Lau and Yotopoulos (1988) have implemented the appropriate fixed effects technique using time-series x cross section data of countries.2 The paper is structured in the following order. Section II outlines an analytical framework. Section III describes the data and discusses the variables. Section IV shows how various agroclimate variables affect government decision on where to build roads, markets and schools and where to provide electricity and canal irrigation. Section V looks at the joint impact of agroclimatic endowments and infrastructure on the growth of financial intermediaries. We do this in a cross-section and. time series framework using district level data from India. Section VI then considers the joint determination of private agricultural investment as functions of agroclimatic endowments, technology change, infrastructure, financial intermediaries and of price and interest policy. In Section VII we estimate the impact of the same variables on fertilizer demand and aggregate crop output. The results' are summarized in the concluding part of the paper. II. Analytical Framework In our framework (see Figure 1), agricultural opportunities in a district are the joint outcome of the agroclimatic endowments of the district and the new technology which becomes available to the district 2/ Bapna, Binswanger and Quizon (BBQ, 1984) used random effects techniques to analyze output supply in time-series x cross section data of districts in India.  AGRICULTURAL OPPORTUNITIES 0 PRIVATE - AGROCLIMATIC agroclimate effects () INVESTMENTS ENDOWMENTS AND INPUT USE - NEW TECHNOLOGY financial \ ~~~~~~~syste ;f f t /- prod. policy effects grntr) function pressure groups ~~~~~~~~~~~~~~~~~~PRICE AND | ~~~~~~~~~FINANCIAL direct effects /INTEREST < t>~ ~ ~~) INSTITUTIONS /.POLICY effeo lcts effects infrastructure effects (IQ| Figure 1: Major Relationships among Agroclimte Radowents, ?1nanclal Institutions, Government Infrastructure and Agrliltural Investment and Output  -5- from industry and from foreign, national and state research systems.3 T:^e same technology is potentially available to all districts., but the exten: to which it is actually applicable to a given district depends on its agroclimatic endowments. For exampli high yielding varieties of wheat are not relevant for districts with high wdnter temperatures or districts with excessive amounts of rainfall and flooding problems. Thus the size and the growth of the set of agricultural opportunities varies across a district according tu their agroclimatic characteristics. Better agroclimatic opportunities such as better rainfall, a higher moisture holding capacity of the soil and a better irrigation potential directly affect agricultural output (relation 1). But better opportunities also increase the economic return to private farm investments such as tractors; draft animals or pumpsets (relation 2). The greater private profitability of agriculture in well endowed regions induces farmers to press government for increased investment in the supportive infrastructure (relation 3). Financial institutions find it more profitable to locate in environments where a good agroclimate and rapid technical change leads to a substantial demand for agricultural investment and-working capital and a high repayment capacity (relation 4) and where good infrastructure reduces their cost of intermediation (relation 5). Private agricultural investment and input use is more profitable the better the agricultural opportunities (relation 2), the better the government 3/ Although technology investments are Lhemselves government decision variables, for the purpose of this paper technology is treated as exogenous to the decisionmaking of specific-government agencies, banks and farmers when they make investment, location and input decisions.  -6- infrastructure (relation 6), the cheaper the cost of financial services (rel&tion 7) and the more favorable price and interest policies are which are pursued by the government (relation 8). Exactly the same factors affect the output supply. (relation 9, 10, 11). This means that agricultural opportunities must be translated into public and private investment 'fforts in order to affect agricultural output (For a discussion see BYBM, 1986 and Mundlak, 1985). The traditional production function approach has attempted to estimate the direct impact of capital stocks (investment) and input use on output, (relation 12), ignoring much of the factors discussed here and all the simultaneity problems (see for example Hayami and Ruttan, 1986).4 Estimation Equations and Econometric Specifications Let Rrjt be the level of the r-th infrastructure variable (say. roads) in district j at time t. As agroclimate variables ar-e strictly exogenous, the dependence of the infrastructure variables on a set of measured agroclimate variables and location factors ai can be estimated simply in a cross section regression - (1) Rrjt Rrjt (0j, Ppj ejt) 4/ A profit-oriented producer will take input and output decisions jointly. In order to deal with this simultaneity, the correct way to estimate the production function (12) is to use the predicted instead of actual levels of capital stocks and other inputs. However, as clearly apparent from figure 1, we do not have any instruments which affect only the inputs but not the output. -Therefore the production function cannot be estimated econometrically.  -7- where p3 is the effect of unmeasured agroclimatic and location factors, and ejt is a time-specific error term. 5 Financial institutions in turn are assumed to locate in districts with good agroclimate and infrastructure, i.e. (2) Bjt - Bjt (Rjt, aj, jj, Tjt, ejt) where Bjt stands for the number of banks operating in the district at time t, Tjt is a region-and time-specific technology index, and ejt is an error term specific to the banking equation. The simultaneity between banks, Bjt, and infrastructure, Rjt, arising from their joint dependence on unobserved agroclimatic variables, , can be overcome if an additive model is chosen such as (3) Bjt aO +aj Rjt + a2 Uj +a3 IJ + Tjt + ejt For the mean overtime in district j this relationship reads (4) Bj. 3 ao + al Rj. + a2aj +a3 /pj + Tj. + ej. Taking the difference of these equations, i.e. by transforming the variables to differences from their means, leads to the following estimation equation in terms of difference from the means (5) (Bjt - Bj.) - al (Rjt - Rj.) + (Tjt - Tj.) + (Ejt - ej.) 5/ Similar ultimate reduced forms can be established for all other endogenous variables in the system, the banks, private investment and output. However, the reduced forms for output, for example, is not very informative as it includes both the direct or technical impact of the agroclimate on output (relation (1)) as well as all the indirect effects via its impact on infrastructure, banks and private investment.  As et is a randomly distributed error torm which is uncorrelated with oj and Rjt, relationship (5) can be estimated by the Ordinary Least Squares technique. The disadvantages of using this fixed effects model is that the direct impact of the agroclimate (i.e. relation (4)) cannot be estimated. 6 These direct effects could only be estimated (in this and all subsequent equations) if the infrastructure variables were randomly distributed acrc-s the districts, i.e. were not dependent on the unobserved agroclimatic variables pi. This could happen if the measurement of the observed agroclimatic variables oj were so good that no unobserved effects were left over to significantly affect the infrastructure investments. In that case a random effects model would be appropriate. We will use Hausman-Wu specification tests to determine whether to use the fixed effects or random effects model and present results accordingly. Private agricultural investment in capital item k (say draft animals or tractors), depends on the agroclimate, the infrastructure and the banks. In addition it depends on policy variables such as the output price P, the fertilizer price Pf, and the real interest rate r, i.e. (6) lkjt - Ikjt (Kjt-I, Pf3jt' jt' B3jt Rjt, Tjt, 'jp Pjt ejt) where Kjt-l is the capital stock of year (t-1). In these equations the 6/ As discussed in footnote 1, it cannot be estimated either in a reduced form equation of the form of equation (1).  -9- simultaneity problem between Bjt, Rjt, and p can be overcome just as for equation (3) by using the fixed effects technique. For the interest rate, however, a simultaneous equation problem may arise if higher investments demand leads credit s npliers to raise the interest rate. In the application .below this problem is minimized by the fact that the interest rate used is fixed by the government of India. Only if the government specifically takes agricultural investment demand (rather than more aggregate credit conditions) into account in fixing the interest rate will the simultaneity problem persist. For the other prices a simultaneity can also arise if increased investment leads to higher output supply (and fertilizer demand) and therefore lower output prices (higher fertilizer prices). The fertilizer price used below As the railhead price set by the government and the endogeneity problem is likely to be minimal. However, in order to overcome the simultaneity problem associated with output prices we use an index of international commodity prices as an instrumental variable for the domestic price index. *As India is a small country in virtually all international agricultural commodity markets, this completely eliminates the simultaneity bias.7 The index of international commodity prices is computed separately for each district using district-specific production weights for the year 1975. Since agricultural wages are clearly endogenously determined with agricultural output and investment we replace this variable with state-specific urban wages as an instrument. 7! India has state trade in agriculture. So the domestic prices do not correspond in any simple way to international prices. Nevertheless, estimates show that domestic prices respond positively to international prices, with a lag of 3 to 4 years.  - 10 - As can be seen from figure 1, the variables entering the aggregate crop supply equations are exactly the same as those entering the investmen: equation and all estimation problems are the same. Obviously no data exist on the district-specific index of available (but not necessarily implemented) technology Tjt. Evenson and Kislev (1975), BYBM (1986) and others used expenditure data, or manpower data for public research and for extension. (Alternatively Boyce and Evenson (1975) has used research publications). However, data for these variables do not Wxist at the district level. Moreover they do not include technology opportunities arising from private industry and seed corporations. Other researchers have used simple time trends to proxy technology opportunities (e.g. Binswanger, 1974). However, simple time trends assume technology opportunities grow at constant and identical rates in each district. But the point is that agroclimatic endowments affect the extent to which new technology option; are applicable to a district. Hence technology trends must differ across districts. If, for example, banks systematically located in districts where the green revolution had raised the input and borrowing requirements of farmers, an output supply function including common time trends and banks would erroneously allocate to banks a part of the output contribution of technology, i.e. the coefficient of banks would be upwards biased. In order to circumvent this problem the district specific technology trend is entered into the models as follows: (7) Tjt - Ebm amj t + btt m The regression include a common time trend and interaction terms of all the agroclimatic variables with the time trend, and the district  - .11 - specific time trend is an estimated function of time and time x climate interactions. 8 For example in the output supply equation one would expect the interaction between time and irrigation potential to be positive, consistent because high yielding varieties of rice and wheat require good control over water and soil moisture. III. Data and Variables The data we have used here are drawn from India. The cross- section units are 85 randomly drawn districts of India. These 85 districts belong to 13 states of India--Andhra Pradesh, Bihar, Gujarat, Haryana, Jammu and Kashmir, Karnataka, Kerala, Madhya Pradesh, Maharashtra, Punjab, Rajasthan, Tamil Nadu and Uttar Pradesh. The 85 districts are part of the 99 districts from 17 states randomly drawn by the India's National Council for Applied Economics for its well known Additional Rural Income Survey.9 The period covered in this paper is the agricultural years 1960/61 to 1981/82. The study period covers the agricultural years 1960/61 to 1981/82, but for some dependent variables with more limited data availability shorter periods are used. 8/ Alternatively one could introduce a trend for each district separately as in Lau and Yotopoulos. 9/ In fact, 80 districts are drawn from NCAER sample, while the remaining 5 districts are the ICRISAT districts. All NCAER districts could not be included because of deficiencies in the data for both primary and secondary districts.  - 12 - TABLE 1: Descriptive Statistics of Variables Variable Number of Mean Standard Observations Deviation Dependent variables Aggregate crop output index 1785 1.192 1.045 Fertilizer consumption, nutrient tons/10 sq. km 1148 22.054 924.045 Net investment in draft animals; numbers per year/10 Sq.Km 304 4.962 15.504 Net investment in tractors; numbers per yearllO Sq.Km 304 0.119 0.260 Net investment in pumps; numbers per yearilO Sq.Km 304 1.343 1.864 Net investment in milk animals; numbers per year/10 Sq.Km 304 13.555 25.489 Net investment in small stocks; numbers per year/10 Sq.Km .304 5.308 15.291 Rural population, numbers/10 Sq.Km 1785 2070.304 1547.327 Time-varyina independent variables Canal irrigation, '000 ha/10 Sq.Km 1785 0.064 0.099 Number of villages with primary schools/10 sq. km 1785 1.139 0.605 Electrified villages, numbers/10 Sq.Km 1785 0.688 0.764 Commercial banks, numbers/10 Sq. Km 1785 0.069 0.108 Regulated niarkets, numbers/10 Sq.Km 1785 0.014 0.022 Road length, '000 km/10 Sq.Km 1785 4.389 4.277 Real interest rate of cooperative societies 1785 4.010 4.485 Aggregate real domestic crop - price index 1785 0.968 0.295 Aggregate Real International Price index 1785 0.687 0.355 Real urban wage (annual earnings) 1785 A186.336 1195.586 Real fertilizer price index 1785 3.413 0.505 Annual rainfall (in mm) 1785 1138.573 986.503 Time-invariant independent variables Length of rainy season in months 85 3.653 1.368 Number of excess rainy months 85 1.236 1.393 Number of cool months (Temp < 180) 85 0.935 1.313 Percentage of district area liable to flooding 85 1.389 3.531 Irrigation potential, percentage 85 30.001 31.897 Urban distance (km) 85 298.441 152.029 Soil moisture capacity index 85 2.350 1.009  - 13 - For each district the aggregate crop output is the index of 20 major crops with district-specific prices of 1975/76 as the base. - Agricultural investment is represented by private investment in draft animals, milk animals, small stock (i.e. sheep and goats), tractor, and irrigation pump. Investment for each period is the difference in the stocks meaiured for each of the agricultural censuses which occurred at five year or six year intervals. It is therefore net investment per year during each of four intercensal intervals. Government infrastructure consists of primary schools, canal irrigation, rural electrification, regulated rural markets, and total road le.ngth. The only rural financ.al intermediary is the number of rural branches of Commercial Banks--the only -comparable data available for the study period 1960-81. The variables that characterize the district's agroclimate environment are the irrigation potential, the length of the rainy season in months, the number of months in a year with excessive rain (where rainfail exceeds potential evapotranspiration),l0 the number of cool months in a year when mean temperature is less than 18 degree Farenheit (this is related to the ability to grow wheat), the proportion of a district liable to flooding, an index of the moisture capacity of the soils in the district, and the district's distance to the nearest major urban center (out of eight centers i.e., Delhi, Bombey, Madras, Banglore, Kanpur, Ahmedabad, Hyderabad and Calcutta). The price variables consist of the annual per capita earnings of industrial workers as a proxy for urban wages, a price index of fertilizers at railheads, the rate of interest charged by the Primary Agricultural Cooperative Societies, and the district specific real 10/ Evapotranspiration is a sum of transpiration via plants and evaporation from soil.  - 14 _ international price index for the aggregate output. The international prices of 17 crops have been converted to Rupees at the official exchange rate and aggregated using district-specific production weights for the agricultural year 1975/76. Domestic price indices using district level farm harvest prices have also been computed for comparison. All the prices or price indices are deflated by the consuner price index for rural workers using 1975/76 as the base year. The interest rate charged by cooperative credit, societies is the expected real rate i.e., the nominal rate for the current year less the average percentage increase in the consumer price index for the previous five year, i.e. inflationary expectations are assumed to form over a five-year period.11 The means and standard deviations of these variables are listed in table 1. The data relating to the agroclimate and the urban distance are single data points for each of the 85 district, because they are the permanent characteristics of each district. The data relating to agricultural output, government investment, prices and Commercial Banks are time-series data covering the period, 1960/61-1981/82. In contrast, the investment data refer to the four intercensal periods 1961-66, 1966-72, 1972-77, 1977-82. When investment equations are estimated the independent variables are their respective means over the corresponding census intervals. 11/ In both the output supply and investment equations, the inflation rates are averaged over five years, except for the years prior to 1965 where data limitation led us to use averages for 2, 3 and 4 years respectively. The rate of inflation did not fluctuate sharply in the late fifties, or early sixties.  - 15 - The following variables require additional explanation: The azzregate output index reflect both variation over time in each district relative to its base year 1975 as well as variations in output across districts relative to a hypothetical average district formed by computing the averages of all variables across districts in 1975. The agregate output is normalized for district size, i.e. it c-mpares aggregate "yields" per unit of geographic area. When fixed effect techniques are used the across- di-trict variability is of course lost. 'Rezulated markets do not include all rural markets but only those where government provides market infrastructure and regulates all trades via a supervised auction system. The government does not regulate the market price but may enter as a purchaser in order to prevent market prices from falling below its guaranteed level. The rural banking system is complex. It consist of traditional moneylenders and traders for whom no data exist. Cooperative credit societies were the first formal institutions to achieve wide rural coverage. They lend largely for short-term production purposes such as fertilizers. By 1969 such societies existed in virtually all districts of India covering 94 percent of the villages in the country and they were providing 5i percent of total formal credit extended to famers. Their number has been declining as smaller societies have been merged in recent years. Yet by 1979 they were providing 49 percent of all formal credit. We include the regulated government lending rate of these societies among the explanatory variables. At this regulated interest rate credit rationing is pervasive and we test whether the rate has an influence on  - 16 - output and investment despite the rationing. Land Development Banks are also cooperative institutions which lend primarily for investment purposes. Between 1969 and 1979 their share of lending in formal credit increased from 15 percent to 19 percent. Their lending rates are closely tied to the official rates of the Cooperative Credic Societies, i.e. the interest variable will capture the effect of interest rate changes of both the cooperative credit societies and the Land Development Banks. The Commercial Banks, in 1960, were not involved in lending to farmers except perhaps to plantations and very wealthy farmers. However, they did considerable lending to agroindustrial enterprises. After their nate 'alization in 1969 they were compelled by the government to expand their lending to farmers and agroindustry with targets set both for number of rural branches as well as the proportion of lending to the agricultural sector. Between 1969 and 1979 the share of commercial banks in formal credit provided to farmers rose from 34 percent to 49 percent at the all India level. In the 85 districts considered here the role of commercial banks rose even farther. Their share in formal credit rose from 52 percent in 1972 to 72 percent in 1.979. At the same time the number of rural and semi-urban commercial bank branches rose from 3,625 to 7,690. Unlike the volume of lending and outstanding loans in any period, which is influenced by farmer demand, the number of branches is strictly controlled by the Banks and'therefore strictly exogenous to farmer decision making. Other than for the joint dependence of farmer and bank decision making on agroclimatic and infrastructure variables, no endogeneity problem therefore arises. Unfortunately the number of Cooperative Societies and Land Development Banks cannot be used in these equations as exogenous measure of the growth of these systems because the process of consolidation has  - 17 - reduced their numbers but not necessarily the availability of their services in the affected villages. Soil moisture capacity is a variable which can be viewed as a substitute for either rainfall or irrigation. For a given rainfall a higher soil moisture capacity means that a crop can withstand more days without additional rainfall. In addition where soil moisture capacity is yery high, a full moisture reservoir in the soil may be able to support several months of a crop cycle without the addition of rainfall or .irrigation. FQr given annual rainfall, payoffs to irrigation investments ara therefore more limited where soil moisture capacity is higher. Irritation potential is defined as the percentage of a district's area inside any type of irrigation command area, i.e.. sum of proposed. command area, command area under construction and already existing command area.12 This-variable has been measured using the Irrigation Atlas of India. Planned command areas are a good indicator o. the remaining potential for canal irrigation in India as they reflect long range plans and any area not yet included in these plans has virtually no potential. IV. Azroclimatic Endowments. Infrastructure. Population and Crop Output In Table 2 we see that the seven measured agroclimatic and location factors explain between 24 percent of the variation in the density of primary schools to 41 percent of the variation in government provided 12/ An irrigation command area is an area whilc receives or is expected to receive water from an irrigation system.  - 18 - Table 2: Effects of Aprocilmat. Endowment. on Infrastructure. Population and Asaresate Crop Output (Observation a 56) Explanatory Rural Regulated Canal Primary Rural Aggregate Variables Road Morket Irrication School Electricity Population Crop Output Cool months -0.222 0.004 0.912 0.303 0.389 285.725 6.229 (-0.587) (1.647)e (1.559) (4.770). (4.293)o (2.261)o (2.417)* Excoss rain 0.229 -4.O02 -0.M1 -0.041 0.009 42S.138 0.019 (0.477) (-0.761) (-4.667) (0.077) (2.679)* (2.679). (O.168) Rainy seson 2.442 -4.064 9.oo7 0.079 0.089 S41.898 0.264 (4.901)e.(-0.140) (0.708) (0.988) (6.775) . (3.86). (2.261)* Flood potential -0.077 -O.01 -O .W14 0.021 -0.064 -10.649 -0.694 (-4.S09) (-1.622). (-1.234) (9.862)- (-1.826)o (-0.226) (-2.574). Irrigation 0.033 0.001 0.002 0.003 6.W69 20.589 0.034 potential (1.93S). (5.872)o (6.706)o (1.127) (2.332). (3.712). (8.262). Soil moisture 0.422 0.004 -0.619 0.116 0.231 -14e.635 0.102 capacity (1.676). (2.827). (-2.187). (2.159)* (-0.982) (-8.034) (0.914) Urban distance -4.003 0.0W04 0.6001 4.0002 -0.001 -0.671 0.034 (-9.118) (2.881)- (2.266). (-0.320) (-1.419) (-0.825) (0.026) Constant -4.218 -..014 -0.018 0.586 0.186 -416.025 -0.804 (-1.692). (-9.97) (-9.384) (1.474) (0;826) (-4.524) (-1.852) Adjusted R-cquarr 0.87 6.8 0.48 0.24 9.28 0.41 0.53 Not.s: t-Statisticu are in parentheses. Asterisk refers to significance level ot 10 percent or better. Rural road corrosponds to the agricultural year 1974, while the remaining variables relate to agricultural year 1981.  - 19 - irrigation and the population density of the region. The explanatory power is thus very substantial, not much smaller than that for output itself (53 percent). The traditional treatment of agricultural infrastructure as exogenous variables in output supply analysis is unwarranted. The variable with the most powerful effect across the equations is irrigatiorn potential, i.e. the proportion of the area which is included in an existing or planned irrigation command area. It significantly increases the density of all infrastructure variables, except schools. It is also clear that population has migrated to, or grown more rapidly in regions with a high irrigation potential, i.e. private and public decisions are influenced by the same factors. For the other variables the impact varies substantially across the government investments. Regions with a fairly cool winter, which are able tRo grow wheat, have higher density of regulated markets, more primary schools, more electrification and higher population density. Population density is also very high in regions with many monthn with excess rain, i.e. the humid tropical zones such as Kerala. Population and roads are higher the longer the rainy season, or said otherwise, government has found it less worthwhile .to build roads in semiarid and arid regions. Areas which are liable to flooding are not well served by regulated markets, roads, and electrification. As discussed in the data section, the results show that high soil moisture capacity acts as a substitute for canal irrigation. But high soil moisture capacity is a positive agroclimatic attribute and thus attracts investment in regulated markets, roads and in electrification. Distance to major urban centers tends to increase the  - 20 - Distance to major urban centers sends to increase the density of regulated rural markets, and also the level of government provided irrigation. One point which stands out from the regressions is that for the purpose of government infrastructure investments, agroclimatic potential cannot be measured by a single variable or an agroclimatic index. Different aspects of the endowment affect the govrernment investments differentially. The total effects--both direct and indirect effects via government infrastructure and banks--of agroclimate and location factors on output supply are also shown in table 2. They suggest that agroclimatic endowments explain 53 percent of the variation in agricultural production of 85 districts in the agricultural year, 1981. Agricultural output supply is high in regions well endowed with water from either irrigation or rainfall. Regions with a relatively cool winter have high agricuitural output while agricultural output is low in regions with a high flood potential. All the effects are as-expected. Using data for the three-census years (1961, 1910, 1981) we investigate in table 3 whether the investment trends over the past two decades were similiarly influenced by the agroclimatic characteristics. This is done in regressions which include time trends according to equation- 7, i.e. interactions between time and the agroclimatic characteristics. The Hausman-Wu test suggests that a fixed effects model is appropriate for explaining variation in public infrastructure ovdr time, while a random effects model can be used for rural population growth.  - 21 - TABLE 3. Effects if Agroclimate Endowments on Growth in Infrastructure and Population. (Obe. a 265) Explanatory Regulated Canal Primary Rural Rural Variable Market Irrication School Electrification Population Fixod Effeets Fixed Effects Fixed Effects Fixed Effeets Random Effects Yoor 0.0006 0.0001 0.008 -0.004 0.283 (1.614) (0.162) (1.623) (-0.222) (t.034) Yoor A cool months 0.0001 -0.000 6.010 0.018 4.643 (1.624). (-0.237). (7.632)* (5.329). (3.229). Year x excess rain -0.000 0.0005 -0.001 -0.001 6.186 (-0.283) (3.269)* (-0.526) (-0.305) (3.640)* Yoar x flood -0.000 0.0001 0.0002 -0.002 -0.163 potential (-2.063)o (1.149) (0.411) (-1.042). (-4.361) Year a lrrigation O.WO 0.60 00 0.66 6.0104 0.235 potential (5.683)0 (2.152)* (-0.456) (2.489). (3.757). Year x soil 0.6of -e.00o 0.005 0.011 -1.847 moisture capacity (0.184) (-0.126) (3.132)o (2.801). (-1.0S5) No. of cool months -96.S23 (0.648) Excess rain -79.314 (40.431) Rainy season 121.471 (6.644) Flood potential 0.2S3 (0.004) Irrigatton potential 1.835 (0.283) Soil moisture -1.406 capacity (0.008) Urban dish, g --S.494 (0.534) Constant -494.96C (-0.852) F-Statiptic 32.152 C.148 63.427 38.991 56.113 Hausen-Wu (Chi-square,7) 16.222 19.274 20.482 22.326 1.364 Note: Asterisk refers to sIgnifIcance level of 10 percent or better on * two-tall teat.  - 22 - The results are consistent with the simple cross section results. They suggest that better agroclimatic attributes such as irrigation potent!al contribute to the growth in public infrastructure as well as population, while unfavorable attributes such as flood potential reduces their growth over time. This clearly indicates that agroclimatic endowments did not only affect the placement of public programs and institutions in the distant past, but also their growth over the study period. V. Development of Commercial Banks In Table 4 the cross-section results indicate that Commercial Banks have tended to locate in areas which are well endowed with water, either from irrigation or from a long and over-abundant rainy season. Such areas are characterized by relatively low risk of agriculture and therefore less repayment'problems for the banks (Binswanger and Rosenzweig, 1986). Th. implies that the banks have avoided areas where drought risk is high. That banks try to avoid'high-risk areas, is also apparent in the negative coefficient of flooding potential. The simple cross section relation for the year 1980 included the indirect effects of the agroclimate on Bank, via the improved infrastructure. The pure infrastructure effects, on the other hand, are estimated and peesented in Table 4 using the cross-section time-series data of 85 districts for the years 1972-80. As the results of the Hausman-Wu test suggest, the fixed effects model appears more appropriate than the random effects model in explaining the variation in the bank growth over time and only the fixed effects results are therefore shown. The  -23 - TABLE 4: Effects of Azcoclimatic Endowments and Government Infrastructure on Commercial Bank Cross-section effects Fixed effects Explanatory Variable (observations * 85) (observations - 765) Year 1980 Canal irrigationa -0.193 (-2.190)* Regulated Marketa 0.196 (3.227)* Primary Schoola 0.026 (0.077) Rural Electrificationa -0.115 (-1.457) Roada 0.821 (4.584)* Year -0.011 (-4.873)* Year x Cool Months 0.002 (3. 835)* Year x Length of 0.002 Rainy season (4.514)* Year x Flood Potential -0.001 (-3.983)* Year x Irrigation 0.0001 Potential (6.639)* Year x Soil Moisture -0.0001 Capacity (-0.093) Year x Excess Rain Months 0.005 (11.372)*. No. of Cool Months 0.016 (1.156) Length of Rainy Season 0.046 (2.735)* Flood Potential -0.010 (-1;992)* Irrigation Potential 0.002 (3.848)* Soil Moisture Capacity 0.0002 (0.068) Excess Rain Months 0.050 (2.985)* Urban Distancea -0.205 (-0.089) Constant -0.132 (-1.570) F-Statistic 6.90 94.792 Hausman-Wu (Chi-square, 12) 51.631 NOTE: t-Statistics are in parenthesis. Asterisk refers to significance level of 10 percent or better. a Coefficients of these variables are in elasticity form.  - 24 - regression clearly shows that Banks are more likely to locate in areas where the road infrastructure and the marketing system are improving. Markets provide both higher incomes to producers and reduce the price risk they face, i.e. they improve their repayment capacity. And roads have an eifect on farmer income, demand for inputs and hence credit and they reduce the credit transactions costs of both the customers and the banks. Of the two variables, roads have the most powerful effect with an elasticity of about 0.83, followed by regulated markets with an elasticity of 0.20. Markets, of course, are a relatively cheap investment which can be increased much more rapidly than roads. Rural electrification does not contribute to Bank gro-th. Indeed it has a negative effect which is statistically significant at the 10 percent level. The time trend and the interaction effects with time confirm that Bank growth has been more rapid in districts with a high irrigation potential, where the rainy season is longer, and where cool months allow for the growth of wheat. This is fully consistent with the hypothesis that banks have systematically located in environments which were favorable to the green revolution technologies; i.e. that banks responded to opportunities.-created by technical change. In addition bank growth was lower where the flood potentially high, i.e. where they face higher risk and where green revolution technology is less applicable because of lack of water control. VI. Determinants of Private Investment The investment data in table 5 relate to average annual levels of investment for each of the intercensus intervals for which we have data.  - 25 - Table S. Effects of Aaroclimatic Endowments. Government Infrastructuro. Commercial Bank *nd Prices on Agricultural Investment (No. of Observations u 304) Investment In Explanatory Draft Milk variable animals animals Small stocks Pumos Tractors Aggregote real interno- 2.098 1.007 1.697 -0.497 -0.076 tional price, logged a (3.709)* (2.368)* (2.163)* (-1.327) (-0.197) Real fertilizer price5 -12.262 -8.396 -7.836 -1.140 -1.127 (-4.292)* (-4.139)* (-2.662)* (-0.834) (-0.799) Real urban wag, a 6.866 3.406 2.106 -0.470 1.127 (6.306). (5.684S) (1.977). (-0.922) (2.284). Real interost rate -0.686 -6.115 -0.302 -0.109 0.092 (.3.688). (-1.163) (-1.691)