.- IF iF .040 D46 go. 95 -3 -. 2 II!1 1 I I~ I I _I 875 SLC008759 /N. / Y_ ARU- 48 For Staff Use Only Estimation of Aggregate Agricultural Supply Response from Time Series of Cross-Country Data Hans Binswanger, Yair Mundlak, Maw-Cheng Yang and Alan Bowers SECTORAL LIBLRAR Ir4TERNATI0,NAL B'ANK REcONSItRUCTIO AD DEVELOPMET tAAYM Oi. t89 Division Working Paper No. 1985-3 September 1985 Commodity Studies and Projections Division Economic Analysis and Projections Department Economics and Research Staff The World Bank Division Working Papers report on work in progress and are circulated for Bank staff use to stimulate discussion and comment. 5 O Jt gIk - _VPERS Copy ESTIMATION OF AGGREGATE AGRICULTURAL SUPPLY RESPONSE FROM TIME SERIES OF CROSS-COUNTRY DATA September 1985 Prepared by: Hans Binswanger Agriculture and Rural Development Department Yair Mundlak, Consultant Hebrew University of Jerusalem Maw-Cheng Yang and Alan Bowers Commodity Studies and Projections Division Economic Analysis and Projections Department TABLE OF CONTENTS Page List of Tables and Figures ............ $................................ iii Preface ..................... 0..... ............... ......... .#* ....... iv rI. INTRODUCTION .......................................... ............1 II. DATA, VARIABLES AND ESTIMATION TECHNIQUES .......................... 5 Price Deflators and Price Elasticities ....................... 10 III. THE STATISTICAL FRAMEWORK .....................s................... 12 IV. THE EMPIRICAL IMPORTANCE OF COUNTRY EFFECT ........................ 16 V. WITHIN-COUNTRY ESTIMATES ......... ....................................... . 22 Cross-Price Effects..*.. * ...... ........... ................ 27 The Shifters ............................................... 28 VI. INTERPRETATION .................. ................................. 39 Length of Run ....... o.*.*.* ................................. 39 Varying Factor Supply. . .................. . ................ 0. 43 VII. ON THE SUPPLY IN THE LONG RUN ........................ ....................... . 47 APPENDIX 1 DETAILED DATA DESCRIPTION ........... ............ 49 APPENDIX 2 WHAT ACCOUNTS FOR THE HIGH SUPPLY ELASTICITY ESTIMATES BY PETERSON ........................... .... 72 - iii - LIST OF TABLES Page Table 1: Degree of Fit (R2) - Selected Equations ...................... 16 Table 2: Comparison of Elasticities - The Importance of Country Effect .......................................................... 21 Table 3: List of Experiments ............... ................ ... .......... 33 Table 4: Price Elasticities Under Different Price Normalizations ...... 34 Table 5: Price Coefficients in Experiments Using Lags ................. 35 Table 6: Experiment G: Nerlove Lags with Undeflated PPP $ Prices ..... 36 Table 7: Experiment K: Basic Specification Without Interaction Terms. ...... ......... ........................ ..... *a........o.o......37 Table.8: Sensitivity Results for Shifter Variables ....o......... - ... 38 Table 9: Time Trend of Agricultural Prices Deflated by US Wholesale Prices 1990-1983...o... o .. .... .-. ..... . .... 48 - iv - PREFACE The size of the long-run response of aggregate agricultural supply to a large and permanent change in the agricultural price regime in developing countries is a matter of significant interest for Bank policy analysis. The study reported herein was an attempt to answer this question definitively by the use of repeated observations on a large cross-section sample of countries, including countries with artificially low and artificially high agricultural price regimes. The repeated observations allowed analysis of within-country variations and thereby controlling for the country-specific factors not accounted for explicitly. The cross-country analysis facilitated a between- country comparison. Such a comparison is thought to better approximate long- run behavior. The major findings of the study, which come mainly from within- country variations in the data, are that (i) the aggregate supply response to output price changes is highly inelastic; (ii) the input-demand elasticities are also inelastic, but are larger than output supply elasticities; and (iii) the major explanation of variations in output is attributed to shifter variables which determine the level of supply functions. The results for the price elasticities are consistent with results from studies on aggregate agricultural output response using individual country data. Aggregate output is therefore demonstrated to be much less price-elastic than output of individual commodities. They are also consistent with short-run behavior. In order to obtain the long-run response it is necessary to explain the variations in the shifters themselves. The shifters represent two kinds of fixed factors, public inputs and private inputs. The framework of supply analysis is only pertinent for the determination of the private inputs. The public inputs are determined by policies which do not follow a consistent pattern and sometimes generate quantity-price relationships which are opposite to those observed for the private inputs. Clarification of this issue of, the need to differentiate public and private responses has arisen only after the completion of the present stage of the empirical work and, in particular, after puzzling over the lack of results from the cross-country analysis. Before further empirical work is undertaken an improved specification of agricultural supply relationships is essential. The development of an improved aggregate supply specification will be pursued as the next stage of the project. Ronald C. Duncan Chief Commodity Studies and Projections Division Economic Analysis and Projections Department Economics and Research Staff I I. INTRODUCTION* The nature of supply response is a subject often encountered in evaluating the effects of economic policies. The concept of supply response concentrates on the output-price relationship. However, output is affected by other factors and that complicates the empirical detection of the supply response. Any measure that affects resource allocation affects both prices and outputs. When dealing with agricultural supply, a distinction should be made between private and public inputs. Decisions on the private inputs are made by farmers jointly with their decisions on outputs. On the other hand, the decisions on the public inputs are made by non-farmers and as such represents a decision disjoint from the decision on output. True, in some sense it can be assumed that public decisions are made in anticipation of their effect on output. Still, it is a decision made separately from the decision on output. This distinction raises a natural question. What part of the change in output can be related directly to changes in prices and what part is associated with public inputs. Aside from the academic interest, this distinction between the effects of price and public inputs is largely inspired by government policies toward agriculture. A major policy problem related to the aggregate supply response is that price regimes in many developing countries discriminate 1/ Hans Binswanger .AGRES), Maw-Cheng Yang and Alan Bowers (EPDCS), and Yair Mundlak, Hebrew University of Jerusalem. Our thanks go to Fataneh Semsarzadeh and Hye-Sook Chung for assistance with the painstaking task of filling in missing values, and to Ron Duncan who managed the project. Our gratitude also to those persons and institutions who supplied data. - 2 - substantially against agriculture. Discrimination takes the form of taxes on production or exports, protection given to other sectors of the economy, and/or overvalued exchange rates. At the same time some of these countries provide the infrastructure, in terms of both physical and human capital, that facilitates growth of output even though prices are suppressed. Agricultural supply response is studied at different levels of aggregation. 1/ Individual crop studies usually show quite elastic response. Such studies provide pertinent information on the composition of output. Thus, in dealing with the question of food versus cash crops, it is important to know the dependence of output composition on prices. This is an entirely different question from that posed above about the level of aggregate output. For example, for India, Bapna, Binswanger and Quizon estimated seven individual crop supply elasticities (and their cross supply elasticities) in a system of equations. The individual own-price supply elasticities ranged from near zero to as high as 0.77. The resulting aggregate supply elasticity is only 0.1. This same elasticity was also found in estimating the aggregate supply elasticity in a single equation with the same data. Bapna found a very similar result with a different data set from India. Direct estimates for aggregate supply response for the developed world are also typically near zero. 2/ An exception to these findings is Peterson who used cross-country data and obtained an aggregate supply elasticity of 2.0. He argued that the low estimates, which have been based on 1/ For a review of some empirical analyses, see Askari and Cummings. 2/ See Griliches for the US. Pandey et al. review findings on aggregate agricultural supply response in Australia, and Colman and Rayner review findings in the United Kingdom. -3- time-series data for individual countries, are invalid because they do not reflect the investment impact of changes in long-term price strategies. The reason is that individual country time series will provide an invalid experiment for the measurement of long-run price elasticities, because countries generally pursue high or low price strategies for decades, with price peaks and troughs around these policies being maintained for only short periods. These price movements would not lead to the kind of investment response which would be seen if a country were to abandon its discriminatory stance for good. This argument ignores the inconsistent policies alluded to above that generate exactly the opposite relationships from that argued by Peterson. Contrary to the private sector behavior, to which Peterson's description applies, the public input may be positively correlated with lower prices. One way to approach the problem empirically is to utilize time series, cross-country data. This is the purpose of this paper where we analyze- the effects of price and various public inputs on the aggregate agricultural supply. The sample consists of annual observations for 58 countries for the period 1969-1978. The public inputs, to be referred to as shifters, are research, extension, adult literacy, life expectancy, roads, irrigation, a measure of comprehensive capital. In addition, a measure of climate is included. The major findings of the study are: A weak positive supply response is obtained from the variations over time for the individual countries (within-country variations). A negative supply response is obtained from the between-country variations. The shifters, as a group, account for most of the variations in supply in the within-country and between-country analysis. These - 4 - findings are in contrast to the result obtained by Peterson. In view of the importance of the issue we analyze the reason for the difference in the two studies and conclude that Peterson's results were obtained using FAO price data that since have been revised by FAO. Our analysis is based on the revised data. Does it mean that prices have no or almost no role in the determination of supply? The answer is no but the role is somewhat more difficult to detect. We indicate that the within-country estimates measure the short-run supply response and as such our estimates are consistent with estimates obtained from the data of the same source for the agricultural production function. On the other hand, the cross country analysis does not provide estimates for the long-run supply response. In Appendix I there is a detailed description of the data used in the analysis--the data sources, the methods of indexation, and the estimation procedures used to fill in missing values. In Appendix 2 there is a full discussion of the differences between Peterson's results and the results obtained in this study and an attempt to explain the differences. The plan of the paper is as follows: Section II discusses the data and the variable definitions. Section III describes the statistical framework. Section IV compares within-country and pooled regression and thereby makes inferences about the importance of the country effects. Section V compares various specifications, using the within-country estimates. Section VI places the results within an analytic framework, emphasizing the short-run supply function. Section VII discusses briefly the issues related to the iong-run supply. II. DATA, VARIABLES AND ESTIMATION TECHNIQUES We will be estimating supply functions for two outputs, aggregate crop output and aggregate livestock output and two input demand functions, fertilizer and tractors. Data constraints do not allow the extension to more inputs, e.g., labor. The system deals with only a small fraction of the agricultural input. Yet, fertilizer can be considered as representing the advanced technology inputs. Two additional alternatives are added. First, output is aggregated to total agricultural output. In this case the system consists of one output and two inputs. Second, crop output is decomposed into area and yield. Consequently, in this case there are two outputs and three inputs. The independent variables consist of prices and different measures of human capital and infrastructure. The latter affect the choice of the implemented technology. The prices include the two output prices, the price of fertilizer and in some cases a measure of wage rates. Given the fact that we estimate a system, system constraints could be imposed. However, as explained above, the system is far from complete and therefore such constraints would either be unjustified or of little help. Furthermore, the cross-equation constraints are quite commonly rejected even in data sets where the unit of analysis is much closer to the actual farm- level producer to whom the constraints apply. Therefore, imposing them on cross-country data, where aggregation problems are severe, seemed to make little sense. In the absence of theoretical restrictions we opted for siimplicity and estimate the equations in double-logarithmic form. The only variables not in logarithms are Literacy and Irrigation, described below, which are measured as percentages. -6- A detailed discussion of the data sources, data transformations and gap filling procedures is given in Appendix 1 but a brief discussion follows. The individual output and price data come from FAO. They were aggregated using multilateral Fisher quantity and price indices. In using multilateral Fisher indices we depart from the previous literature on cross- country production functions which has used wheat equivalent indices with international commodity prices for some specific base year or with an average of commodity prices for several countries. The multilateral Fisher indices use average world market prices and production over the period 1961-81 as base weights. Thus, the basis is less volatile and this minimizes the impact of year-to-year fluctuations in world commodity prices on the index for each country. For each year and each country the Fisher index is computed relative to the international base quantities and prices. Because the Fisher index uses both base weights and individual country-by-year weights in its computation, the indices have greater country specificity than a Laspeyres index would have, or the wheat equivalent indices normally used in international cross- country production function studies. 1/ The other dependent variables are tractors, and fertilizers, measured as total nutrient tons of all N, P and K 1/ For a discussion of the merits of multilateral translog indices in cross- country comparisons, see Caves, et al. We were unable to use these indices as many countries do not produce all commodities, leading to many zero observations. Because translog indices use logarithms of quantities and prices they are not defined when zero observations occur. However, the logic of the multilateral indices is that base quantities should be defined as averages over the entire period and the index should use both the base weights and individual country-by-year weights to increase the level of country specificity. Our multilateral Fisher index does both of these. - 7 - applied. The prices of fertilizers are indices which weigh the nutrient tons by the unit costs of nutrients in those straight fertilizers for which data were available in a given country. Both the tractor and fertilizer data are from FAO. In addition to prices, the equations contain the following shifter variables: Research is introduced in two alternatives. 1/ The first is one where man-years of research and the cost of research per man-year are entered as separate variables. Because the quality of researchers and their cost differ across countries, the research cost variable is used to "clean out" the effects of differences in researcher quality and/or cost. 2/ Its coefficient has no other interpretation because quality and price effects are confounded and only the "research years" variable is to be interpreted. In the second specification the stock of research expenditures per hectare of agricultural land is entered as the only research variable, i.e., no attempt is made to separate the quantity and quality or cost dimensions. Extension is measured in the number of extension agents. The farm population is used to normalize this variable because the impact of an extension agent depends on the number of people he can contact. As we do not have the number of farmers we used farm population as a proxy variable. 1/ The data on research and extension were supplied by Ann Judd and Robert Evenson. 2/ It should, however, be pointed out that this is at best an approximation; because of restricted mobility of researchers, the cost is not a good indicator of quality. -8- In a number of cases we have also included the following interaction terms: Research x Extension (RES x EXT) and Extension x Literacy (EXT x LIT). The hypotheses tested here are that extension has a higher payoff where the research effort is high, and that the payoff to extension is lower the better educated the farmers. The other "human capital" variables entered are Life Expectancy as a health investment measure, and Literacy as a schooling variable. 1/ Ideally, one would like to have a measure of health of the farm population rather than of the population as a whole. And it is the education of farm operators which is more relevant than the education of the population as a whole. Both of these variables are therefore crude proxy variables. The agricultural endowment of a country cannot be easily measured by any single land variable, because land quality and climate differ enormously. Moreover, the FAO land-use statistics are partly based on how land is actually used, not on its potential. Agricultural land can be expanded at the expense of forests, for example. The variable Agricultural Land is therefore only used to normalize research expenditures and irrigation variables. It includes the FAO classification arable land, land under permanent crops, and meadow and pasture land. Potential Production is introduced as a crude attempt to capture a country's innate agricultural potential. It measures potential dry matter production in each country and is based on the study of Buringh et al., which 1/ The data are obtained from the World Bank. is the only worldwide study of agricultural potential currently available. 1/ Buringh et al. do not present their results on a country basis, but by major soil types. Country-specific measures were developed by mapping the soil types onto country boundaries and aggregating potential production of each soil type by the respective areas in each country. This is a very crude method. 2/ Because potential production is a variable which does not vary from year to year, it drops out of the analysis when country effects are allowed for. The other agricultural endowment variable included is Irrigation, the FAO estimate of area irrigated at least once per year. This variable is normalized by per unit of agricultural land. Rural Population Density is included as a separate endowment variable to allow for the Boserup hypothesis that an increase in population density leads to agricultural growth. Two mechanisms are believed to be at work: higher population density allows greater specialization of the population into farm and non-farm activities and among the various farm outputs, leading to gains from specialization. Moreover, population density reduces the per capita cost of providing infrastructure and a variety of services to the population. There are many kinds of infrastructure and services such as roads or extension. Some of these are included as separate variables. The inter- pretation of the population density variable, therefore, is that it captures those economies of higher population densities not already captured via the specifically included variables. 1/ The recent and more detailed study by FAO excludes the developed world. 2/ Buringh et al. would not necessarily approve of our use of their study in this way. - 10 - For roads we have included as separate variables the Road Density, road length deflated by potential agricultural land, and Pavement, the percentage of roads paved. 1/ The final "endowment" variable included is GDP, average per capita income over the three years preceeding the sample period. 2/ Following Mundlak and Hellinghausen, this variable is viewed as a proxy for the overall capital endowment of the economy. The higher the economy's capital endowment the greater the ease with which agricultural investments can be financed, implying a greater capital stock in agriculture. This capital stock consists of variables which are included in the regression as well as those which are omitted. Since the two groups of variables are likely to be correlated, leaving out this measure of comprehensive capital would bias the coefficients of the variables actually included in the regression. Price Deflators and Price Elasticities When dealing with cross-country data, the problem of exchange rate conversion arises. Nominal exchange rates cannot be used because currencies can be artificially over- or under-valued via government intervention. We have explored alternative specifications. The first is to use relative, rather than absolute, prices by which the whole issue of exchange rate conversion can be sidestepped. The first variant of this is to follow Peterson and deflate all output prices via the fertilizer price. One disadvantage of this procedure is that in developing 1/ The data are published in World Road Statistics of the International Road Federation. 2/ Source: World Bank. countries fertilizer has been in short supply and there was an excess demand at the official prices, causing the deflator to be biased downward. The degree of such bias varies with time and may be difficult to characterize. In addition, fertilizer is often not a major input and it would be preferable to deflate by the price of a more important input. This led us to use the urban wage rate as the normalizing variable. We assumed that urban wages would be the best indicator of the opportunity cost of agricultural labor. The disadvantage here is that labor quality may differ systematically across countries, and that differences in wage rates may measure the quality differences as well as differences in opportunity cost. We return to this hypothesis below. The other main alternative is to use purchasing power parity exchange rates (PPPR) to convert domestic currency prices into dollars. Kravis et al. have estimated such exchange rates for all the countries included in the analysis, although for only a subset of the countries were these PPPRs based on actual sample surveys. For the other countries PPPRs were based on prediction from multiple regressions on country characteristics. Moreover, for all countries, changes over time were estimated via the difference between US and domestic inflation rates. Thus there may be a serious accuracy problem with using PPPRs. - 12 - III. THE STATISTICAL FRAMEWORK The model consists of several equations which have the same presentation: Yhit = 'hit -itsh + Uhit i = 1,...,I country index, t = 1,...,T time index h = 1,...,H equation index. It is assumed that the disturbance is distributed as: uhit - N(O, a°) uhit are independent across countries and over time but not across equations. However, the vector of explanatory variables, xit, is basically the same in all equations. These variables are taken to be exogenous and therefore the cross-equation correlation of u contributes no additional information and the system is estimated by single equation techniques. Within this framework there are several variations. The intercept, uit is decomposed into country and time effects. The implications of such a decomposition are now well known but nevertheless some aspects are reviewed here because they help to interpret the empirical results. 1/ Most of the analysis was conducted by allowing for a country effect, so that the intercept is written as pit :- p +.p.i In this case, the estimator of 3 is referred to in the literature as a "within" (country) estimator because it is obtained from within-country- variations of the variables in question, namely, yit - y. or xit - X. whe yZ yit etc. Such an esrimator does not t 1/ See Mundlak (1961) and (1978). - 13 - utilize the between-country variations. The country effects represent variables not introduced explicitly into the analysis. Such variables may not be known to the researcher, or known but unquantifiable. The omission of country effects from the analysis can be analyzed in the same way that omission of variables is analyzed. It will bias the estimates of the coefficients whose variables are correlated with the country effect and at the same time will increase their precision. Thus, when the effects are correlated with the variables, it is desirable to allow for country effects because otherwise we will get precise estimates of wrong coefficient. In terms of the present analysis, if the productive countries have a lower price for the agricultural product, failure to allow for country effects will result in a negative bias of the price coefficient. The supply then may seem to be inelastic or even negatively sloped. Similar to country effect, there may be time effect as well. The time effect largely represents changes in technology. Such changes are assumed to be captured in our study by the variables representing human capital, infrastructure and comprehensive capital. But, if such changes are not fully captured by those variables, estimates not allowing for time effect will be biased. The analysis of the expected bias is similar to that discussed above. Since productivity increases with time, as do all the shifters, one would expect, on the whole, a positive bias for their coefficients. On the other hand, if product prices tend to decline with time, neglecting the time effect would tend to bias downward the price coefficients. The opposite is true for input prices which tend to increase with time. Time effects can be allowed for in the same way that country effects are allowed for. The intercept is written as pit = 4o + pt and the variables adjusted for the effects - 14 - are: Yit - y.t, etc., where yit = It=l Yit* Finally, the time and country effects can be allowed for jointly. In this case, the intercept is written as 40 + vi + litnd the observations adjusted for the effects are: yit - Yi - Y.t + Y. The foregoing discussion centered on the variations in the intercept. There can, however, be variations in the slopes, i.e., in the 3 coefficients. Such variations can be random, or they can be related to systematic variations correlated to variations in some variables. To illustrate, we can write for a given coefficient: B = it0 + rz + w where z is a variable (or a vector of variables) and w is the non-systematic component in B. Substituting this equation for a in the regression: y = V + xB + u yields y = . + W0x + xzi + (xw + u) The systematic part of this equation contains an interaction term, xz. In addition, the error term now contains an additional component, xw. This component is still uncorrelated with x but its variance is not. In this study several interaction terms, have been tried but no effort was made to allow for the effect of such inclusion on the variance structure. A large number of independent variables were included in the study. Some of these variables are correlated. For instance, many of the variables include various forms of capital goods. A country which is more affluent is likely to have more of the various components. This is simply saying that the data may show multicollinearity which, as is well known, causes large sampling - 15 - variance of the estimators and therefore low reliability. To overcome this, we used the method proposed by Mundlak (1981) which combines principal component and multiple comparisons. Basically, the method imposes the largest possible number of linear homogeneous constraints that can be imposed on the model jointly. (The emphasis is on jointly). Said differently, the method extracts and disposes of the maximum amount of noise in the data. Suppose that initially the equation consists of k variables (algebraic rank). Using the procedure, it is found that it is possible to impose r restrictions of the form CS = 0, where C is a r x k matrix of rank r. By this it is meant that the hypothesis CB = 0 is not rejected at a preassigned level of significance. Then we say that the statistical rank of the x matrix, the matrix of the independent variables, is k - r. The method produces the matrix C with the highest rank, given the particular sample. The statistical rank is reported in some of the tables. The difference between algebraic and statistical rank gives the rank of C that can,be referred to as the rank of the null space. - 16 - IV. THE EMPIRICAL IMPORTANCE OF COUNTRY EFFECTS In this section we deal with a specification that was selected on the basis of the within-country estimates as explained in the next section. The specification itself is described in Table 2. The equation was estimated using four different estimates: Pooled data (P), Within-country estimator allowing for country effects (C), allowing for time effects (T), and allowing for country-time effects (CT). The values of the R2are presented in Table I for four equations: area, yield, crop output and aggregate output. Table 1: Degree of Fit (R2) /A Selected Equations Equation P C T CT Area .8792 .9986 .8578 .9986 Yield .7925 .9805 .7896 .9810 Crop output .8883 .9954 .8527 .9955 Aggregate output .9173 .9977 .8957 .9978 /A The P specification includes MPDM which is omitted from the T-specification. It is for this reason that the R2 of P is larger than T. The variables included in the equations are listed in Table 2. The most general model is CT. Imposing no time effect on this model leads to the C model. An F-test of the null hypothesis of no time effect does not lead to the rejection of this hypothesis in the four equations presented in Table 1. Thus, in what follows we concentrate on two specifications, C and P. The difference between the two is the inclusion of a country effect in C. Note, however, that P includes a country-specific variable, MPDM. This variable is not included in the C specification because it remains constant over time. Thus, the country effect, which is statistically significant represents variables over and above our crude measure of physical potential, MPDM. - 17 - The estimated coefficients of the two specifications are presented in Table 2. As indicated in Section III, omission of the country effect biases the coefficients whose variables are correlated with such effect. The direction of such correlation can be detected by comparing the coefficients of the same variable in the two specifications. For instance, the coefficients of the aggregate product price in the aggregate output equation are 0.0452 and -0.5992 for the C and P specifications, respectively. The difference, [-0.5992 - (0.0452)], is negative. That indicates that countries with a high country effect have lower prices. A high value for the country effect means that, other things being equal, the output in such a country will be larger than that obtained in a country with a low value for the country effect. A similar result is observed for crop output. Tracing this to the two components of crop output, it is seen that the source of the bias is primarily in the area equation. That implies that countries which, other things being equal, tend to plant larger area have lower prices. A similar conclusion applies to the yield equation but there the difference is by far smaller. On the surface it may suggest that large area and high yield cause lower prices. This suggests a simultaneous equation bias due to the omission of a demand equation from the analysis. This possibility is not pursued here, it will be treated elsewhere. The present discussion deals with other issues. The coefficient of fertilizer price is positive in all equations, though in most cases it is not significantly different from zero. Furthermore, a comparison of C and P equations indicates that in all cases the bias due to country effect was positive. It thus appears that low output prices, high fertilizer prices and positive country effects go together. This is particularly so for the acreage - 18 - equation. This implies that low ratio of output to fertilizer prices was combined with strong acreage response. Since international prices are common -to all countries, and abstracting from differences due to transportation, it appears that countries with larger country effect in the area equation taxed agriculture more heavily. In contrast to the price effects, the effect of the shifters are strong and significant and in fact they account for most of the variations in output. The shifters are discussed in more detail in the next section. At the present we are mostly interested in the relationships between the shifters and the country effect. It is to be noted that the coefficients of most of the shifters are significantly different from zero in the pooled equations and their effect on output is positive. An interesting exception is "scientist man-years." We return to this variable in the next section. In evaluating the research and extension variables it is important to note that they appear alone and also in interaction terms. Life expectancy behaves in an inconsistent way whereas adult literacy seems to have a positive contribution. The physical infrastructure variables, with the exception of roads, have robust and strong positive influence on output. It is certainly clear that it is not all a human capital show. If anything, it is capital in all its various forms, human and physical. Of particular importance are two variables, population density and GDP per capita. Note that the GDP affects yield and not area whereas population density affects both. We return to this below. When country effects are taken into account the results change qualitatively and quantiratively. Many of the coefficients become insignificantly different from zero, some change signs, and the others change magnitude. Thus, clearly there is a correlation between the country effects - 19 - and the various explanatory variables. The role of population density and GDP is largely unaffected by the introduction of country effect. Furthermore, quantitatively, the value of the coefficients of GDP is changed relatively little in all equations except for aggregate output. This indicates a positive correlation between overall capital availability and the production of livestock. On the other hand, population density is positively correlated with the country effect in the area equation and that effect carries into the crop output equation. In the within-country regressions, irrigation (measured as area receiving water at least once) has a large positive coefficient on yield but a negative one on area; the net effect on crop output and aggregate output is still a positive coefficient. We will come back to these results later. However, by comparing with the pooled regression it seems that irrigation is positively correlated with the country effect in the area equation, but negatively correlated with the country effect in the yield equation. What this means is that countries where irrigated area constitutes a larger proportion of their total arable land tend to have a negative shift in the yield equation, i.e., other things equal, have lower yields. Conversely, other things equal, they utilize more area. One interpretation of this is that countries with low yield potential have responded by constructing more irrigation. Paved roads and literacy are positively correlated with the country effects of all equations. The performance of the research-extension variables is more heterogeneous, and it will be discussed in the next section. In conclusion, it is clear that we were unable to reproduce Peterson's results and the cross-country comparison gives negative price - 20 - coefficients. On the other hand, the public inputs, in the form of human and physical capital, have a strong effect on output. In addition, two variables which affect the economic environment, total capital abundance, in the form of GDP, and population density have a strong positive effect on agricultural production. There is a strong correlation, in most cases positive, between the country effects in the various equations and the various variables. The analysis of the C equation, the within-country estimator, takes this effect into account. By so doing, we do not utilize the between-country variations. At the same time, we avoid some of the country-specific errors associated with the construction of the data. All country-specific errors in the data are eliminated under the within-country estimation. Table 2: COMPARISON OF ELASTICITIES- TIIE IMPIORTANCE OF COUNTRY EFFECT EPEWENI wIA11h AREA DEPINDENI VARIABLE: CROP cuIPUn VOIAKEWithin-country Estimator Pooled Within-country estimator Pooled VAJIAIII ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~VAR I AKE CR1? PRICE 0.0401 -0.461 sa CROP PRICE 0.01,03 a-0.6005 off IVE[SIDCK PRICEf 0.0093 0.5126 *aLIVESIOCE pRICE 0.0021 0.4313 8 fERitilIE PR IC~E 0.0221 a.0.0,610 FERJILIIERa PRICE 0.0093 0.0829a SCIEMIISI MAN YR5 0.0491 -0.6533 a.SCIERIISI MAN 9R! -0.0649 40.5061 RESEARCH COSI -0.0351 0.0390 RESEARCH CEISI -0.1215 as0.1510sa [iEHSIOm -0.0184 a.-P.1304 EIIEKSIOH 0.0313 -0.3711 a EMLIil 0.000 5. 0.0110 615 C11.1 -0.0001 0.0132aa RESIEII 0.0090 -0.0931** RESIEII -0.0021 -0.0809a. LIfE EIPECIAIiCJ 1.2110 a'2.839? 'ilifE EIPECIANCI 0.4924 0.8122 a 181 1641108 -0.1593 II .9842 *lINRISAII08 1.35i, a.. 1.8644 s ROADS 0.0025 -0.0168 ROADS 0.0031 0.0381 PAVEIIE11 -0.0004 0.0033 aF AVEIIENT -0.0001 0.004? 9 POPULAIION INSIDE 0.1421 4a .303k11 POPULATII ONSII511 0.2113 all 0.4916 " A01t1LIJIIERAEY 0.0032 i*0.0115 . ADLItII LI lEIACY 0.0048 oil 0.1100** GOP1 L00199 -06.0191 GOp 0.1110 ate 0.1036is I?OP 0.2445 'a POP 0.2963 a RANK 13 Ii RAW 13 Il SIAIISIICAI. RANK63I SIAIISTIEAL RAW 61Is I-SQUARE VAMIES I-SQUARE VP&IIS 0.9996 0.8192 0.9934 0.19 0.0893 KEPEIIIfR VARIA11.ia YIELD DEPENOERI VARIAILEs MEA JE1611 BiRPUI VARIAILEWithin-country estimator Pooled VRAL Within-country estimator Pooled CROP PRICE -0.0113 -0. I OR ao A66REfiAIE 111EE 0.0452 -0.5911 Its LIVESIIIC PRICE -0.0520 -0.1011 GOB FERTILIZER PRICE 0.0101 0.0391 fEiliILIE PRICE 1.0024 0.0331 SCIENIIGI MN VI 0.0363 -,O-.4120 too sciENIISI NlAN IRS -0.0963 of 0.1111 a.. RSEARCH C06T -0.0201 0.05861 RESEARCH COSI -0.0606 0.1021 a' iIENsION -0.0298 -0.3249 eat EliTNSION 0.1064 all -0.2061 *.. 111.1 1.0000 0.0121 ia' [milli -0.0006 soa 0.0024 aaRESKEIT 0.0081 -0.0808 ale RESiEDI -0.0124 ae. 0.0065al LIFE EIPEETANCT 0.4130 -0.3120 LIFE CIPECIAICY -4.4129 of -2.1584 a.R 14 A08110 1.242B a.. 1.5014 off IRR16AIION 2.6444 '0.1431 i ROADS -0.0069 4.0533 of ROADS 0.0242 0.0323 * AVEEIIKI 0.0003 0.0046 off PAVEtIENI 0.0001 0.0022 a POPUJLAIION 00511 0.3916 $la 0.4881 moo PIJPIJAIION DN5119 0.1556 a.0.1868 aaADLtl LIIERAET 0.0028 af 0.1061 Is# ADULI EiIERAEY 0.0005 0.0415 ac SP 0.2240 a 0.6103 esi GDP 0.2099 a. .1160 'a IDP 0.2468 l WDP ~~~~~~~~~~~~~~~~~~~~~0.0303 a lAW 12 1 RANK 13 17 STAT ISI1CAL. RAID 66 Is SIATISIICAl. RAIl 60 12 R-5OUARE VALUES I-SQUARE VALUES INS (1.5 ILS as * 0.9911 0.*113 0.1805 0.2925 - 22 - V. WITHIN-COUWNTRY ESTIMATES Often, empirical results depend strongly on the specification of the estimated equation. This section summarizes results obtained under various specifications which should help to place the discussion within an appropriate empirical framework. Table 3 summarizes the specifications tried. Specifications A to J primarily explore alternative price variables. Table 4 summarizes the results of using different price normalizations. The columns of Table 4 report the own-price coefficients and the coefficients of some other independent variables for each of the dependent variables, enlarging the number of equations to seven. Each price variable referred to is the level of prices of the preceding year. No further lags were included. Each of the estimation equations contained all of the foLlowing shifter variables in addition to the variables shown: Irrigation, 1/ Research Man-Years, Research Cost, Extension, Literacy, Life Expectancy, Road Density, Pavement and Population Density. The fertilizer price variable was also included in those specifications where prices were not already normalized by the fertilizer price. Consider first the issue of deflating by wages. Experiment A deflates prices by wages. Experiment B uses PPPR prices and introduces the wage variable separately. Normally one would expect wage increases to reduce output. But only in the crap area equation do we find the expected sign. In the equations for crop output, yield, livestock output and aggregate output, the wage variable has the positive sign. Such a sign would only be appropriate if wages measured labor productivity rather than cost. In this case, other 1/ Excluded sometimes from the crop area and livestock equations. - 23 - things being equal, higher productivity implies higher demand for labor and hence higher wages. We therefore decided to leave out the wage variable and instead include the per capita GDP variable as a better proxy for labor productivity. In what follows the results from experiments A and 3 are no longer considered. The remaining specifications for which we show results in Table 3 are as follows: Experiment C uses the basic specification with PPP prices and per capita GDP. Experiments D, E and F use producer prices relative to the fertilizer price, with experiment D otherwise identical to experiment C. In experiment E we explore whether the results on price elasticities are sensitive to the replacement of the GDP variable by a time trend in quadratic form. This substitution of an average trend (time) for a country-specific trend (CDP) has hardly any influence on the resulting price elasticities. Nor did it affect the results with respect to the shifter variables substantially. This indicates that we have succeeded in replacing the time trend with an economic variable without any sacrifice in precision. Such a replacement makes it possible to use GDP for cross-country comparisons, which cannot be done with a time variable. In experiment F, per capita GDP is again included rather than the time trend. The own-price variables were interacted with all the shifter variables to allow country-specific factors to influence the price elasticity. While the resulting average elasticities differ somewhat from the earlier results, these differences are not dramatic. Summarizing across experiments C to F we find the following: 1. Short-run aggregate output supply appears to be remarkably price inelastic. Own elasticities of output supply in no case exceed 0.06 for any of the output variables. - 24 - 2. All crop supply elasticities have the correct sign and are often statistically significant. 3. The livestock supply elasticities have a statistically signi- ficant negative sign and vary between -0.8 and -0.16. Given that stock adjustment is a feature of livestock supply, a negative sign for the price of the previous year is not necessarily alarming, and the issue will be pursued below. 4. The fertilizer demand elasticity is remarkably stable at -0.16 to -0.17. It is the largest of the elasticities found in absolute terms. 5. The tractor elasticity with respect to the output price varies from 0.09 to 0.13 and is significant. Thus; both inputs are seen to be more price responsive than the resulting output levels. Intuitively it can be attributed to the fact that the input elasticities represent expansion as well as substitution effects, whereas the aggregate supply elasticity represents only expansion effects. 6. The key results are not sensitive to the price normalization, i.e., whether PPP prices are used or prices normalized by fertilizer prices. There are minor differences, however. With PPP prices, crop area is price-responsive, but not crop yield. With prices relative to fertilizer price, the reverse is the case. - The question arises whether substantially higher elasticities result when lags are introduced. This is explored in Table 5. Experiments C and H incorporate simple distributed lag specifications. The full results of - 25 - experiment C are given in Table 6. In this experiment it is assumed that the lagged dependent variable captures all the effects of the shifter variables and they therefore have been excluded. Apart from the lagged dependent variable only the relevant price and cross-price terms are included. In experiments I and J we use free-form lags instead of distributed lags. A price term is included for each of the previous three years. The total supply elasticity is the sum of the elasticities from each of the three lagged prices. For both sets of price normalizations, the use of the free-form lag results in price terms lagged by one year having the same sign as the price terms reported in Table 4 (experiments C and D). These two sets of experiments make it possible to interpret the role of the lagged dependent variable. If it represents the expected price obtained by adaptive expectations its results should not differ significantly from the regressions with free-form, third-order lagged equations. If anything, the latter should give better results. But this is not the case. Moreover, the difference in the number of variables is substantial, but there is little difference in the statistical rank. This indicates a considerable redundancy in the formulation of the free-form price lags. The interpretation of the role of the lagged dependent variable is that it represents the technology and the level of capital items-, including unobserved components, existing close to the date of making decisions with respect to activities in year t. Of course, it does it with some error, but apparently it does it well. This interpretation must be kept in mind in the application of any long-run supply elasticities derived from the distributed lag estimator. To emphasize this, long-run elasticities derived from distributed lag analysis will be qualified as "long run." - 26 - The distributed lag experiments G and H come to somewhat different conclusions. When using output prices relative to fertilizer prices the "long- run" yield elasticity is 0.17 as compared to 0.13 in experiment C. The area elasticities are the same. Thus, the difference in the yield elasticity carries itself to the crop output--0.3 in experiment H as compared to 0.22 in C. But all other supply or demand elasticities are not statistically significant from zero in H, in contrast to G. Compared to a priori expectations, the experiment G gives che best behaved results and are therefore fully reproduced in Table 6: 1. All supply and demand elasticities have the correct sign. 2. As measured from the within-country variation the "long-run" supply response of any of the output aggregates with respect to its own price does not exceed 0.33. 3. While livestock is less elastically supplied in the short run, its "Long-run elasticity" of 0.32 exceeds that of crops of 0.22. Since livestock output requires capital investments, a larger difference between short- and "long-run" response is not unexpected. Aggregate agricultural output has a supply elasticity of 0.28, in between that of crop and livestock. 4. Fertilizer demand is quite responsive to the output price (+0.47) but less responsive to its own price (-0.08). Such a result is consistent with the hypothesis that in many cases the use of fertilizer was limited by supply availability. 5. Machinery stock is very inelastic with respect to the output price in the short run but very elastic in the "long run" (+1.22). However, this latter result is very unreliable since - 27 - the short-term coefficient on which it is based is not significant. With all qualifications, note that these results are consistent with those of Table 3 in that the fertilizer and machinery inputs are more price responsive than the outputs. The results with respect to prices for fertilizer demand in experiment G seem to be the cleanest and most consistent. Cross-Price Effects In the examination of the own-price effects we have seen that only in the distributed-lag experiments with PPP prices were all the own-price effects largely consistent with a priori expectations. We, therefore, will only discuss the cross-price effects for this specification. There are 12 cross-price effects in Table 6, of which eight are not statistically significant. The fertilizer price effect in the crop output and crop yield equation is positive and significant, i.e., it has the wrong sign. This result is consistent with the possibility that fertilizer utilization is largely supply-determined. The fertilizer price coefficient is also positive and significant in the livestock-supply equation. As livestock is usually less fertilizer-intensive than crop production, this result is not an inappropriate sign. The fourth significant cross elasticity is that of fertilizer demand with respect to the aggregate output price and it is positive, as we have already seen in Table 4. This result is also consistent with the possibility that fertilizers are to some extent supply-determined. On balance, therefore, we must conclude that these data and/or our techniques do not support the estimation of substantial cross elasticities at such highly aggregate Levels. - 28 - The Shifters As can be seen from Table 3, the experiments with shifter variables are primarily concerned with the treatment of research and extension. The other shifter variables were introduced in the same manner in all equations. The variations with research and extension involve (1) the shift from man- years and research cost per man-year to total research expenditures, and (2) whether interaction terms were present between the research variable (man- years or total expenditures) and the extension variable and between the extension and the literacy variables. In Table 7 we present the full set of results of experiment K, with PPP prices, research man-years and with no interaction terms. In Table 8 we show the sensitivity of the results to the changes in specification, i.e., whether prices are lagged or not, whether PPP or relative prices with fertilizers were used and with respect to the various specifications of the research and extension variables. Results from the seven experiments C, D, E, I, J, K, and L are summarized in Table 8. A sign of + (-) in an entry means that at least in one of the experiments the coefficient of the variable was positive (negative) and significant at the 10% level. An entry of zero means that in at least one of the experiments the coefficient was not significantly different from zero. A robust result means that only one of the three possible signs appears as an entry. We discuss these results for the independent variable in order of robustness of the findings. GDP. The mean GDP per capita over the previous three years is inter- preted here as a measure of comprehensive capital in the economy. Economies with larger capital-labor ratios will tend to have more capital-intensive activities, which will also affect their choice of technology (Mundlak 1984b). It is also possible to interpret the GDP as a demand variable, but in that - 29 - case it would not replace the other public capital items as discussed in the lagged price experiments. CDP is positive and highly significantly in all equations. From Table 7 we see that the largest GDP elasticities (0.45 and 0.42 and 0.37) arise in the tractor stock, the livestock supply and the fertilizer demand equations, respectively. The elasticities are lower in the crop output and crop yield equations (0.16 and 0.15, respectively). Aggregate output has a GDP elasticity of 0.23. The lowest elasticity, 0.03, arises in the area equation. The fact that the capital inputs (tractors and fertilizer) and the output of livestock (which requires high levels of livestock capital) have high GDP coefficients, while the crop area elasticity is very low, is consistent with an interpretation of the GDP variable as a capital- availability variable rather than as a final demand variable. Population Density. Rural population density clearly has a positive and significant effect on all output variables, except perhaps for crop area. It also increases the fertilizer input, but its effect on the tractor stock is sometimes positive and sometimes not distinguishable from zero. This last finding is consistent with the idea that tractors are a substitute for labor. Ignoring the tractor equation, the population density elasticity is the highest for fertilizer demand and livestock output (0.53 for both), followed by aggregate and crop output (0.28 and 0.27). The crop output effect is partitioned into nearly equaL area and yield effects (0.12). The high impact on fertilizer demand is consistent with the idea that distribucion of fertilizer is cheaper in densely populated areas where commerce is weLl developed. But the main effect may reflect the fact that the densit- of the population affects the cropping pattern--the more dense the population the Larger the proportion of crops which are intensive in inputs other than land - 30 - (in particular, fertilizer). This explanation also accounts for the high elasticity of livestock. The strength and consistency of the population density variable is completely consistent with Ester Boserup's hypothesis on the relationship between population density and agricultural productivity. For all other variables the results are much less stable. Research. As mentioned before, the research costs variable is only included to clean out cost differences in costs per research man-year. Our attention thus focuses on the research man-years and the research expenditure variables, only one of which was included in each of the regressions. The only robust conclusion one can draw from Table 8 is that research increases fertilizer demand. A somewhat less robust conclusion is that research increases aggregate crop yields, but perhaps tends to reduce crop area. Because of these apparently contradictory impacts, the results do not show a positive effect on crop output. As the results with respect to livestock output are also contradictory, research appears not to influence aggregate output in a positive way. These results appear to contradict a lot of earlier research on the impact of research on agricultural output. Further work will be required to discriminate among three hypotheses: (i) Our findings are correct and earlier studies falsely attributed to research the effect on agricultural production of left-out variables, such as population density, capital availability and infrastructure. (ii) Resea,rch is not measured accurately enough by our variables. (iii) There is too much heterogeneity in our data set to allow the effect of research to show up. Grouping of countries into more homogeneous groups is required. - 31 - Extension. With our crude extension variable we cannot show an effect of extension on either crop area or aggregate output. (An effect on mechanization was not anticipated and indeed does not show up.) Extension may, however, have an effect on crop output, crop yield and livestock outpuc. Note that our extension variable cannot measure quality of extension. Irrigation. Irrigation clearly has a positive effect on aggregate output and tractor demand. The latter effect is easily explained by the extra power requirements of more intensive cropping. It also appears to increase livestock output. The effects on yield and crop output are not consistently positive. However, the finding of no effect on crop yield comes from experiment I (with free-form lags and PPP prices) in which the regression had only a statistical rank of 3, and most variables had a low level of significance. Therefore, we can safely conclude that irrigation has a positive yield effect. Irrigation may have a negative effect on crop area, however. If aggregate demand is limited, newly-irrigated areas may compete with non- irrigated areas and the latter may decline. Just as in the case of research, the higher yields may substitute for area under crops. The aggregate output effect of irrigation may therefore not be as dramatic as usually assumed. Literacy. This variable appears to have a positive effect on crop yield and crop output. (Again the sole non-significant coefficient in the crop yield equation comes from a regression with a statistical rank of only 3.) Effects on aggregate output and fertilizer demand may be positive as well. Literacy clearly has a negative effect on crop area. Moreover, it consistently seems to reduce livestock output and tractor demand. - 32 - Life Expectancy. An improvement in this variable appears to lead to increased area and greater tractor demand. Positive effects on crop output, livestock output, aggregate output and fertilizer demand are also often found. Effects on crop yield are contradictory. But our results indicate a positive effect of life expectancy on agricultural outputs and input use more clearly than for several other shifter variables. Roads and Pavement. The most robust effect of both these variables is on tractor demand, with pavement having a particularly strong effect. Livestock output is also unambiguously associated with improvements in this measure of road quality. The following effects are not always statistically significant, however: road density tends to increase crop area, crop output and fertilizer demand, and these effects accord well with a priori expectations. Less in accordance with a priori expectations are the findings that road quality reduces crop area but increases yield, with the negative area effect dominating the yield effect so that crop output and aggregate output are also reduced. To summarize the results for the "within-country" estimators, fairly clearcut and positive effects on agricultural output are found for CDP, population density, irrigation and life expectancy. Research, extension, literacy and the road variables either have little effect, once the other shifter variables are introduced, or are so poorly measured that this effect does not show up clearly. Aggregate price elasticities are low for all agricultural output measures. In most experiments they are measured at less than 0.1. "Long-run" elasticities may be somewhat higher, but clearly not exceeding 0.3. Input demand elasticities are consistently higher than aggregate output supply elasticities. 3 33 - Table 3: LIST OF EXPERLMENTS Experiment Price Research Interaction GDP, Code Prices Lags Variable Terms Time or Wage A relative to I year man-years Research X Extension Neither urban wage and cost Extension X Literacy B PPP I year Wage C GDP D relative to fertilizer I year GDP E Time F all shifters with own GDP price research X extension extension X literacy G PPP lagged no shifters dependent H relative to no shifters fertilizer I PPP free expenditure none GDP form lags J relative to man-years and research X extension fertilizer cost extension X literacy GDP K PPP 1 year none GDP L relative to expenditure research X extension fertilizer extension x literacy GDP Table 4: PRICE ELASTICITIES UNDER DIFFEKENr PRICE NORMALIZATIONS Deflated Relative to Wages PPP Adjusted Domestic Price Prices Relative to Pertiltzer Prices With all Interacttions Own Price Own Price Wage Own Price CDP Own Prtce CUP Owo Price Time (Time)2 Own Price EXP. A EXP. B EXP. C----- EXP. D------ EXP. e -- --- -------_- EXP. F------- Crop Area 0.062*** 0.054*** -0.090*** 0.023** 0.030*** 0 0.053*** 0 0.003*** 0*** 0.025** Crop Output 0.083*** 0.049*^* 0.032*** 0.025*** 0.144*** 0.046*** 0.205*** 0.050*** 0.011*** 0.001*** 0.033 Crop Yield -0.005 0.030*** 0.032*** 0.003 0.169*** 0.059*** O.139*** 0.033*** 0.007*** 0.001*** 0.053*** Livestock -0.165*** -0.137*** 0.197*** -0.156*** 0.447*** -0.097*** 0.381*** -0.081*** 0.008*** 0.001*** n.a. Aggregate Output 0.017** 0.096*** 0.081*** 0.047** 0.230*** 0.018*** 0.255*** -0.004 0.011*** 0.001*** -0.004 Fertilizer Demand -0.128*** -0.158*** 0.049** -0.166*** 0.361*** -0.167*** 0.257*** -0.160*** 0.038*** -0.001** -0.171*** Tractor Demand -0.031* 0.200*** 0.110*** 0.130*** 0.459*** 0.089h** 0.439*** 0.086*** 0.006*** 0 n.a. - significanit at 0.01 level ** - significant at 0.05 level * - significant at 0.10 level Table 5: PRICt COEFFiClIENrS IN EXPERlMENTS USING LACS PPP Adjusted Domestic Prices With Free Form lag and Puce Reldtive to Fertilizer PltLs__ With Oitly Lagged Depenident Shifter Variables S Rdauk With Otily Lagged Depeadeiit Witih Free from lag med Shlifter Vaxtables S Ru-k L.P. C EXP. I EXP. It EXP. D_ Short Louug I Lag 2 Lags 3 L.ags Total Short l.oog I lag 2 lagb 3 lags Trotal Crop Ared 0.060*Ak 0.128 .029*** ^.018** -.032*^ -.021A** 5 0.(151** 0.125 0.0l0**k 0.010*** 0.016*A* O.UJ6*AA 4 Crop Output 0.168*** 0.218 .072*** .087 A* .047*** .206 3 0.181A * 0.300 0.01d** 0.022*A* 0.028 A* o.168AA* 3 Crop YieLd 0.106*** 0.130 .050*** .042*^ -.009 .083 3 0.12*1)55 0.171 0.(l05*** -0.00IA** 0.002* 0 .006A A 2 LIvestock 0.038** 0.324 -.092*** -.051 .013 -.130 12 0.1()2 0.043 -0.028** -0.047*** -0.)28*5* 0.l03AAA 9 Aggregate Output: Own Prtce 0.179*** 0.280 .088^** -.032 .008 .064 7 -0.004 -0.011 0.016 -0.017 0.O04 0.003 8 Fertlltzer Price 0.015 0.024 .006A** .029** -.005 FerLiltzer Dealudld: Owuu Price -0.039*a -0.081 -.124A**A .028 -.110*** -.205 9 -0.046 -0.114 -0.161** 0.048 _O.| k*AA -0.248AA 8 Ourpuit Price 0.226*** 0.468 -.058*** -.054 .212*** .099 0.161*A* -0.048 0.135A^* 0.248AA* Traccor Demand: 0.101*** Output Price 0.018 1.219 .175*** -.190 -.008 -.103 8 0.011 * 1.(25 0.101^** 0.023 O.l(J*A* 0.227A*A 9 *5* :- SSignificant dt 0.01 level 5* SStg1tfIca.1t at 0.05 level Sitg1ulfIla1it at 0.10 level Table 6: EXPERIMENT C: NERLOVE LAGS WITH UNDEFLATED PPP $ PRICES /A Dependent Crop Price or Livestock Fertilizer Lagged Statistical Variable Row Aggregate Price Price Price Dependent Rank Crop Area Coefficient 0.060*** -0.005 -0.004 0.530*** 3 Long-run elasticity 0.128 -0.011 -0.009 Crop Output Coefficient 0.168*** 0.002 0.033** 0.227 3 Long-run elasticity 0.218 0.003 0.043 Crop Yield Coefficient 0.106*** 0.006 0.025 0.187 3 Long-run elasticity 0.130 0.007 0.031 Livestock Coefficient 0.001 0.038 0.016 0.884*** 4 Long-run elasticity 0.011 0.324 0.136 Aggregate Output Coefficient 0.179*** 0.015 0.361** 3 Long-run elasticity 0.280 0.024 Fertilizer Demand Coefficient 0.226*** -0.039*** 0.517 2 Long-run elasticity 0.468 -0.081 Tractor Demand Coefficient 0.018 -0.011 0.985** 3 Long-run elasticity 1.219 -0.742 /A Long-run own-price elasticities are underlined. * Significant at 10%. ** Significant at 5X. *** Significant at 1%. t N 90 00.E4 0@O0 00 0i r U ¢ X P t r > i X D 0 r ?? $ U D il z r 0 D;* D | p §' 1 l ° a* Aa *S ;;I - H 1 I *i VI b El in en | 28 i! tn | . th "0 & ":l th {° | l ^ ga - 11 RZ 11 8to | I;o ^ orfi b t . I i v : 4~ : I i U NOY 00~ O~ 1 0 Il 0- 04 s r i on p i* on 0TT On F,~~~~~~~~~~~~~~~~~ A 0 a -_E g"8§8§~3e io*oi-~ oFtf ,1 fiIn I~~~~~~~z J 4hin0,. 1 ; t " .g st | e i1 x z 4; ° ° 1; 7'T|3 l O.;| St Xi. Ao co s uuZ t [ fi-i I t0 ?t _ ? PI PI U U_ U i5! Im v an 00 1 -1-0,:6~~~~~~ C; 4 C; C; 0 a y'1 0 09 0 0 0 IOn I !~~~~~~~~~~~ ;~ ~ ~ ~ t I I I IF 14 s II! 2 1I Table 8: SENSITIVITY RESULTS FOR SHIFTER VARIABLES Irrigation Research Research Research Extension Literacy Life Roads Pavement Population CDP Manye'ar Cost Expenditure'B Expectancy Density Crop Area 0 _A 0, _- 0 _ + 0, + 0, _ 0, + + Crop Output 0, + 0, - - 0, - 0, + + 0, + 0, + 0, - +e Crop Yield 0, + 0, + 0, - + 0, + 0, + +, - 0, +, - 0, + + + Livestock +A , +, 0, + 0, + O, + _ O 0 - + + + Aggregate Output + 0, - 0, - 0, 0, + 0, + 0 0, - +F Fertilizer Demand 0, + + 0, + + 0, +, - 0, + 0, + 0, + 0, - + + Tractor Demand + 0,- + 0, + , - -+ + + - + + Experiments included: C, D,E, I, J, K, L. + : At least one of the results is positive and significant at 0.10 level. - : At least one of the results is negative and significant at 0.10 level. O At least one of the results is not significant at 0.10 level. /A Irrigation was only included in specification I and K. /B Researchi expenditure included only in specification I and L. - 39 - VI. INTERPRETATION To place the discussion within an analytical framework we repeat the following findings: 1. Consistent with virtually all the earlier literature we find that aggregate supply responses, when measured from within- country variations, are quite low. 2. Input demand responses with respect to the output price exceed the output supply elasticities. 3. A much greater proportion of the variation in output is associated with the productivity-related shifter variables such as capital availability, population density, irrigation, education and life expectancy. Also, to complete the picture it is noted that the literature reports considerably stronger supply response to individual products than to aggregate output. All this can be comprehended within a consistent theoretical frame- work if we consider the empirical results to be estimates of the short-run supply function. By this, it is meant the supply conditional on fixed factors. The discussion can then be generalized to deal with the long-run response. Length of Run To a large extent, studies of supply response are carried out within the microframework, applying results which hold for an individual firm directly to the sector as a whole. To place the discussion within such a framework, write the restrictei profit function: (1) i (p, w, K, T) = max (py-wv : y,xeT) Y,v - 40 - where y is a vector of I outputs; x is a vector of J inputs decomposed to variable (v) and fixed (K) components: x = (v,K) with dimensions (a,b), a+b = J; T is the feasible technology set; p is the vector of product prices; w is the vector of factor prices. It can be decomposed to conform to the decomposition of x. However, where ambiguity does not exist, such a decomposition is not made explicit. Using Hotelling's Lemma the product demand and factor supply functions are written: (2) yi(.) = , C(.) = (p, w, K, T) api alI Vj(.)= _ _ The strength of the response of y and v to changes in prices depend on the decomposition of x. The case when all inputs are variables (a ] and b = 0) is referred to as the long run and is denoted by: (3) y"(p, w, T) v (p, w, T) K (p, w, T) Empirical analyses are based on dated data. The decomposition of x co v and K is done according to the ease of changing the inputs within the period of analysis, usually a year. Consequently, the empirical analysis of (2) produces a restricted or short-run response. The relationships between the - 41 - restricted supply as given in (2) and the unrestricted supply as given in (3) is given by the identity (4) y(p, w, K ,T) _y *(p, w, T) Hence ayt ayi ay. iK api api j Kt gpi Thus, the long-run response is the sum of the short-run response and the impact of the price change on output via its impact on investment in "fixed" factors. Using logarithmic differentiation, the result can be expressed in terms of elasticities. a In Kif (6) C u = e. + B*. - L U1 ri 1 ijJ 3a In Pi where C i and eriare the unrestricted and restricted elasticities, respectively, and St. = a ln yi/a ln KM. Sometimes e i and E iare referred ij j 1.ul. ri. to as the long-run and short-run elasticities. Note that the relationships in (6) are obtained under the identity in (4). The quantitative importance of the foregoing discussion can be illustrated by a simple example, using a single output production function. For simplicity, let the production function be homogeneous of degree p < 1. The corresponding cost function can be written as (7) c = O(w)Yl/u with marginal cost - 42 - (8) dc _ @(w) (1/p) - I dy Vi Unless otherwise indicated, competitive conditions are assumed and hence dc/dy P. To derive the supply elasticity, we use (7) and (8): (9) dCIn y = d In = l/d[ln O(w)-ln u + (- -1) In y]/d In y diln p d In (dc dy U The sensitivity of the supply elasticity to the degree of the homogeneity of the function is illustrated by some numerical values: u .95 .9 .8 .6 .4 .2 .1 E 19 9 4 1.5 .67 .25 .11 To relate this to our discussion, write the underlying production function as: Y = F(v, K). The restricted maximization leads to (10) 3 a In y w.v. j - a ln vj py - I where M. is the share of factor j in total output and 8. is the production elasticity of factor j. If the production function is homogeneous in v, the a degree of homogeneity is E M.. If it is not homogeneous, it can be j=l J approximated locally by a homogeneous function of degree e, where e is the scale elasticity defined, using obvious notations, by: a F.d(v.X)/dX x d F(Xv, K) | = j e(v,K (vK)=dX-= j=l F(v, K) - 43 - a a a (11) E F. v./F(v, K) = E 3 = E M. j=j J J j=l J j=l J For an early discussion of such an elasticity see Carlson. When the function is homogeneous, e is constant everywhere. Otherwise, it is defined locally and as such it depends on the classification of inputs to v and k. In any case, what is important for the present discussion is that it can be evaluated as the sum of the factor shares of the variable inputs. If we consider labor, land and capital to be largely fixed in the short run then the value of e is relatively small. There are several estimates of such elasticities for the global agricultural production function. Mundlak and Hellinghausen suggested elasticities of 0.14 for land, 0.16 for livestock, 0.09 for fertilizer, 0.06 for tractors, 0.01 for irrigation and 0.38 for labor and 0.16 for unobserved capital, public and private. These are somewhat different from the results obtained by Hayami and Ruttan, but for our discussion convey a similar .message. Thus, if agricultural labor, land and livestock are fixed in the short run, how much can the supply change in the short run in response to price? The sum of the variable elasticities is 0.16 and the supply elasticities = 0.86 = 0.19. This result depends somewhat on the treatmenc of the public inputs. In any case, the qualitative conclusion remains unchanged. Varying Factor Supply The division to variable and fixed inputs is to some extent arbitrary. Such a dichotomy suggests a zero supply elasticity for the fixed inputs and infinite elasticity for the variable inputs. The latter assumption is made, generally implicitly, in many of the oroduction analyses using derivatives of the profit function. It holds true for the individual firm but not for the industry as a whole. For instance, the recognition of the - 44 - importance of fertilizers which accompanied the green revolution could not be matched with unlimited supplies at the going price. Hence, fertilizer is to some extent fixed. On the other hand, the supply of the fixed inputs can be changed if prices justify it and if time is allowed for. But even then, in many contexts, the supply of some important factors of production may be in inelastic supply. For example, the labor supply studies of both Rosensweig and Bardhan imply rural labor supply elasticities of less than 0.3 for India. Even when a migration response is added, the supply elasticity does not rise substantially beyond 0.3 (Quizon and Binswanger). Taking these considerations into account, the analysis can be generalized by introducing the factor supply functions: (12) x. = S.(w. , T.) where Tj represents the technology (or tastes when dealing with labor supply) in the sector producing input j. Let s. = a In x./3 ln w.. The smaller are the factor supply elasticities, the smaller is the product supply elasticity 1/ This can be seen by incorporating (12) in the analysis and rewriting (9) as: (13) {d In ¢(w) + (1 _1-1 d ln y 1J Evaluate the first term on the RHS: a ln b(w.) d ln w. a In x. (14) d In Iw) d ln y J a ln w. d ln x a ln y 1/ For a discussion of this, see Friedman and for applications see Brandow and Floyd. - 45 - Evaluate the RHS of (14) term-by-term, using (7), a In c = a[ln t(w) + 1 In y]/3 in I.. a I (w) 3 in w. 3lnw. and using Shephard's Lemma 3 In c x c a y ln w; -j a In x. Next, d In w./d In x. = 1/s.. Finally a = E. is the elasticity of xj i i j ~~~~alny - jy along the expansion path obtained under constant w. Then, 1 ~~~~~-1 (15) E = [(- -1) + E E. a./s.] . s.;O, a=J jy .J 3 3 A value of zero for sg is ruled out in (15) because it is assumed that all inputs are allowed to vary. However, sj can be small for some j, thereby making the supply elasticity small. This is true even if we take the production function to be linear homogeneous, so that 1, and Ej = 1 for all j. In this case (16) E = 1/E z = 1, a = J, b = 0. If some inputs are fixed so that s* = 0 for some j, the expansion path is not linear anymore. In this case, the result is modified. To simplify, suppose there are only two factors, x1 = v, x2 = K and y = B1, 1-y = 82. In a ln x a In - /8,, aln x2/3 In y = 0 and a = I Then (17) C =( )+s ] - 46 - = 81[(i - + s 1 s~~~~~ This expression is comparable to (9), and shows clearly how the supply elasticity is modified by the factor supply elasticity of the variable factor. Thus, for 3 = 0.5 and s, = 1 the product supply elasticity is 1/3 instead of I when s, = 5. To conclude, the estimates obtained for the aggregate supply elasticities from analysis of country data are consistent with the analysis leading to the short-run supply funccion under the assumption that labor, land and some of the capital items are fixed in the short run. That brings us to the question of what causes the changes in these and related variables in the long run. - 47 - VII. ON THE SUPPLY IN THE LONG RUN To place the analysis of the long-run supply within an appropriate empirical background we present in Table 9 the time trend of the terms of trade of selected agricultural commodities for the period 1900-1983. The prices of these commodities are divided by the US WhoLesale Prices index. The figures in the column headed "b" are the slopes of the semilog regressions of prices on time. The t-ratios are given in the column headed "t" and the fit is indicated by the R2. It is seen that the terms of trade thus measured have deteriorated for cereals and cotton at a rate of about 0.5-0.7 percent per annum. That implies a decline by a factor of 1/2 or more over that period. True, the fluctuations are quite wide and hence the trend accounts only for a small fraction of the total variations. Yet it is statistically significant. What is more important is that, even if there were no trend, during this period agricultural output increased very dramatically. Thus, any analysis that would relate output to prices would conclude that either the supply is negatively sloped or, if no trend were observed, that it is perfectly elastic. To explain the long-run supply, it is necessary to explain the change in the shifters themselves. This is not done here empirically. However, we can use the foregoing framework to comment on the process. With some qualifications, equation (15) can serve to link between the short and the long run. The qualifications are: (1) the equation only applies to the private inputs; (2) it is obtained within a comparative statics framework. Since comparative statics analysis is timeless whereas observations are dated, there is a need to link between the two. - 48 - Table 9: TIME TREND OF AGRICULTURAL PRICES DEFLATED BY US WHOLESALE PRICES 1990-1983 b t R2 Coffee +0.01 6.8 0.36 Cocoa +0.003 1.4 0.02 Tea -0.0006 0.5 0.003 Sugar -0.007 3.7 0.14 Wheat -0.007 7.2 0.39 Maize -0.006 6.1 0.31 Rice -0.006 5.5 0.27 Cotton -0.005 5.4 0.26 Wool -0.001 8.9 0.49 Rubber -0.03 13.0 0.67 Source: EPDCS - 49 - Changes in aggregate supply involve changes in resources. Much of the increase in agricultural output was obtained by expansion of capital inputs. In relating such an expansion to prices it should be noted that the total investment in the economy is bounded by overall savings. Therefore an increase in agricultural investment takes place at the expense of investment in non- agriculture and as such it depends on the differential rate of return between agriculture and non-agriculture, as well as on other variables. This process is not revealed by comparative statics analysis. To capture it, a different analysis is required as illustrated in Mundlak (1979) and Cavallo and Mundlak (1982). A change in agricultural prices affects the rate of returns in agriculture and thereby its share in total investment. The essence of this process is that an increase in the pace of agricultural investment has an increasing cost due to the increase in the shadow price of capital goods. In terms of (15) that can be expressed in terms of low values for the factor supply elasticities (sj) related to capital inputs. But it should be noted that within this framework sj is a function that depends on the pace of agricultural investment which, by itself, is endogenous within the economic process. Also, the level of employment in agriculture cannot be explained within the framework of comparative statics. Since agriculture is a declining industry, it generates surplus labor as reflected by lower returns to labor in agriculture than in non-agriculture. As a result, a dynamic process of off- farm migration determines the labor supply in agriculture (Mundlak 1979). The time rate of migration depends, among other things, on the magnitude of the intersectoral wage differential. This is the major channel through which changes in agricultural prices affect labor supply. Again, putting it in terms - 50 - of elasticity of labor supply as required by (15), the value of such elasticity will depend on the length of period under consideration, and in any case will be endogenous. All this leads to the conclusion that using (15), the move to long- run elasticities cannot be done by innocently assuming all factor supply elasticities to become very large, and consequently obtain a large product supply elasticity. The above discussion dealt with the long-run response arising from the supply of private inputs. The empirical analysis indicated that the public inputs have had a very substantial effect on supply. The levels of such public inputs are not determined within the same framework which applies to the private inputs. We do not have, at the present, a framework to explain the level of public inputs and it is therefore taken to be exogenous in this analysis. Note, however, that if indeed countries follow inconsistent policies of applying public inputs in agriculture to compensate for discriminating price policies toward agriculture, it is possible to generate negatize supply response from cross-country analysis, as was the case in our analysis. This can be derived immediately by a proper interpretation of (6). Finally, agricultural output has historically been strongly affected by technical change. Recall that our empirical findings, once accounc was taken of shifter variables, detected no time effect on supply. Also GDP, our measure of comprehensive capital, performed empirically as well as a time trend. This is not a coincidence. As explained in Mundlak (1984b), the process of technical change is closely related to that of capital accumulation. This process has two major aspects--the invention of new techniques and their implementation. Both aspects, but especially implementation, depend to a large degree on the pace of capital accumulation. - 51 - To conclude the discussion, the study of long-run supply response requires a substantially more detailed investigation of private agricultural investment and migration. It also requires an analysis of the process by which governments allocate resources to the shifter variables which appear in the single-equation supply function of the sort analyzed above. The simple aggregate supply functions, when estimated from the within-country variations, produce very low short-run elasticities with respect to price. The short-run input demand elasticities with respect to output and own-price are higher than the aggregate output supply and these results are consistent with conventional theory. - 52 - APPENDIX 1 DETAILED DATA DESCRIPTION Two sets of indexes were computed and are available in the data base. Most estimation was, however, done using the Multilateral Fisher index. Wheat Equivalent Index (1) Quantity Index: In order to aggregate agricultural production of many diverse commodities for each country most previous studies have used a wheat equivalent unit. The average world price of each commodity over the period 1961-81 was calculated by using the world export unit value weighted by world production of each commodity. The world wheat equivalent price was derived as the ratio between the world average price of each commodity to the world average price of wheat. The quantity level of each country is then the summation of the products of the production and the wheat equivalent price of each commodity. The above computation can be summarized as follows: t iwt iwt P = iw z y t iwt P. 1W lw Y =; E. * y rt I l irt Where: Yrt wheat equivalent quan-tity level of country r in year t. II = world wheat-equivalent price of commodity i. Yir.t - production of commodity i in country r in year t - 53 - P. average world price of commodity i P1w average world price of wheat (2) Price Index: Each country's implicit domestic currency price index was calculated by dividing total production value at farm level by the total wheat equivalent quantity level. This can be expressed mathematically as follows: p = irt Yirt rt Y Where: Prt = domestic currency price index in country r in year t Pirt producer prices of commodity i in country r in year t. We can also define an explicit price index as E _ £ irt * i Yirt _ £irt rt i E z Yirt . a E. r irt Where Xirt is the proportion of commodity i in country r's output in year t valued at wheat equivalent units. By inspection, it is seen p = p E rt rt i.e., the implicit and explicit price indices are identical. Note: In order to be consistent with production and producer prices, the export unit value of rice was converted to that of paddy. The conversion factor used was 65%. Export unit values of shelled groundnuts was also converted to grourinuts-in-shell. The conversion factor used was 71%. Where producer prices were available only for sugarcane or sugar beet, these prices were converted to raw sugar prices as follows: sugarcane prices were multiplied by a factor of 10, and sugar beet prices were multiplied by a factor of 7.7. - 54 - Fisher Multilateral Quantity and Price Index The base prices and quantities were calculated by averaging world prices and production over time. y. y. /TN Yio t iwt E Yiwt Piwt 10 Yiwt t Where N is the number of countries and T is the number of years, i.e., Yio and Pio can be viewed as the annual output levels of the "average" country and the average world price levels, respectively. The number of countries covered was 60 and the period covered was 1967-78. The Fisher multilateral price index was calculated by using the following formula: Z E pF = ( . Pirt * Yio* iPirt *Yirt rt Pio Yio ePio Yirt 10 1 10 r P L * pp ) ½ Where pL and Pp are equal to "Laspeyres" and "Paasche" formulae, respectively. The quantity index was calculated by deflating total production value at farm level by the Fisher price index in each country. The formula used was F i Pirt irt Y- rt PF rt F where Y Fisher quantity index in country r in year t. rt - 55 - Estimates for Agroclimatic Variables The agroclimatic variables MPDM (potential dry matter production without irrigation) and FWD (reduction factor caused by water deficiency) for each country were estimated from Tables 1-7 in Buringh et al. The MPDM for each country is the summation of potential production from each identified soil class. Mathematically, for country i MPDM. = (P. * MPDM.) 1 J ij where Pij = proportion of soil class j falling into country i, estimated by overimposing maps with country boundaries onto soil maps from Buringh et al. FWD for each country is a weighted average of FWD in each soil class. Mathematically, for country i X (A. * P. * FWD.) FW J j 1J J FW =i (A. * P) j 1J where FWD = FWD in soil class j, from Buringh et al. Ai area of each soil class. Calculation of Coefficients of Variation (CV) CV of prices is calculated by the following procedures: 1. Use 1967-72 CV of prices as instability measures for the 1967-72 period. 2. Compute for each successive year from 1973 as the movinL coefficient of variation, i.e. - 56 - CVlt = coefficient of variation of the year Plt-1' Pil t-2' Pil t-31 Pil t-4, PIl t-5 PIl t6 For example: CV173 = cv(PIl72, P1171, P1170, P1169, P21163d P167) CV174 = cV(PI173,PI172, P1171, P1170, P1169, P168. Data Base We have set out below the variables put into the data base for the study and given the source of the data. DATA DESCRIPTION AND SOURCE ------------------------------------------------------------------__---------_ NOTATION DESCRIPTION SOURCE REMARK AE1 Agricultural extension Evenson/Judd, millions of expenditure Yale University 1980 dollars EWI Agricultural extension Evenson/Judd workers Yale University ALl Adult literacy World Bank EPD Data Bank CV1 Coefficient of Variation World Bank calculated crop price index EPDCS (wheat equivalent) CV2 Coefficient of Variation World Bank calculated of livestock price EPDCS (wheat equivalent) CV3 Coeffici.ent of Variation World Bank calculated of fertilizer price EPDCS (in US$) CV7 Coefficient of Variation World Bank calculated of total price index EPDCS (wheat equivalent) - 57 - CV8 Coefficient of Variation World Bank calculated of crop price index EPDCS (Fisher) CV9 Coefficient of Variation World Bank calculated of livestock price index EPDCS (Fisher) CVIO Coefficient of Variation World Bank of total price index EPDCS calculated (Fisher) ERI Purchasing power parity World Bank local currency/ exchange rate EPD Data Bank US$ FC4 Fertilizer consumption FAO tape in tons FDl FWD-reduction factor Buringh, Van estimated from caused by water Heemst & Staring, the data in deficiency "Comparison of the Buringh's paper Absolute Maximum Food Production of the World", 1975 MMI MPDM - potential dry Buringh et al. estimated from matter production the data in without irrigation Buringh's paper. FII Crop price index World Bank calculated (Fisher index) EPDCS FI2 Livestock price index World Bank calculated (Fisher index) EPDCS FI3 Total price index World Bank calculated (Fisher index) EPDCS FP4 Fertilizer prices FAO tape US$/ton. Some estimates by EPDCS FP5 Fertilizer prices in FAO tape local currency/ local currency ton. Converted from FAO data, using official exchange rate GYI GDP in local currency World Bank EPD in millions Data Bank local currency - 58 - GY2 GDP in agriculture World Bank EPD in millions Data Bank local currency GY3 Agricultural share World Bank EPD in % in total GDP Data Bank IR1 Irrigation FAO tape in hectares LD1 Arable land FAO tape in hectares LD2 Land under permanent FAO tape in hectares crops LD3 Meadow and pasture FAO tape in hectares land LD4 Forest and woodland FAO tape in hectares LD5 Other land FAO tape in hectares LD6 Cultivated land World Bank calculated from EPDCS FAO data, in hectares. LFI Life expectancy World Bank EPD in years Data Bank PDI Rural population World Bank calculated density EPDCS population/ hectare PI1 Wheat equivalent crop World Bank calculated price index EPDCS local currency/ton PI2 Wheat equivalent World Bank calculated livestock price EPDCS local index currency/ton PI3 Wheat equivalent total World Bank calculated price index EPDCS local currency/ton PO1 Total population FAO tape P02 Agricultural FAO tape Derived from population FAO ratio P03 Total labor World Bank EPD force Data Bank P04 Agricultural labor World Bank EPD force Data Bank 5 59 - P05 Rural population World Bank EPD Data Bank RDI Road density International Road in kms/sq. km Federation, World Road Statistics PV1 Percent of International Road in % road paved Federation, World Road Statistics QF1 Fisher index of World Bank calculated crop output EPDCS QF2 Fisher index of World Bank calculated livestock output EPDCS QF3 Fisher index of World Bank calculated total output EPDCS QL1 Crop output in World Bank calculated wheat units EPDCS in tons QL2 Livestock output World Bank calculated in wheat units EPDCS in tons QL3 Total output in World Bank calculated wheat units EPDCS in tons SRESEXP Research stock in US$ Everson/Judd, in million Yale University 1980 dollars SSMY Stock of scientists Everson/Judd, man-years Yale University TRl Number of tractors FAO tape WG1 Agricultural wages International Labor Some estimates Office, Yearbook by EPDCS, local of Labor Statistics currency/month WG2 Urban wages ILO, Yearbook Some estimates of Labor Statistics by EPDCS, local currency/month Estimation of Missing Values In this section we describe how the missing values were estimated. This information is given in detail so that others using the data are fully aware of its derivation. - 60 - Wages. Data on agricultural and non-agricultural wages were collected from the International Labor Office, Year Book of Labor Statistics (various issues). There were four different time units, i.e., local currency per hour, day, week or month. All of them were converted to the same time unit--local currency per month by the following methods: L. Use hours of work per month in the non-agricultural sector if the data were available; 2. Otherwise, use the following conversion factor 1 day 8 hours 1 week = 5 days 1 month = 4.3 weeks If wage data were available for male, female, and total employment, the total value was used. If only male and female data were available, male data were used. If data were classified into skilled and non-skilled labor, non-skilled data were used. The following observations were missing: Non-agricultural wages: Bangladesh 1968-69 Brazil 1977-78 Burundi 1968-72 Costa Rica 1968-72 _Egypt 1978 Malaysia 1968-70 Nigeria 1968-71 Portugal 1969-70 Syria 1978 Thailand 1972-78 Zimbabwe- 1968-71 Agricultural wages: Argentina 1978 Australia 1968-78 Bangladesh 1968-71 - 61 - Brazil 1968-78 Burundi 1968-72, 1976 Cameroon 1975-78 Colombia 1971-75 Ecuador 1968-78 Egypt 1978 El Salvador 1968-69 France 1978 Greece 1968-78 Guatemala 1968-78 Nigeria 1968-71, 1976-77 Panama 1968-78 Peru 1968-78 South Africa 1968-78 Switzerland 1968-70 Syria 1968 and 1978 Thailand 1968-78 Trinidad and Tobago 1968-78 Venezuela 1968-78 Regression analysis showed that there is a close relationship between non-agricultural wages (WG2) and the consumer price index (CPI). Therefore, the domestic consumer price index was used to estimate the missing observa- tions for non-agricultural wages, except for Portugal where agricultural wages (WGI) were used. The regression results and data periods were as follows: Bangladesh 1970-80 WG2 = 41.59 + 2.98 * CPI R2 = 0.86 (1.7) (7.5) Brazil 1962-76 WG2 = -160.97 + 161.09 * CPI R2 = 0.99 (-4.8) (34.9) Burundi 1973-80 WG2 = 1487.21 + 83.19 * CPI R2 = 0.94 (2.8) (10.1) Costa Rica 1973-80 WG2 = -457.80 + 28.76 * CPI R2 = 0.95 (-2.4) (11.0) - 62 - Egypt 1961-77 WG2 = -7.38 + 0.60 * CPI R2 = 0.98 (-8.2) (28.9) Malaysia 1971-78 WG2 = -72.50 + 5.26 * CPI R2 = 0.93 (-1.6) (8.6) Nigeria 1972-80 WG2 = 10.08 + 0.53 * CPI R2 = 0.89 (2.2) (7.5) Syria 1961-77 WG2 = 86.16 + 2.25 * CPI R2 = 0.77 (6.2) (7.1) Thailand 1961-71 WG2 = 44.71 + 15.05 * CPI R2 = 0.30 (0.15) (1.9) Zimbabwe 1972-81 WG2 = -6875.46 + 427.33 * CPI R2 0.98 (-4.2) (20.3) Portugal 1971-80 WG2 = -1575.73 + 2.10 * WG1 R2 = 0.99 (-6.7) (39.1) Missing data for agricultural wages (WG1) were much more extensive that for non-agricultural wages. In some cases no data were available. The data gaps were filled in by the following procedures. 1. Where agricultural wages were not available, using countries in the same region the following regression was run-,- WR =f~ (xl, x2, x3, x4, x5, x6 where WR = WG1 - 63 - Xi = Agricultural income per capita Non-agricultural income per capita X2 ALl X3 = (LD1 + LD2)/QLI X4 = TRl/QLl X5 = ARl/QLl X6 = AEl/QLl WGI WR * WG2 For African countries with missing observations, data for Kenya, Tanzania and Zambia were used giving the following regression result: WR = 0.40 + 0.22 xi + 0.001 x2 + 0.04 X3 (12.1) (1.9) (1.2) (2.5) -7.17 x4 - 0.99 x5 -33.66 x6 (-0.3) (-0.3) (-1.9) R2 = 0.69 For Latin American countries with missing values, data for Argentina, Mexico, Costa Rica and El Salvador were used, giving the result, WR = 0.39 + 0.34 xl - 0.004 x2 - 0.39 X3 (4.6) (6.9) (-3.2) (-3.5) + 174.69 x4 + 12.90 x5 + 103.24 x6 (5.6) (0.4) (1.6) R2 = 0.77 For Thailand, data for Malaysia and the Philippines were used. The estimated regression was, - 64 - WR = 0.42 - 0.18 xi + 0.007 x2 + 0.09 X3 (2.6) (-1.0) (2.9) (0.4) + 337.89 x + 29.05 x - 121 79 x (1.7) (0.4) (5.2) R2 = 0.98 For Greece, data for Israel and Turkey were used. The estimated regression was, WR = 2.06 + 3.44 xl - 0.02 x2 - 2.52 X3 (2.7) (3.0) (-1.6) (-3.0) - 180.42 X4 + 273.36 X5 + 1438.73 x6 (-2.9) (2.6) (5.3) R2 = 0.97 2. Where a few years' data only were not available, the relationship between agricultural wages and non- agricultural wages within the country was used to estimate missing observations. Regression results and the period covered were as follows: Argentina 1961-77 WG1 = -6.27 + 0.80 * WG2 (-0.4) (350.3) R2 1.0 Colombia 1961-70 WGI = 33.88 + 0.36 * WG2 -(2.3) (17.8) R2= 0.98 El Salvador 1970-80 WGI = 14.68 + 0.25 * WG2 (1.8) (8.0) R' -0.88 - 65 - France 1961-77 WG1 = -323.89 + 0.84 *WG2 (-4.6) (16.3) R2= 0.95 Syria 1968-77 WGl = -10.0 + 0.65 * WG2 (-0.6) (7.4) R2= 0.89 Switzerland 1971-81 WG1 = -224.25 + 1.08 * WG2 (-2.2) (19.8) R2= 0.98 3. In the case of Bangladesh the relationship between agricultural wages and the consumer price index was used. WG1 = 24.11 + 3.12 * CPI (1.2) (11.0) R2= 0.94 4. In the case of Australia the ratio between agricultural wages and non-agricultural wages in New Zealand was applied to estimate agricultural wages in Australia. Producer Prices. The missing producer price observations are presented below. The missing observations were estimated by one of the following procedures. 1. On the basis of the producer price of a similar commodity in the same country. For example, producer prices of soybeans in Pakistan for 1967 and 1968 were estimated by means of the relationships between producer prices of - 66 - lentils and soybeans in Pakistan. A regression for the period 1969-78 was run to establish the relationship between the two price series and the relationship was used to estimate the missing years. 2. On the basis of producer prices of the same commodity in a similar, neighboring country. For example, producer prices of sugar in Uruguay for 1967 and 1968 were estimated via the relationship between sugar prices in Uruguay and Argentina for the period 1969-78. 3. On the basis of the relationship between domestic prices and world export unit values of the same commodity. For example, producer prices of tea in Korea for 1967 and 1968 were estimated by means of the estimated regression between Korean tea prices and world tea export unit values for the period 1969-78. If the whole series were missing, the ratios between international prices of similar commodities were used to estimate the missing series. For example, jute prices in Peru were estimated by means of the price ratio between jute and cotton in the international market. - 67 - PRODUCER PRICES - MISSING DATA, 1967-78 ---------------------------------------------------------------__------------_ COUNTRY COMMODITY YEAR --------------------------------------------------------------__-------------_ AUSTRALIA Sugar, raw 1967 & 1968 Broad beans, dry 1967 & 1968 BURWNDI Wheat 1978 Rice 1978 Maize 1978 Sorghum 1978 Potatoes 1978 Beans, dry 1978 Peas, dry 1978 Cottonseed 1978 Cotton lint 1978 Tobacco 1978 Beef & veal 1978 Cow milk, whole fresh 1978 Mutton & Lamb 1978 Coat meat 1978 Indigenous pigment 1978 Hen eggs 1978 Honey 1978 Poultry meat 1978 Millet 1978 Sweet potatoes 1978 Cassava 1978 Groundnuts in shell 1978 Castor beans 1978 Bananas 1978 Tea 1978 Coffee 1978 Palm kernels 1978 Palm oil 1978 CAMEROON Honey 1967 & 1968 CANADA Sugar, raw 1967 & 1968 CHILE Soybeans 1967 & 1968 Jute 1967 & 1968 COLOMB3IA Oats 1967 & 1968 Chick-peas 1967 & 1968 Lentils 1967 & 1968 Oranges 1967 & 1968 Honey 1967 & 1968 - 68 - ECUADOR Grapefruits 1967 & 1968 ECUADOR Honey 1967 & 1968 EGYPT Citrus fruits, n.e.s. 1967 & 1968 EGYPT ~~~~~~Jute 1967 &1968 EL SALVADOR Lemons 1967 & 1968 Sweet potatoes 1967 & 1968 Castor beans 1967-68 FINLAND Sugar, raw 1'967 & 1968 FRANCE Sugar, raw 1967 & 1968 GERMANY, F.R. Sugar, raw 1967 & 1968 GUATEMALA Sugar, raw 1967 & 1968 ISRAEL Tangerines 1967 & 1968 JAPAN Jute 1967 & 1968 KENYA Plantains 1978 KOREA Tea 1967 & 1968 MALAWI Chick-peas 1967 & 1968 Rapeseed 1967-7896 NEW ZEALAND Jute 1967-78 PAKISTAN Sunflower seed 1967 & 1968 Soybeans 1967 & 1968 PERU Jute 1967-78 PHILIPPINES ~~~Beans, dry 1978 PHILIPPINES Cottonseed 1978 Onions, dry 1978 Oranges 17 Tangerines 1978 Lemons 1978 Grapefruits 1978 Cotton lint 1978 Beef & Veal 1978 Cow milk, whole, fresh 1978 Mutton & lamb 1978 Pigmeat 1978 Hen egg9s 1978 Horsemeat 1978 Poultry 1978 - 69 - Soybeans 1978 Castor beans 1978 Bananas 1978 Pineapples 1978 Plantains 1978 Coffee 1978 Rubber 1978 ZIMBABWE Tomatoes 1967 & 1968 Tangerines 1967 & 1968 Cassava 1967 & 1968 Coffee 1967 & 1968 SPAIN Sugar, raw 1967 & 1968 Coffee 1967 & 1968 SYRIA Maize 1978 Sorghum 1967 & 1968 Potatoes 1978 Beans, dry 1978 Broad beans, dry 1978 Peas, dry 1978 Chick peas 1978 Tomatoes 1978 Onions, dry 1978 Oranges 1978 Lemons 1978 Apples 1978 Pears 1978 Peaches 1978 Grapes 1978 Tobacco 1978 Beef & Veal 1978 Cow milk, whole, fresh 1978 Mutton & lamb 1978 Coat meat 1978 Hen eggs 1978 Honey 1978 Poultry meat 1978 Bananas 1978 Citrus fruit, n.e.s. 1967-68 & 1978 UK Horsemeat 1967-78 Sugar, raw 1967 & 1968 URUGUAY Sugar, raw 1967 & 1968 ZAMBIA Tea 1967 & 1978 - 70 - Fertilizer Prices. Fertilizer prices were estimated by taking the weighted average of the three fertilizer groups--nitrogenous, phosphate and potash fertilizers. Consumption of each fertilizer group was used as the weights. Prices of each group were estimated by the same procedures as described above for producer prices. For countries for which price data were available but not consumption, the weights were borrowed from countries at similar income and agricultural development levels. These countries are as follows: Australia - New Zealand Cameroon - A group of African countries Canada - USA Costa Rica - Mexico El Salvador - Ecuador France - Spain and the Fed. Rep. of Germany Greece - Israel Guatemala - Ecuador Malaysia - Thailand Nigeria - A group of African Countries Panama - Ecuador Zimbabwe - A group of African countries Trinidad and Tobago - Mauritius United Kingdom - Ireland Yugoslavia - Austria Zambia - A group of African countries - 71 - The missing data for fertilizer prices for the period 1968-78 were as follows: Argentina: 1968-78 Australia: 1968-78 Bangladesh: 1970 Brazil: 1968-73 Burundi: 1968-69 Cameroon: 1968-78 Canada: 1968-78 Sri Lanka: 1968-76 Chile: 1968-71 Colombia: 1968-74, 1977 Costa Rica 1968-78 Cyprus: 1975-77 Ecuador: 1968-71 El Salvador: 1968-78 France: 1968-76 Greece: 1968-78 Guatemala: 1968-78 India: 1968-71 Ireland: 1977 Kenya: 1968-71, 1974, 1978 Malawi: 1968 Malaysia: 1968-77 Nigeria: 1968-78 Panama: 1968-78 Peru: 1970 Philippines: 1968-72, 1977-78 Portugal: 1968-75 South Africa: 1968-73 Tanzania: 1968-71, 1974-75, 1978 Thailand: 1968-69, 1972, 1976-78 Trinidad and Tobago 1968-74, 1976-78 Turkey: 1970-72, 1977-78 U.K. 1968-76 Uruguay: 1968, 1971-72, 1974 Venezuela: 1972-74, 1976-78 Yugoslavia: 1968-78 Zambia: 1968-78 Zimbabwe: 1968-78 - 72 - The missing data, for fertilizer prices were estimated by the following procedures. 1. Pick three straight fertilizers in each nutrient group for which data are available. 2. Fill gaps for consumption and prices in each series. Consumption of a straight fertilizer was regressed on consumption of its nutrient group and international or domestic prices. For example: Qurea a + b * QN + c * PN where Qurea = Consumption of urea QN Consumption of nitrogenous fertilizers PN = International or domestic price of N Price gaps were filled in a similar fashion. For example: Purea =a + b * PN + c * PS where Purea = Price of urea PN = International price of N ps = Domestic prices of a nutrient of the same group 3. Calculate weighted average price of each nutrient group. 4. Ext-rtapolate the price of -.each nutrient group to remaining year-sby a regression on: (a) internacionai prices (b) straight nutrient price for a similar country. - 73 - 5. Use consumption of each nutrient group as weights to calculate weighted average price of the overall fertilizer as ,P. * C. 11 1 P = F C. where Pi = price of each nutrient group Ci = consumption of each nutrient group The international prices of fertilizer and the world weighted average prices are shown below. - 74 - INTERNATIONAL FERTILIZER PRICES …----------------------------------------------------------__----------------_ N P K Weighted Average /A …( --------------(US$/TON)---- 1966 194.1 249.3 55.2 162.9 1967 172.4 234.4 51.2 148.6 1968 142.4 189.5 48.2 123.8 1969 121.4 194.5 44.2 112.8 1970 105.0 214.5 62.9 111.6 1971 100.0 214.5 65.3 109.3 1972 128.9 339.0 67.2 146.6 1973 206.1 498.6 85.3 220.1 1974 686.5 1,515.8 121.5 663.1 1975 430.4 1,007.1 163.3 449.2 1976 243.5 453.7 110.5 241.2 1977 277.0 483.7 102.4 265.9 1978 314.8 488.5 113.3 293.8 1979 375.9 728.0 154.0 380.3 1980 482.8 897.5 232.8 488.7 /A Weighted by world consumption of fertilizers. Source: World Bank, EPDCS - 75 - APPENDIX 2 WHAT ACCOUNTS FOR THE HIGH SUPPLY ELASTICITY ESTIMATES BY PETERSON In Appendix Table 2.1 we set out the relationships between the variables used by Peterson (P) and those used by us (WB). The only set of observations for which this comparison can be made is the 1968-70 period average for those 36 countries represented in both data sets. Simple cross- country regressions with intercept were run between the logarithms of the variables in column (1) and those in column (2). The regression coefficient in column (3) should be close to 1 if the two variables correspond closely; also, t-values (column 4) and R2 (column 5) should be high. As can be seen, the wheat equivalent (WE) indexes of the two studies are very closeLy related, with a regression coefficient of 0.93 and an R2 of 0.92. The WB Fisher output index is also closely related to the WB wheat equivalent output index, with coefficients and R2 very close to 1. Therefore, the relationship of Peterson's WE output index and the WB Fisher index is also reasonably close. Differences in output measurement should therefore not be expected to cause the discrepancy in results. The correspondence between Peterson's research variable and the WB research variable is also fairly good. Both variables were originally compiled by Evenson and his collaborators. Peterson used the total number of publications per ha, while we have used research costs per ha of agricultural land. Peterson's climate variable is simply the long-run average precipitation. Our agroclimate variable is a measure of water availability 1/ 1/ Called "Factor of Water Deficit" by Buringh, et al. - 76 - computed from the maps and tables of Buringh, et al. It is computed as rainfall, adjusted for evapotranspiration demand of the particular climate, i.e., less rainfall is required in cold climates than in hot ones to achieve the same water-availability effect. As can be seen, these two variables are only loosely correlated. The biggest differences are in the prices. Here there is no correlation at all. The raw output price index and the raw fertilizer price index are only slightly better correlated, but that is probably because they are in local currency units, i.e., there is substantial variation in these series which simply reflects the differences in exchange rates across countries. We discuss the source of the discrepancies in the price series below. In Table 2.2 we compare results of Peterson-type cross-country supply functions obtained with the different data series. Columns 1 to 4 indicate the source of each variable. Columns 5 and 6 report the price coefficient and its t ratio, and column 7 the R2. Other coefficient results are not reported. In run 1 we reproduce Peterson's supply functions, with and without the inclusion of the research variable. Because we can use only 36 countries and one of his cross sections, these results do not exactly match Peterson's published results. Nevertheless, if research is omitted, we arrive at a supply elasticity of nearly 2 which is highly significant. (With research included the supply elasticity is still about 1.4.) When we replace Peterson's precipitation variable by our water availability variable (run 2), the price coefficient drops from3 1.94 to 1.70. Thus, the agroclimate variable does not account for the difference. However, when in run 3 we use both our price and precipitation variable, the price coefficient drops to zero. - 7 7 - In run 4 we start from a regression with all our variables, and find an elasticity of less than 0.2. When we replace our price series by Peterson's prices (run 5), the elasticity jumps to 1.53. And when we use our output variable and Peterson's price and precipitation 7ariables (run 6) we obtain a supply elasticity of 1.98. Finally, in Table 2.3 we explore whether the choice of countries and years influence the results. The first two supply equations use WB data. The first regression used the means of 11 years for 59 countries and finds a supply elasticity of 0.45. When going to the 1968-70 average and 36 countries, that elasticity drops to 0.17. While this is a considerable drop, the change in data points does not account for the huge difference compared to Peterson's findings. The elimination of 18 countries from Peterson's data set (row 4 to row 3) hardly affects the supply elasticity. It also appears from Table 2.3 that water availability is a more powerful explanatory variable than precipitation. Its coefficient is about three times larger and (for comparable degrees of freedom) has a much larger t value. To summarize, the discrepancy in the findings between Peterson's study and ours is attributable almost exclusively to differences in the price variables. 1/ Output variables correspond closely. Discrepancies in research and climate variable do not affect the price coefficients substantially. Nor is the difference associated with the many shifter variables Peterson left out. 1/ Chhibber's study also extended Peterson's specification to include variables representing differences between countries; these new specifications resulted in a lowering of the own-price supply elasticity to about 1.0. - 78 - It is sufficient to substitute our price variable into a regression with Peterson's output variable to generate a very low supply elasticity. Conversely, use of Peterson's price variable with our output variable results in the high elasticity found by Peterson. Comparison between WB price data and Peterson's data Of the countries common to the two data sets, there are 11 countries which show serious differences between the WB wheat equivalent price index and Peterson's index. An actual comparison of this data can be seen in Table 2.4. We have calculated the wheat equivalent price index manually for these 11 countries and the result is the same as the WB computer printouts. We have further examined the price data used by Peterson and by us and have found that there are major differences in the producer prices used. Peterson used the FAO 1975 publication, "Agriculture Producer Prices" whereas we have used the FAQ 1982 producer price tapes. There are three major differences between these two data sources: 1. While output is measured in metric tons in the 1982 tapes there is no uniform unit in the FAQ 1975 price publication. Prices are often given in odd units such as per 3.75 kg, per 10 eggs, per 100 liters. Units for the same commodity vary among countries. The conversion of these prices to metric tons could create serious accuracy problems, particularly in the livestock sector where "head" or "live weight" rather than "meat weight" is used for producer prices. The 1982 producer price tape has the same unit as the production tape, i.e., the FAQ has done all the conversions before publication. - 79 - 2. The 1982 producer price tapes cover more commodities than the 1975 publication. 3. Some prices are different. An example is given below for Ireland's producer prices for 1969 (Irish pounds/ton): 1975 Publication 1982 Tape Wheat 31.7 31.7 Barley 25.3 25.3 Oats 22.6 22.6 Potatoes 26.6 19.8 Sugar, raw 53.0 53.0 Beef 357.6 340.1 Pig meat 412.0 269.6 Milk 25.3 25.3 Eggs 267.2 235.2 Wool, greasy 366.1 303.1 The wheat equivalent price index for Ireland for the period 1968-70 was calculated using producer prices in the FAO 1975 publication and applying our method. The result was 2.'01/10'0 kg. Peterson's index was 1.88/100 kg whereas our index- (using 1982 tape data) was 2.16/100 kg. Further, the fertilizer price for- Ireland for the period 1968-70 was 4.34 pounds/100 kg whereas Peterson's price was 5.23 pounds/100 kg. Table 2.1: RELATIONSHIlPS BETWEEN VARIABLES IN PETERSON'S DATA SET AND WORLD BANK DATA SET - 1968-70 AVERAGES First Variable Second Variable Reg. Coeff. t R2 (1) (2) (3) (4) (5) Outputs P WE index (TAO69L) WB WE index (QL3LD8) .93 21.1 .92 P WE index WB Fisher index (QF3LD8) .80 16.19 .88 WB Fisher index WB WE index 1.06 34.22 .97 Prices P WE price ratio (IPR69L) WB W4E price ratio (PR1L) .40 3.84 .29 P WE price in (WEP69L) WB WE price in local local currency currency (RAA1) .26 2.27 .13 P fertilizer price in WB fertilizer price in local currency (FERT69L) local currency (FPL) .28 2.27 .13 WB WE price ratio WB Fisher price ratio (PR2L) .96 9.60 .72 Shifts Precipitation (PRC69L) Factor of water availability (FD1) .56 2.25 .12 Publications (TRE69L) Scientist man years (SSMYL) 1.02 9.14 .69 Notes: Abbreviations in brackets give name of variables in the merged data set. All variables, except for FD1 were used in their logarithms. Other abbreviations: P = Peterson's data set, WB = World Bank data set, WE - Wheat Equivalent. Table 2.2: RECONCILIATION OF PETERSON AND WORLD BANK FINDINGS Source Variables (1) (2) (3) (4) (5) (6) (72 Run Dependent Price Agroclimate Research Price coeff. t value R (a) 38 Common Countries, 1968/70 average 1 P P P P 1.39 3.72 .58 P P P 1.94 5.69 .58 2 P P WB 1.70 4.51 .47 3 P WB WB .03 .08 .15 4 W,B WB WB WB .14 .96 .86, WB WB WB .17 .52 .28 5 WB P WB 1.53 4.25 .53 6 WB P P 1.98 5.74 .50 (b) 59 Countries, 11 years of data 7 WBM WBM WBM WBM .21 4.59 .86 8 WBD WBD WBD .02 1.49 .57 9 WBD WBD .02 .83 0 10 WB WBM WBM .45 4.38 .29 WBD .02 .09 .71 Abbreviations: P Peterson, WB = World Bank WJBM = Country meanis of main World Bank data set WBD = CouL1try deviations of main World Bank data set Table 2.3: PETERSON'S SUPPLY FUNCTION USING DIFFERENT DATA SETS AND VARIABLES Price Term Agroclimate Term R2 Main WB data set WB variables .45 1.65 .30 59 countries, mean (4.41) (10.26) of 11 years V Merged data set WB variables .17 1.81 .28 38 countries, (.52) (3.27) 68/70 average Peterson variables 1.94 .44 .49 1 (5.69) (1.61) c Peterson data set Peterson variables 1.81 .55 .59 54 countries, (8.47) (3.11) 68/70 average Notes: Dependent variable = Wheat equivalent indices of output (P: TA069L, WB: QL3LD8) Price variable = Wheat equivalent price - divided by fertilizer price (P: IPR69L, WB: PRIL) Agroclimate variable = P: Average annual precipitation PRC69L WB: Water availability FD1 t - values in parenthesis ~~2i' ~~~~ '.'c~~ ' - 1 , i'-'i, 1,r'' :' q;6 T CI7. 0 ' 617 I :c 0 9 I Io .IST ('- t 13/q S I 16 'i-i1 C/..1 I Ir19n>iln 1" 00~~ II ~ ' rj91 '.I,- I T 889 1 Wo T 7"'1 tj/r /;'' clC r 'l0 14 T I 3 £6'9 0 I 1 17 CI It USn Q Z' : 1: K* 61-1.' C 82 176C I-'1 1, ? 0 c)' I.1 1 o', i'TC$'U , R -6 7 wrig,m 1 1' 0Iin * - ''~~~~ - tic' 1 ('1,~~~" '£ '2 1 '' C El,I? I C :-(- 9C I C-0 C' 6Z 0II`C 0 7'1 T ': 9.1 1 o C 9 tN( f1M 'zt: , C- I ': ~ .i '1( I -,I 'l/,'22( 4o 19/.`F 6061 !'2 17 C I'T t'2 'IC 67U N i I; I*"'*/I ~~~17 /'2t' (Ir I8C I7T2~O I2/7/.'IIi - £'2 P(76147/i0 C1 0 I'6'0T C/ 9T TCI *OC 817 NPU I f '~ ~T 1117' AA' 69 ~i?(,t' /~ 6~917 .2 178 _ TLI, *. ,7 r' '::C1 9 /2* C' `C C6IA LYDJ4r I ''.;:n oo - i~~~~~/o'A 00 *C i~~ 08'' '':'., 8C R9CA T(''2;( £F3uRCO . 17'.726 I'' 13'2 f' * 261' 0 '11 6C I i - *217 ~~~~;'~~''~~ CC9 01; *CC~~~I- 0:00(, '0 01 1(3 Cl t 0 cC'? 10 '0 `/9C9: 'R 6'6 rL I I77 617£ I /.L *'? -' I 1I~U TA 01'IL 82 r21 QC1 'CC - I I 9 0 PC '26 ':': 1 ? 17711, 'C' T0:C2 '0 CT. OC l '6 '0 I Pur 'I Z i" RI? I NUI! HJS '"17' QC - I'2 17'2.T. I 0''~~~~~(Wl 69C l If *,L,0 IC 61179 191- t I9',_I 7 '0 Cr21 01) L'.7 C- fC I C2''2CC '9 9 6 'iv CIAm .10N7 ~~ ~~ ':'i~~~ - v: II. I (' I 'N 0 6 *0, I l tC 9101 ("2/f,-I Z:, / tRhI:' 17';o'2 1Z;CI? 0 17 C " I 7/'CT CI ''FIT C'3 TI)! '211 UY1-4113 - - / "~ '1 v; '~"n 00 -60'2' ooo~~(1 81 6I"' ':' / RL'.- 0 219; 8* "1 '28'L ICL II'11 I SsNI - I 00 MI I 00 Z17, 9 IT(CC '':' 0, El' fI / C':1 10F- 'I- 0 t,q`-IC '0 ~1 *0C 818 q', 'C6I7 ') 0Ir I 6,- 1 I'P12 7 '7'2L CP *t 000 iq 0. I T pI I (C 269 9'. Ie,1 11 '0C 6 '6r '271 t,'7I T 11701-IC T '61'I / N''- I, T7 II l' (4 ICCC IC ':"' 'Ci T 1 C1 fI': 9PC -;e,T '2rC 1 f 0 C/i (C 9O :' i CI, IC Z 0 C7 '6 9! 6 'CC T :'9'117 T T/: t '09 N 176 AlULI Au ''2 /.,177' 66 '~~1.C 60 ('6 ': 9/ 1 IC 61 ( CI'0-( ' 602 C '6 T29/', (I :I C'2/.' 'CI ('T'6)1 69 90U-1V3-11AN p , r k 'IC CP-P VI '00I 'C6' Cl: TC 0 IC TC IC 92 0.( '0 T/.L/.'26C '7, I 'C 7t '6 , e 796C C 6C '29 (11J3Th1 9 '2."C 0 I'AC 60/Iq 01' - 1. 2 ('6'I 0 0 9CCT'' 13~1 0C''-'0 "'2C T fIC'': 0 '2tC 69' 'Cl IIO '01 CZI 'TT9 T ' 1' 'CLC ~ ~~0' ''2 17? 1/. t'/. I 06 '0 (' 1 C Ci .C'' /1I A1C18I '0 ' '171 Z '2I 2 '6J 9C0' '.17I t 'IC T ("2 (INVAP .I 6 '0111 ''C ''21' C~~~~'2 1 ('(3 ''2C I ~~ ('0.'f '0 C2 / 66(1C T I 1 7r.: (''1117': '0 I_ 2'2/)t T(' '( 6 ''2 TC6' T. 6162 'T 6, 2'' 6 7 '1VI(T I 6 ('6 '06- 618''I? T 00 '7, T!2 17 'C' C17. 162 9T'tpI . I_/A,T 7(, 17 COC V 'T ' 1, 6'2 6C 'Z: Z9113 '(' 9 '('1 t1C ATi-V1- 1 17'' 097 ZI C3 02 IC8C Z OC (' 111313 f2' 2 FC( "I) CC'266 I0 r'' t;, ''. Z199'7 6 6 ) 11, ''P2 iM l"2 '2'2''2 /61 '26 '61 6CC'2 '') IC /.L6171 '2211 6C('8C(' '':' C/CL'/.C 'C' 8 'LCI, [' 9 '6-('I61 '1 £ '6 (rC T I NUJ 6 '1.1011 0':' 'i':"~ ':' 179C9 ':'c' 'I,- 9'2 LC2' I'C'R 6 T I 0 IC ' IC90 ' t' C6'/ I; o'( '91i, 86£. '; T 1 T' T ) 8 'C 6C LO YI' C '9C9T ('6 '9'2I C9C17 ('0 'Ct'11 10CR'')~~t- ('C899 /. I T f)'0C~213 '0 72''2C ('119'9 T 19'2''C ' 1IP 2 1 -I /9.-VM '2-1-4 691~1~I 1 (Id 93>J:I AW'35 6921YJ C~t-1 1 .Yci 69Y.L I 11CJO I'1CI '691 A.9 V - V GI- JI iUNJ 1-17 U0Tl1 fl C' U0i1, 7,~q~.~ 1!Z I~; nonr qT InqgTA 6, gj UOI2l,( t 'l' 1AZ W 6Uo@Tiwi9;jD__I ~ _____________________ _______________t______ I 112I, 0 I -j -01) -9 T f f -7lT ( R T)onb nny 2UIA'cpf I1J ~~ ~npUi ~~'$'d lndan0 1~~~~T19A 69'-111TJZ A 6 0) Cl ( c7, IEl 0: OW jvnW L, ON NOSHJ]d J NOS fVI4O III 6 9r"-( Tt -T9 t 1 f i, q Z fq_:Nj - 84 - TA069: total agricultural output in wheat equivalents per hectare (quintals/ha.) QL3LD8: total output in wheat unit per hectare. QF3LD8: Fish index of total output per hectare. IPR69: international price ratios - kg of fertilizer that can be purchased with 100 kg of wheat equivalents. PRI: wheat equivalent total price index deflated by fertilizer prices. PR2: Fisher price index deflated by fertilizer prices. TRE69: total research SSYM: stock of scientist man-year. PRC69: precipitation/year. FDl: FWD - reduction factor caused by water deficiency. FRT69: Weighted average prices of fertilizer in local currencies (local currencies/100 kg). FP5: fertilizer prices in local currencies (local currencies/MT). WEP69: weighted average prices of farm products (wheat equivalent price per kg). PI3: wheat equivalent total price index. Note: Data for the above variables are 1968-70 average. - 85 - REFERENCES Askari, H. and J.T. Cummings, Agricultural Supply Response, New York, Praeger Publishers, 1976. Bapna, S.L., Aggregate Supply Response of Crops in a Developing Region, New Delhi, Sultan Chand & Sons, 1980. Bapna S.L., Hans P. Binswanger, Jaime B. Quizon, "Systems Output Supply and Factor Demand Equations for Semi-Arid Tropical India," Indian Journal of Agricultural Economics, 39(1984):179-202. Bardhan, Pranab K., "Determinants of Supply and Demand for Labor in a Poor Agrarian Economy; An Analysis of Household Data in Rural West Bengal," in Binswanger and Rosenweig, eds., Contractual Arrangements, Employment and Wages in Rural Labor Markets in Asia, New Haven, Yale University Press, 1984. Bond, Marian E., "Agricultural Responses to Prices in Sub-Saharan African Countries," International Monetary Fund Staff Papers, 30(1983):703-26. Boserup, Ester, Conditions of Agricultural Growth, Chicago, Aldine Publications, 1965. Brandow, G.E., "Demand, for Factors and Supply of Output in Perfectly Competitive Industry," Journal of Farm Economics, 44(1962):895-9. Buringh, P., H.D.J. van Heemst and G.J. Staring, "Computation of the Absolute Maximum Food Production of the World," Agricultural University, Wageningen, Netherlands, January 1975. Carlson, Sune, A Study of the Pure Theory of Production, Augustus M. Kelly Inc., New York, 1956. Cavallo, D. and Y. Mundlak, Agriculture and Economic Growth in an Open Economy: The Case of Argentina, IFPRI, Washington, D.C., 1982. Caves, Douglas W., Laurits R. Christensen and W. Erwin Diewert, "Multilateral Comparisons of Output, Input and Productivity Using Superlative Index Numbers," Economic Journal, 92(1982):73-86. Caves, Douglas W., Laurits R. Christensen and W. Erwin Diewert, "The Economic Theory of Index Numbers and the Measurement of Input, Output and Productivity," Econometrica, 50(1982):1393-1414. Colman, D. and A.J. Rayner, "Feed Demand Elasticities and Agriculture's Aggregate Supply Elasticity," Journal of Agricultural Economics, 22(1971):125-9. - 86 - Chhibber, Ajay, "Long Term Agricultural Growth in Developing Countries: A Framework for Comparing the Role of Price and Non-Price Policies," FAO Discussion Paper, Rome, September 1983. Desai, M. Gunvant, Sustaining Rapid Growth in India's Fertilizer Consumption: A Perspective Based on Composition of Use, Washington, D.C., IFPRI Research Report #31, 1982. FAO, UNDP and International Institute for Systems Analysis, "Potential Population Supporting Capacities for Lands in the Developing World," Technical Report of Project on Land Resources for Populations of the Future, 1984. Floyd, G.E., "The Effects of Farm Price Supports on the Return to Land and Labor in Agriculture," Journal of Political Economy 73(1965):148-58. Friedman, M., Price Theory, Chicago: Aldine Publishing Co., 1962. Griliches, Zvi, "Estimates of the Aggregate US Farm Supply Function," Journal of Farm Economics, 42(1960):282-93. Hayami Y. and V.W. Ruttan, "Agricultural Development: An International Perspective," Johns Hopkins University Press, 1973. Johnson, D. Gale, "The Nature of the Supply Function for Agricultural Products," American Economic Review, 40(1950):539-64. Kravis, Irving B., Alan W. Heston and Robert Summers, "Real GDP Per Capita for More Than One Hundred Countries," Economic Journal, 88(1978):215-42. Mundlak, Y. "Empirical Production Function Free of Management Bias," Journal of Farm Economics, 43(1961):44-56. _ "On the Pooling of Time Series and Cross Section Data," Econometrica, 46(1978):69-85. , Intersectoral Factor Mobility and Agricultural Growth, IFPRI, Washington, D.C., 1979. , "Cross Country Comparison of Agricultural Productivity," Rehovot, The Center for Agricultural Economic Research, Working Paper No. 8105, 1981. , "Endogenous Technology and the Measurement of Productivity," Rehovot, The Center for Agricultural Economic Research, Working Paper No. 8410, 1984a. , "Capital Accumulation, The Choice of Techniques and Agricultural Output," IFPRI, Washington, D.C., 1984b. - 87 - fundlak, Y. and C. McCorkle, "Statistical Analysis of Supply Response in Late Spring Potatoes in California," Journal of Farm Economics, 38(1956):553- 69. dundlak, Y. and R. Hellinghausen, "The Intercountry Agricultural Production Function: Another View," American Journal of Agricultural Economics, 64(1982):664-72. Ierlove, M., The Dynamics of Supply, Baltimore, Johns Hopkins University Press, 1958. ?andey, S., R.R. Piggott and T.G. MacAulay, "The Elasticity of Aggregate Australian Agricultural Supply: Estimates and Policy Implications," Australian Journal of Agricultural Economics, 26(1982):202-19. ?eterson, W.L., "International Farm Prices and the Social Cost of Cheap Food Policies," American Journal of Agricultural Economics, 61(1979):12-21. Quizon, Jaime B., and Hans P. Binswanger, "Income Distribution in India: The Impact of Policies and Growth in the Agricultural Sector," Agricultural Research Unit Discussion Paper 21, World Bank, Washington, D.C., April 1985 (revised). losenweig, Mark R., "Determinants of Wages Rates and Labor Supply Behavior in the Rural Sector of a Developing Economy," in Binswanger, Hans P. and Mark R. Rosenzweig, editors, Contractual Arrangements, Employment and Wages in Rural Labor Markets in Asia, New Haven, Yale University Press, 1984.