Aspects of savings behavior in rural India

FT-rIES IN DOMESTIC FINANCE No. 31 ASPECTS OF SAVINGS BEHAVIOR IN RURAL INDIA By Surjit S. Bhalla* Princeton University and Brookings Institution Public and Private Finance Division Development Economics Department * This paper has benefitted from the views of*several people, for which I am grateful. I would especially like to thank Suman Bery for his interest in the project as well as for his comments. Discussions with him were extremely useful. I would also like to thank Philip Musgrove for helpful discussion, and Montek Ahluwalia, Alan Blinder, and Graham Pyatt for comments on an earlier draft of this paper. The responsibility for any remaining errors is mine. December 1976 PREFACE While we have some idea about the saving behaviour of pure enterprises on the one hand and of pure households on the other, we do not know much about the saving behaviour of household enterprises - which include farm households. Since such hybrid enterprises are likely to account for a major part of saving of the so-called household sector (which comprises pure households as well as household enterprises), their saving behaviour is of particular significance from the point of view of policies for resource mobilisation. One can venture some conjectures about the nature of this be- haviour with regard to farm households. For obvious reasons, saving and investment decisions would be vitally related; the dominant saving motive would be to finance invest- ment. Further, investmeht decision would be based on the perception of profitable investment opportunity. Thus, given such opportunities, invest- ment would be constrained by capacity to save (income above some minimum conventional subsistence level) and hence one could expect saving and investment to be an increasing function of income. However, given the limits imposed by scarce factors like land and managerial ability, it is likely that profitable investment opportunities for a given household, after a certain level of assets and income is reached, may level off; at this stage the constraint on saving would be investment and the average saving rate may not increase. Though these conjectures are not explicitly discussed by Mr. Bhalla, they seem to be corroborated by this study on Rural Savings Behaviour, particularly the aspects of this study relating to a compara- tive analysis of the saving behaviour of households affected by the Green Revolution and those whose technological horizon did not widen. However, the survey results analysed in this paper are for only three years - and that too for a period immediately after the onset of the Green Revolution. It may be interesting to study rural saving-investment and flow-of-funds behaviour over a longer period, particularly to identify the impact on this behaviour of policies relating to (a) land reform, (b) technical change and agricultural extension, (c) primary education and, (d) credit. The Division is currently formulating a project of this type based on an analysis of the rural survey data of the Reserve Bank of India for the years 1951-52, 1961-62 and 1971-72. V.V. Bhatt TABLE OF CONTENTS Part I 1. INTRODUCTION . . . . . . . . . . . . . . ... . . . . . . Part II - A DESCRIPTIVE STUDY 1. Description of Data - NCAER Survey . . . . . . . . . . . . . . . . 7 2. Concepts and Definitions - Savings and Income . . . . . . . . . .10 3(a).Aggregate Estimates and the Choice of a Savings Definition . . . .13 (b).Estimates of Savings and Income by Occupation. . . . . . .. . . . 21 4(a).Pattern of Investment. . . . . . . . . . . . . . . . . . . . . . .25 (b).Non-Monetized Patterns of Investment . . . . . . . . . . . . . . .27 5. Asset Levels and Saving Rates. . . . . . . . . . . . . . . . . . .31 Part III - AN ANALYTICAL STUDY 1. The Savings-Income Connection. . . . . . . . . . . . . . . . . . .35 2. Permanent Income - Definition and Measurement. . . . . . . . . . .42 (a) Permanent Income - A weighted.Average . . . . . . . . . . . .42 (b) Permanent Income - A Modified Earnings Function . . . . . . .47 3. Permanent Income Measures - Consumption Function Results . . . . .58 4(a).Savings Model - A Derivation . . . . . . . . . . . . . . . . . . .63 4(b).Savings Model - Propensities to Save . . . . . . . . . .. . . . . 67 4(c).Savings Model - Effect of Trnasitory Consumption . . . . . . . . .73 5. Towards a New Savings Function . . . . . . . . . . . . . . . . . .75 Part IV - SPECIAL TOPICS 1. Investment Opportunities and Savings . . . . . . . . . . . . . . .84 2. Sources of Income and Savings. . . . . . . . . . . . . . . . . . .98 Part V 1. Summary and Conclusions. . . . . . . . . . . . . . . . . . . . . 110 APPENDIXES 1. Criteria for Selection of Observations . . . . . . . . . . . . ..116 2. Sources of Income Equation - Derivation of Formulae. . . . . . . 118 References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121 1 I. Introduction Economic policies in developing countries are geared towards two major goals--economic growth and a more equal distribution of income. Both of these goals are affected by the level and pattern of domestic savings. The connection between economic growth and savings is predicated on the assumption that long-run income growth in developing countries is constrained by a lack of finance rather than of investment opportunities. The bulk of this finance in most developing countries comes from domestic savings--hence, their rele- vance. Less clear is the connection between aggregate domestic savings rates and income distribution. There is a strong possibility that there is no relationship between the two. Only under very special assumptions does progressive income redistribution affect savings and (through savings) growth. Both the permanent income and life cycle theories of consumption behavior explicitly postulate that the savings rate for a household (or an economy) is independent of the level of permanent income. A simplified Keynesian savings function allows average savings rates to increase with income but postulates a constant marginal savings rate. This rules out any changes in the aggregate savings rate with moves towards a more equal distribution of income. However, if the marginal savings rate is allowed to increase with income, then aggregate savings will decrease with progressive income redistribution. Thus, the quad- ratic Keynesian function is the only one of the received theories of savings behavior to allow for an income distribution effect on savings. 2 -Aggregate. savings in an economy are composed of private savings (households and uninco;poi;ated enterprises), corporate savings and savings. By- the government sector. Private sayings form a major part of gross domestic savings in -most developing countries. In a recent survey of the ECAFE countries, Alamgir (1974) reports that private. savings were often 60-70% of national savings, with.South.Korea an outlier.at 35%. In India for 1970-71, '(the period for which- data in the present study were collected) the household sector alone accounted for 78% of gross domestic savings. Thus, a study of household savings may be informative about movements in national savings. Extensive analysis of how households vary- in thei saving behavior has been undertaken for the.U.S. and other data-ricli countries. These studies have centered on the. testing of three major theories of consumer behavior-Keynes' absolute income hypothesis, Friedman's theory of permanent income and the Ando-Modigliani-Brumberg's life cycle. hypothesis-- and have involved an examination of both. time series and cross section data. Indeed, a major purpose of the. non-Keynesian theories is the reconciliation of time series evidence that the overall saving rate of the household sector is independent of the level of income with cross- section data that shows that at any given time richer households have higher saving rates. The evaluation of these theories in the less-developed countries (LDC's) has, however, been restricted almost exclusively to time series data, with- some researchers extending the aggregation of data to study inter-country behavior. This macro testing of consumption theory, for wah-t is assentially-a theory of individual hehavior, has undoubtedly been forced on the researchers through the- paucity of micro data in developing countries. Only a handful of household studies exist for developing countries. 3 Paniar'-s. C1961) study is a collection of results fr7om various surveys on te. average., saving rates in rural India. Friend (1661 uses grouped data for Indian househblds7, and Kelley-Williamson (168) analyze rural and urban data from the. Jogjakarta region of Indonesia. Ramanathan (1969) explores the savings behavior of urban households in Deli, India, and Betancourt (1971) uses extensive cross-section data for Tboth the urban and rural areas of Chile. Two major projects in recent years have tried to generate information and analysis on household expenditure behavior in the LDC's. One is the ECIEL study on urban households in Latin America. Studies resulting from this project are those of Crockett-Friend (1973) and Musgrove (1974a, 1974b). The other major project is the Ohio State University study on Agricultural Capital and Technology, which has dealt to a large extent with Taiwanese data (see Ong. et. al (1974)). Though this listing of household studies is not exhaustive, it is indicative of the small numbers involved. By contrast, the Mikesell-Zinser (1973) review of the savings literature shows that time series studies number well over a hundred. This paper is concerned with.the analysis of household saving behavior in rural India. As its basis- it uses the data collected by the National Commission for Applied Economic Research (NCAER) on some 4,000 households in rural India. The data are unique for a developing country in that the information was collected from a panel of households for three consecutive years--1968-69, 1969-70, and 1970-71. The paper is divided'in three. parts. Part II of the paper is concerned with. a descrt tive- analysis of savings behavior as revealed 1 by the NCAER data. The nature of the data is discussed in Section I1.1, 1This part of the paper can, and should, be. skipped by any reader not interested in "numbers." Part II of the paper was written to offer readers a "feel" for the data. 4 and the definitions of two important variables--sayings. and income--are discussed in II.2w The. dat, contains independent information on savings (change in net worthY, income and consumption of each household. Thus, two estimates of household, savings are possible, Sd (change in net worth), and Cincome-minus consumptionj. The relative merits of using either definition is discussed and aggregate estimates of savings constructed, and compared with national estimates in Section 11.3. (The SdW definition performs-much Better than the S definition). The Y.-C pattern of investment is discussed in Section 11.4. Since the NCAER data contain information on the non-monetized components of investment, (presumably an important component of investment amongst rural Indian households) this aspect of saving is studied in somewhat greater detail (Section II.4(b)). Section 11.5 concludes the discussion on "descriptive" results by assessing the relationship between sources of income, use of technology, level of assets and saving rates. Parts III and IV form the analytical parts of this paper. In Part III, a detailed investigation of the savings behavior of rural (cultivator households) is undertaken. The. analysis is concerned with questions related to the issue of "redistribution.vs. growth." Ia particular, answers to two important questions are sought - (a)' what concept of income best describes the savings behavior of a household: current income or permanent/life-term income and (b) what form does the relationship take-linear, quadratic, proportional; etc. The difficult question of the appropriate income concept is 1The method of selection of observations is discussed in Appendix I; Part II of the paper discus3es.the reasons why only the cultivator sub-sample is analyzed. 5 discussed, and the panel. nature of the data exploited, to yield two different estimates of permanent.income. Consideratimas of discount rates, expected income, and expected growth of incom& dictate. one measure of permanent income, Y where i is the suBjective discount rate. A different procedure, namely an earnings function, is used to derive another estimate,of permanent income, Y . In addition to the Px usual parameters (physical assets, level of technology, etc.), this method estimates the impact.that (permanent) individual differences have on the earnings of a household. By allowing for individual differences, Y is able to determine permanent income, more accurately , pm- than has previously been possible with data from LDC's. 1 Though not unique in its construction1, it is likely that such an estimate of permanent income has not been used before in a study of consumption behavior. These measures of permanent income are used together with measured income to determine which concept of income provides a better explanation of savings behavior, and to test the validity of the proposition that saving rates are independent of the.level of permanent income. The question of the proper functional form to represent the relationship between savings (S) and permanent income (Yp) is also examined in detail. In Section TII, a "new" savings function2 is proposed n tested; one which postulates a non-linear dependence of saving rates on permanent income. This new. function is compared with. other nonlinear relationships and found to yield more plausible resulta. 1See Gordon (1976) and Lilliard (19751. 2 Essentially, this function. allows saving rates to increase non- linearly with income, and approach. a constant asymptotic value. 6 Section TV discusses two. "special topics" in the theory of savings behavior--the- effects of investment opportunities, and of the. sources of income on savings propensities. The NCAER data are particularly appropriate for the. testing of the: "investment opportunity" effect since the survey was conducted at a time. when the riih-yie.lding varieties (YV) of foodgrains were- being adopted in raral India. Though often talked about, the. effect of investment opportunities on savings has never been directly tested. Section IV.1 discusses the difficulties involved in constructing an appropriate index of investment opportunities, and makes a preliminary attempt at testing its impact on savings. Section IV.2 tests for differences in the propensitias to save from two sources of income - agricultural and non-agricultural. Past studies have attributed differences in observed propezisities to differences in share of "profits",' investment opportunities, etc. The Friedman hypothesis of permanent/transitory income offers an alternative explanation - namely, that differences in the variability in the sources of income account for differences in the propensities to save. This hypothesis is discussed and explicitly-tested in this section. The major conclusions of the study are summarized in Section V. 7 SECTION 11.1 - The 'Data The National Council for Applied Economic Research (NCAER)under- took a survey (known as the Additional Rural Income Survey (ARIS)) of 5,115 households in 1968-69 to gather data on .the distribution of in- come, and the pattern of consumption, savings and investment of these households. The sample was selected according to a multi-stage strati- fied probability design in order to provide a representative cross-section of the rural Indian population. Higher income households were over- sampled. The survey was repeated in 1969-70 and 1970-71 on the same households,and the final version of the data refers to a core sample of 4,118 households. The NCAER collected detailed information on the age-sex composi- tion of the households, the composition of income (by sources and by occupation), the consumption pattern (food, clothing, unexpected expen- ditures, etc.) and the ownership of and expenditure on selected assets (farm land, farm equipment, irrigation, consumer durables). Detailed data relating to transactions in the capital market are available for the third year only. In addition, "background" data on the villages where the households reside are also available. The NCAER survey is rare in that it collected independent informa- tion on the consumption, income and savings of households. About 25 food items, 10 non-food items (fuel, clothing, education, medicines, etc.) and "other" items (marriages, funerals, unexpected travel, etc.) comprise the information on consump tLon expenditures. Independent estimate of savings is available by using the "change in net worth" concept, i.e., additions to and subtractions from the stock of assets held by the household. Thus, two estimates of savings (S) can be 8 derived using the NCAER data: the "residual" estimate S (income- y-c consumption) and the "direct" estimate, S dW (change in net worth). Aggregate estimates obtained by using the two definitions of savings are compared in Section 11.3. The panel nature of the NCAER data has advantages even for one period analysis. Surveys are often susceptible to measurement error, a form of which is individual response error. If these errors are significant, and caused by factors like ignorance or non-compliance, a panel study should minimize them, since the respondents should get used to the questions over the years. Thus, it might be reasonable to assume that the information for the third year, 1970-71, is more accu'e rate than would have been available if the survey had no prior history. At the same time, data for the later years may suffer from bias result- ing from the conditioning of the respondents. The eff"ct of repeated questioning on the reliability of the data is, therefore, uncertain. These caveats notwithstanding, it is believed that the third-year data are the most reliable of the three years and have, consequently, been used as the basis for most of the analysis and hypothesis testing in this study. The group most extensively analyzed in this study are families 1 engaged in cultivation for all the three years of the survey. The savings behavior of the non-cultivators is analyzed in considerably less detail. (See Section 11.3). The.emphasis on the behavior of 1 A household was defined as a cultivator if it engaged in any kind of self-cultivation on owned or leased land. Major occupations of non-cultivators were self-employment non-farming, and agricultural labor. cultivators was dictated by a number of factors. This group comprises about 75 petzent; of the total sample and is interesting to analyze be- .cause of its firm-household nature. The data base is also more exhaustive for the cultivato,s--no asset or credit information is available for the .non-cultivators. Non-cultivators also comprise the poorest sec- tors of the population and have very little saving capacity. (See Table 3). Out of a total sample of 4,118 households, 2,532 were cultivators for all of the three years and 790 had the status of non-cultivators. The remainder changed status at least once during the three-year period. Summary information on the savings behavior of non-cultivators and 'mixed status' households is presented in Section 11.3. The rest of the paper is concerned only with 'continuous' cultivators and the sam- ple analyzed consists of 1,980 households. Appendix I spells out in detail the procedure used to select these units out of the available sample of 2,532 households. 10 SECTION 11.2 - Concepts and Definitions Definitions affect estimates, and it is important that the basis for these definitions be clearly outlined. This section discusses the definitions used by NCAER for two important variables-- income and saving. XA) Income - The income of a household is defined as the total of the earnings of all the members of a household during a reference period. 1 This income can be business income (farm or otherwise), wages, rents (land and house property), interest and dividends on financial invest- Ments and pensions and regular contributions.2 The definition of income includes any non-monetized investment undertaken by the household, as well as imputed rental income from owner- 3 occupied housing. Though theoretically correct, it can be argued that nou-monetized investment and imputed rentals are not perceived as income by a household, and hence should not be included in an analysis of savings behavior. No stand is taken on this issue, and the above definition of income has not been adjusted. Empirically, it is doubtful whether adjust- ment would make much difference since these particular sources of income account for a very small proportion of total income. The treatment of "Pensions and regular contributions" as part of income does raise some questions. This category includes in it any re- mittances sent/received by the household. The definition of gross income, 1In computation of farm income, all farm output, whether marketed or not, is valued at market prices. 2Income tax, which is negligible for most rural households, has not been deducted to arrive at a figure for net disposable income. Lack of data prevented this theoretically appropriate deduction.. The distinction between traded and non-traded inputs and the methods used to impute value to non-monetized investment are discussed in Section II.4 (b)-Non-Monetized Forms of Investment. 11 according to NCAER, excludes any remittances sent by the household but includes remittances received. If one is interested in the estimation of national accounts parameters, then inclusion of remittances sent and received would amount to double counting. But the question remains as to what is the proper concept for analyzing the behavior of a household. By excluding remittances sent from the income of a household, one is making the implicit assumption that these remittances are in the nature of a tax for the household and its omission results in the proper con- cept of "disposable income." In other words, this definition of income is -assumed to be the one relevant for household behavior. The inclusion or exclusion of remittances sent in the income con- cept does not in any way affect the absolute savings estimate of a household, though saving rates are obviously affected. Though it was felt that remittances sent should not be excluded from gross income, no correction was made to the NCAER data. This decision was made for the following reasons: (a) Data on remittances sent is not available for the first and second years of the survey, so definitions of income would be inconsis- tent for the three years. (b) Only a very small proportion of cultivators (2.5 percent) sent remittances in the third year and most importantly, (c) the alternate definition of income had a negligible effect on the marginal and average propensities to save. (B) Savings - The saving of a household is defined as the change in net worth and computed as the difference between the change in the value- of assets and the change in liabilities. This figure is adjusted for capital transfers. In other words, household saving, Sdw is defined 12 to be: S = PA + FA -dL -K dWI where - PA - Gross change in the value of physical assets FA = Gross change in the value of financial assets dL = Net change in liabilities ] Net inflow of capital transfers The savings estimate includes via PA any purchases of consumer durables, and non--monetized investment that is undertaken by the house- hold.Savings in the form of currency or gold and silver are not included due to lack of reliable data; nor has any adjustment been made for capi- tal gains or losses incurred by the household. Depreciation on assets is also ignored. The savings estimate, SdW, is a direct estimate. The NCAER survey also gathered information on the consumption expenditures of a house- hold. Thus, it is possible to form a residual estimate of savings, S , which is gross income minus reported consumption. The results Y-c according to the two definitions are compared in the next section; S yields estimates of savings which are almost twice the magnitude y-c of SdW* Comparison with independent estimates of savings suggests that SdW is the more reliable estimate; hence, it is used in the analysis of Parts III and IV of this paper. . ... .... 13 Section 11.3 - Aggregate.Estimates and the. Choice of a Savings Definition The NCAER Survey is rare in that it collected.iadependent information on the- consumption (C, savings (S) and income CY) of rural households. The survey organizers did not attempt to reconcile any discrepancies in the estimates. Crhe variables are related by the identity, Y = C + S). Consequently, two estimates of savings (and consumption) can be derived from the data - change in net worth, SdW, and "residual" savings, S . If measurement errors are absent and consistent definitions of the variables are followed, the two measures should yield identical es;timates of savings. Prior belief would indicate that measurement errors are present, and likely to be higher for SdW this under the assumption that greater error is associated with the reporting of assets than of consumption items. However, if the assumption is made that these errors have zero mean and are independent of the true values, aggregate estimates of SdW and SY-c should still be approximately equal. Two considerations suggest that this may not be likely with the NCAER data. The first expected difference is due to matters of definition. The S estimate dW excludes savings in the form of cash aid jewelry whereas S being a Y-C residual, includes these items. Thus, S is expected to be larger Y-C than S dW. The second expected difference is more fundamental and is the result of the particular aims of the. NCAER survey. Assessment of invest- ment and changes in assets of rural households was one of the major goals of the survey, whereas enumeration of particular consumption expenditures was not. Only thirty-six specific items, including the cateogry "other expenditures" were used to estimate consumption. This methodology 14 increases the likelihood that a certain fraction of consumption expenditures was systematically excluded from enumeration. Thus, if S is computed as Y-C, the excluded consumption items will be- erroneously included in savings, thereby boosting up the S estimate. Y-C Both. considerations (exclusion of cash/jewelry, lack of complete enumeration) imply that S should be systematically higher than SdW Y-C Given these caveats, it is inteesting to compare the relative magnitudes yielded by the two estimates (Table 1). Examination of the data for all households, (4118) shows that the rural sector has positive savings with SdW estimate ranging from 2.9 percent to 6.9 percent, and S from 12.7 Y-c to 16.5 percent. These estimates are based on the- "raw" data, and are likely to include measurement and transcription errors. Lack of availability of interview schedules prevented the correction of possible errors in the data. However, one can indirectly guard against transcription errors by 'selective sampling'. In particular, 'inconsistent' and 'uarealistic' observations can be eliminated to- yield a clean sample of observations. In Table 1, consistent observations refer to households which had reported savings, S as being less than income for any year of the 1 survey. Exclusions of these "high" savers causes an expected decline in S for all the three years of the survey. Though naminal changes dW occur in other years, SdW declines significantly from 2.9 percent to 0.6 percent for 1968-69, with incomes remaining relatively unchanged. This radical change in the estimate of Sc points to the sensitivity Given the definitions used, it is impossible that FdW is greater than income. Negative incomes, though plausible, were also excluded from consideration, since the weighting process would tend to under- estimate aggregate rural income. 15 TABLE 1 INCOIE (Rs.) AND SAVING RATES (%) AVERAGES - NCAER SAMPLE '1968-69 1969-70 1970-71 Popu- Popu- Popu- Sample lation Sample lation Sample lation All Observations (4118) Income 3841 2314 3975 2501 4208 2649 Savings 4.9 2.9 10.4 5.0 19.2 6.9 dW Savings 21.8 12.7 21.6 14.7 24.3 16.5 ~y-c * Consistent Obs. (4013) Income 3846 2322 3949 2486 4165 2619 Savings, 3.0 -.6 9.3 4.2 12.1 5.4 Savings, 22.3 13.6 21.4 14.6 24.1 16.3 y-c Selected Obs. (3835) Income 3869 2338 3935 2479 4151 2604 Savings, 3.5 .6 9.6 4.8 13.1 5.8 dw Savings, 23.3 14.8 21.7 15.0 24.5 16.7 y-c Notes: 1) - Observations with negative incomes or with reported savings greater than income in any of the three years are excluded. 2)- Consistent observations with "outliers" excluded i. e. ob- servations with saving rates less than - 150% or greater than 75% in any of the three years. 3) - The definition of savings, SdW includes purchases of consumer durables and non-monetized investment. Savings, S , is a residual estimate of savings and is obta.ned by sutacting 16 reported consumption from income. (For a. complete definition of these measures, see p. 10-12 of the text.) 4) - Population estimates have been derived by using popula- tion weights. High income households were oversampled in the NCAER survey. I--rg iii 17 of results when population weights are used. The weighting process can magnify errors; sanple estimates are affected less by exclusions since each observation has the. same weight. A further attempt to refine the data led to the creation of a third category-selected observations. These are consistent observations with. extremely high CSdW > 75 percent) and extremely low (5 dW < -150 percent) households excluded if they reported such rates in any year of the survey. Saving rates of -500 percent at low incomes and 90% or so at low to medium incomes did suggest the likelihood of a transcription 1 or measurement error. Interestingly, this select set of observations did not yield results very different from the set of consistent observations. Also, estimates for S are unaffected by either selection criteria Given that observations were excluded on the basis of SdW, these results suggest that errors in SY-C are independent of errors in SdW* Comparison of SdW and S indicates that the latter estimate is consistently higher--compare 5.4% to 16.3%, 1970-71. As previously indicated, this is to be expected, but it is doubtful that excluded items from SdW--cash and jewelry--account for a major share of the difference in the two estimates. Thus, it is likely that omitted expenditures from the consumption estimate are causing this large difference. Independent information on consumption and income was used to assess the reliability of extreme. S values; in most cases, these values dW were radically out of line with.the reported.income and consumption ex- penditures. Admittedly, some of the observations might indeed have reflected correct values; in the interest of 'objectivity,' an arbitrary cut-off value was used on both the high and.low side. 18 If independent, and reliable, estimates of rural savings vere available, then one could directly test the. (aggregate.) accuracy of S and S . Unfortunately, no independent estimates exist for dW Y-C savings in the rural sector. However, estimates of Fousehold saving for rural and urban sectors combined are available from national accounts (published by the Central Statistical Organization, CSO). The Reserve Bank of India, (RBI), though employing a different methodology, also published estimates of household savings. Though both CSO and RBI estimates are for all households, their magnitudes can nevertheless set bounds on the estimates for rural households. What is known about the rural and urban sectors is that the latter has higher per capita income. If saving rates are independent of permanent income (measured income and savings/consumption at the aggregate level can be an estimate of permanent values if it is assumed that the transitory components have zero means), then all three estimates of savings should be equal. If, however, saving rates increase with the level of permanent income (and the analysis of Part III suggests that this is the case), then the CSO and RBI estimates, by virtue of combining the richer urban households, should yield higher saving rates. Th_Vs, the results should show that the NCAER estimates of aggregate saving are less than, or equal to, the CSO/RBI estimates. Table 2 compares the saving rates according to the different surveys. The S estimate is consistently lower than the CSO and RBI 'Differences in methodology amongst the three surveys may also cause a variation in saving rates. Investigation of this issue is beyond the scope of this paper. 19 TABLE 2 COMPARISON OF SAVING RäIES - NCAER AND OTHER SURVEYS 1968-69 1969-70 1970-71 NCAER SURVEY S -0.6 4.2 5.4 dW S 13.6 14.6 16.3 y-c * CSO ESTIMATE 8.7 9.4 10.1 (excluding currency) S .7.0 6.4 8.1 RBI ESTIMTE Notes 1) Only consistent observations--4013--have been used in computing the NCAER estimate. 2) The CSO and RBI estimates are on the basis of net disposable in- come whereas the NCAER estimate is on gross inc~ne. Lack of taxation data for NCAER household prevented the =mputation of a disposable income estimate. The margin of err= is not expected to be significant; little taxation occurs in the rural households of India. 3) * - Estimates published by the Central Statisti=1 Organization, Department of Statistics, Ministry of Planning, ,1overnment of India. ** - Report on (brrency and Finance, 1971-72, p~iished by the Reserve Bank of India, Bombay, 1972. 20 estimates and the S estimates are consistently higher. The magnitude of SY-C appears to be abnormally high-14 to 16 percent-a rate which. is 50 percent more than the Gupper boundl CSO estimate. The SdW estimates seem to be of the right magnitude, though. the 1968-69 estimate-".6 percent-is unexpectedly low, especially in comparison with the estimates for 1969-70 (4.2 percent) and 1970-71 (5.4 percent). Given that observed incomes did not increase drastically, one can only conjecture that underestimation of savings seems to be more pronounced in the first year of the survey. Though not conclusive, the results of the. comparison of SY, SdW and the CSO/RBI estimates, indicate that S dW is a more accurate indicator of household savings. Even if differences in the aggregate value of S and S were not observed, statistical consideration would argue in favor of the use of Sdw in an econometric analysis of savings behavior (Parts III and IV). Since income is a major determinant of savings, use of S rather than SdW, would yield biased estimates of coefficients if incomes are measured with error. A regression involving SdW and Y does not suffer from a "cozmon" measurement error and is to be preferred.1 All these considerations support the view that S should be dw selected as the variable for analysis. Consequently, unless otherwise stated, the rest of the paper will deal only with results obtained by using the 'change in net worth' definition of savings. Analagously, if consumption were the dependent variable, one would use the direct estimate. pf consumption,.C, rather than the residual estimate, CY- 21 Section II.3Qh) - Estimates of'Savings and Income by Occupation Differences in savings behavior, by occupational groups, is revealed by Tables 3 and 4. Table 3 presents estimates of saving rates for three different groups in the econamy--cultivators, non-cultivators and those that changed status during any of the three years of the survey. Tables 4(a) and (b) show cross-tabulation figures for two of these groups. The same story is told by all the tables - groups with higher incomes have a higher saving rate. Also, the tables reveal that non- cultivators (landless laborers, artisans, etc.) comprise the poorest sector of the rural economy - their average income is only Rs. 1768 a year, and they possess little saving capacity, 0.4%. (Note the high percentage - 48% - of non-cultivator households who report a zero saving rate, SdW - and the abnormally high saving rate, Sy, for these households). The results of this section are consistent with those observed in most survey data - saving rates do tend to increase with measured income. The question of the proper behavioral determinant of savings, i.e., measured or permanent income, and the independence of saving rates with respect to permanent income, is discussed in Part III of the paper. The next section discusses the pattern of investment and saving behavior of only the cultivators of rural India. Appendix I outlines in detail the method of selection of observations for analysis. 22 TABLE 3 INCOME (RS.) AND SAVING RATES (%) AVERAGES - OCCUPATIONAL GROUPS 1968-69 1969-70 1970-71 POPU- POPU- POPU- SAMPLE LATION SAMPLE LATION SAMPLE LATION Cultivators (N-2317) Income 4596 2895 4746 3169 4998 3302 Savings 3.7 1.0 10.9 7.0 14.2 7.8 dW Savings 23.1 14.0 24.2 18.8 26.3 19.6 Y-C Non-Cultivators (N=765) Income 2407 1602 2258 1593 2419 1768 Savings 1.3 0.4 5.9 -1.0 7.6 0.4 dW Savings 20.5 15.3 13.8 5.5 16.4 9.2 Y-C Mixed Status (N=747) Income 3108 2089 3135 2142 3301 2186 Savings 3.9 -0.1 5.9 3.2 12.2 4.8 dW Savings 25.8 16.7 15.5 11.9 22.1 15.3 Y-C Notes: 1) - See notes 2,3 and. 4. of Table 1 for method af selection af obser- vations, definitions, etc. 2) - A household was classified as a cultivator if it cultivated any land (owned ar leased in) during the year af the survey. The classification above, cultivator/non-cultivator, is based an the occupation o the household for all three years af the survey. Mixed status , therefore, refers to households which changed occupations in any ane of the years af the suzvey. 23 TABLE 4(a) CROSS TABULATION OF SAVINGS (dW DEFINITION) AND INCOME - N-CULTIVATORS, 1970-71 INCOME (Rs.) 1000- 2500- All Saving Rate % <1000 2500 5000 >5000 Obs. -150 - -50% 3.2 2.4 4.2 0.0 2.8 - 50 - 0% 17.5 20.3 11.2 9.2 17.0 0 ·% 70.0 53.0 28.7 15.8 48.4 0 - 10% 6.9 15.1 28.0 17.1 18.0 101 - 40% 1.9 8.4 25.9 47.4 13.8 40 - 75% 0.1 0.7 2.1 10.5 2.0 Mean Income- 727 1624 3517 8151 2418 Mean Savings -3.2 -1.6 4.5 22.0 7.6 # of Observations 160 404 143 76 783 Notes: 1) Cells represent percentage of households and the coluinns add up to 100% 24 TABLE 4 (b) CROSS-TABULATION OF SAVINGS (dW DEFINITION) AND INCOME - CULTIVATORS, 1970-71 2000- 5000 All Savings Rate % 2000 5000 -10000 >10000 Obs. -150 - -50% 6.5 2.7 .8 .5 2.8 -50 - 0% 29.4 19.2 8.8 5.4 17.4 0% 29.6 19.4 11.0 2.5 17.8 0 - 10% 23.6 31.6 24.4 14.4 26.3 10 - 40% 10.2 25.7 466 51.5 30.2 40 - 75% .7 1.4 8.4 25.7 5.5 Mean Income 1431 3260 6938 14835 4954 Mean Savings(%) -7.6 3.2 14.6 28.1 14.1 # of Observations 432 855 491 202 1980 Notes:- 1) Cells represent percentage of households and the columns add up to 100% 2) For method of selection of observations, sce Appendix I. 25 .Section II.4(a) - Pattern of Investment The question of portfolio choice (composition of investment) is an important one but its study is, unfortunately, beyond the scope of this paper. For descriptive purposes, Table 5 is presented, which summarizes the investment pattern on the part of cultivator households in 1970-71. Housing investment accounts for a considerable share--23 percent-- and 52 percent was invested in various forms of farm investment, e.g. land improvement, irrigation, livestock, etc. Financial assets repre- sented only 15 percent of total investment. Non-monetized investment (see pp. 26-27 of text for a discussion about its definition and valuation) accounted for 3-6 percent and consumer durables were 8 percent of total investment. The couiparison of HYV and non-HYV households shows that the two groups have similar patterns of investment, though HYV households have higher levels of income, investment and savings. Since irrigation is necessary for successful adoption of the new technology, it is not surprising that HYV households have a greater share (19.6 percent) of this investment than non-HYV farmers (15.5 percent). The other significant difference is in financial assets--19 percent (HYV) and 12 percent (Non- HYV). This could be a reflection of both the wealthier nature aud the better access to institutions on the part of HYV households. 'The gross investment figures reported by NCAER have been adjusted for the net inflow of gifts; the reported components of investment, how- ever, have not been adjusted. Consequently, the proportions reported in Table are expressed as a ratio to gross change in assets, i.e. gross investment plus gifts. This allows the components to add up to 100 percent. Gifts, incidentally, formed only a small component (<1 percent of aggregate investment). 26 TABLE 5(a) 00MPOSITION OF INVESTMENr - CULTIVATORS, 1970-71 All HYV Non-fYV Investment Type Households Households Households Land Improvement 11.8 7.7 16.9 Farm Equipment 6.2 6.4 5.9 Irrigation 18.6 21.2 15.4 Livestock 13.2 10.3 16.8 Other Farm Assets 1.6 1.9 1.2 Housing 23.4 24.3 22.2 Consumer Durables 8.1 7.2 9.2 Financial Assets 15.4 18.6 11.5 Self-Employment 1.7 2.3 .9 Non-Farming Non-Monetized 3.6 1.7 6.1 Investment Savings rate (%) 14.1 18.8 10.4 Savings Investment 91.7 97.3 85.1 Ratio (%) Land Owned (acres) 10.8 12.5 10.0 Gross Incom- (Rs.) 4954 6778 4103 # of Observations 1980 630 1350 Notes: 1) - Non-monetized investment is included in figures for land improvement and housing - hence, the colunns add up to more than 100%. 27 SECTION II.4(b) - Non-Monetized Forms of Investment Family labor is used for purposes of production and investment. Activities like on-farm land preparation, housing construction, digging of wells, etc. are undertaken by a farm household but not necessarily within a market framework. This use of own labor as a nn-priced in- put for investment is defined to be 'non-monetized investment' (NMI); such forms of investment are also simultaneously savings and income. Little, however, is known about either the magnitude or the determinants of non-monetized investment. Savings estiates have tra- ditionally not included this component and have tended to undezesti- mate the savings and investment in the rural economy. The NCAER survey is the first survey to collect data on the non-monetized forms of house- hold investment on a national basis. Estimates (in value terms) of family labor input are reported for 'land improvement' and 'housing Con- struction.' (The former accounts for approximately 90 percent of the total). Analysis of NCAER data should be useful for assessing biases in figures of investment, saving and income that ignore this component. However, the very nature of NMI makes this a difficult task. Even though the presence of NMI can be established, its valuation poses ser- ious measurement problems. The method, and figures, of valuation will seriously affect any estimate of NMI. What price should be attached to a non-traded good or factor? The NCAER survey took the stand that the labor used in non-monetized forms of investment was indeed a traded 'factor and imputed the prevailing wage in the village to family labor inputs. The assumed presence of a dual labor market in rural areas makes this a difficult proposition to accept. The problems with this 28 measure notwithstanding, it is also unclear as to whether.NCAER imputed separate 'costs' to female and child labor. If male wages were imputed to this labor, then the NCAER figures might be over-estimates since a fair amount of labor in NMI is likely to be of female and child variety. Also, wages vary considerably by season, and it is suspected that most of the NMI occurs during the slack season--a period for which the oppor- tunity cost of labor is not easily defined. These caveats suggest that the NCAER data estimates of NXI should be viewed as merely being suggestive of possible orders of magnitude. It should also be noted that estimates of MI are of a 'gross' kind and not necessarily indicative of net capital formation. A subjective judgement needs to be made as to what proportion of this investment is of a replacement nature. Given that the bulk of NKI represents the use of labor in 'land improvement' and 'housing ecstruction,' the replacement component of this investment is likely to be high. Estimates of NMI, based on NCAER data, are presented in Table 5(b) The share of non-monetized investment in total income for all farm house- holds is 0.6 percent--the share in investment and savings is about 4 percent. Almost 25 percent of the sampre population undertook non- monetized forms of investment and for these households the shares in income, investment and savings are 2 and 11 percent respectively. Classification of the data by farm size yields an interesting pattern to non-monetized forms of investment. The role of such invest- ment in total savings and investment appears to decline with farm size. As a proportion of savings, it is 15 percent for small farmers (<5 acres) and 2.5 percent for large farmers (,15 acres). The ratios follow a similar decline.for households with positive amounts of-MI. A plausible 29 TABLE 5(b) PATTERN OF NON-MONETIZED INVESTMENT - CULTIVATOR HOUSEHOLDS, 1970-71 Land -5 3< Land: 15 Land> 15 (acres) Investors -Investors Investors Investors All Ouly All Only All Only All Only Non-Monetized Inv. 28 115 20 86 27 112 47 165 (NMI) Grass Investment 760 1056 256 527 765 845 1699 2161 (i) Grass Savings 697 1024 130 232 645 778 1856 2603 (SdW Gross Income 4954 5549 2838 2917 5098 5375 8696 9833 (Y) Land Size (Acres) 10.8 11.3 2.7 2.8 9.3 9.1 28.7 27.5 NRI/I (%) 3.7 10.9 7.8 16.3 3.5 13.3 2.8 7.6 NMIS (%) 4.0 11.2 15.4 37.1 4.2 14.4 2.5 6.3 NM/Y () .56 2.1 .7 2.95 .53 2.08 .54 1.7 No. of Observations 1980 487 809 185 742 181 429 121 Notes: 1) See p. 27-28 of text for discussion of issues pertaining to the definition and valuation of non-monetized investment. 2) Households have been classified according to land o~nership. 3) 'Investors only' are those households with positive hon-monetized investment in 1970-71. 30 explanation for this pattern is as follows: Small farmers are the most likely to have an 'excess' of family labor, and this lower 'price' encourages a greater share in an investment that primarily uses this input. These results suggest that estimates of savings which exclude non- monetized forms of investment are likely to seriously underestimate the jgros savings of the rural population, and particularly its small farmer component. The global 'underestimation' is likely to be greater than the aggregate figures shown in the Table, since the NCAER survey over- sampled medium and high income households. Small farmers constitute 60 percent of the cultivating population (44 percent of the sample) and are unlikely to have incomes in the 'medium' and 'high' ranges. In conclusion, the pattern of non-monetized investment that emerges is: (a) It is a relatively small proportion of savings, income and investment for the aggregate sample of households, though these shares are somewhat higher for the investing population. Xb) Positive amounts of such investment are undertaken in equal proportion by all farm sizes in the popilation (20-25 percent), and (c) For small farmers (5 acres), NMI is a significant share of income, investment and savings. 31 SECTION 11.5 - ASSET LEVELS AND SAVING RATES Wealth plays an important role in most models of consumer behavior. In part III, asset levels (and their composition) are explicitly intro- duced into the computation of permanent income of a-household; in this section, for descriptive purposes, the relationship between asset levels and saving rates of cultivators is explored.1 The computation of the asset figure for households is limited by the availability of data. Information on only a selected list of assets is reported in the NCAER survey -- financial liabilities, irrigation equipment (tubewells, persian wheels, etc.), farm equipment (bullock carts, tractors, etc.), other farm assets (godown, cattle shed, etc.), 2 livestock and housing. Purchases of consumer durables are also re- ported, but no information is available on their stock. Indirect infor- mation is available on the value of land owned by the household. Thus, the net worth of a farm household can be approximated as Net Worth = Value of (Irrigation Equipment + Farm Equipment + Other Farm Assets + Livestock) - Financial Liabilities + Land Value No information is available on the assets of non-cultivator house- holds. 2 The reliability of reported values for houses was deemed question- able by NCAER. Consequently, these values have not been used in the computation of asset figures. If the (reasonable) assumption is made that the value of housing (and other assets which are ignored) is proportioned to other assets owned by the household, then differences in asset levels are still meaningful since they are only affected by a scale factor. 32 The values of all these assets, except land value, are reported for the beginning of 1970-71.1 The figure for land value was constructed by using the information on the amount of irrigated/unirrigated land owned by the household. The prevailing prices for these two types of land are available for the village in which a household resides. If irrigation and soil is assumed to be of homogenous quality within a village, then Land Value Price x Area + PriceUI x AreaUI where I, U1 represent irrigated and unirrigated land, respectively.2 Table 6 describes the relationship between saving rates (both definitions) and asset levels for cultivator households. Since occu- pational differences may affect the pattern of asset accumulation of a household, the sample has been divided into two groups on the basis of the primary source of income - self-employment income (mostly land cultivation) or wage income (agricultural and non-agricultural labor). Further, the groups have been separated into HYV and non-EYV cultivators, 1 1If "missing values" were reported for any of these assets, .a zero value was assumed. 2 Though the assumption of homogenous soil quality within a village may be reasonable, the assumption regarding homogenous irrigation is questionable. Farms can vary radically in the method of irrigation used -- wells, tube wells, canals, etc. - and these differences are reflected in land prices. Due to lack of appropriate data, the land values are regrettably not adjusted for irrigation quality differences. 33 TABLE 6 SAVING RATES (%) AND ASSET DISTRIBUTION (Rs.), CULTIVATORS, 1970-71 Asset Distribution of: Self-Employment Households Wages Households <2500 2500- 5000- 20000- >50,000 <2500 2500- >5000 5000 20000 50000 5000 All Households: Income, Y 1765 2567 4299 7739 14,233 1957 2527 5363 Saving Rate, SdW/y -1.6 5.4 10.1 17.4 27.0 3.0 -0.9 12.1 Saving Rate, Sy-c/y 5.6 11.5 -2.0 32.3 43.7 10.8 8.4 20.5 No. of Observations 141 268 796 400 107 109 54 67 Family Size 5.2 6.3 7.5 8.5 10.5 6.0 8.0 9.0 HYV Households: Income, Y 2249 2670 4225 8386 16,053 3558 3411 7088 Saving Rate, SdW/y -1.1 6.0 8.4 19.8 30.3 12.5 1.4 15.9 Saving Rate, S 11.7 8.9 21.1 35.1 48.0 21.6 15.5 21.6 No. of Observations 16. 41 254 208 69 12 8 14 Family Size 6.0 6.4 7.3 6.3 10.2 6.7 8.9 10.4 Non-HYV Households: Income, Y 1703 2549 4335 7038 10,928 1759 2373 4855 Saving Rate, -1.7 5.3 10.9 14.2 18.1 675 -1.5 10.8 Saving Rate, S 4.6 12.0 22.4 28.7 32.3 8.04 6.6 20.3 No of Observations 125 227 542 192 38 97 46 53 Family Size 5.1 6.3 7.6 8.9 10.9 6.0 7.9 8.5 Note: 1) For method of calculation of assets, see page 31of text. 2) Households have been classified by the primr source of income-- self-employment (mos tly land cultivation) or wages (agricultural and non-agricultural). 34 according to the status reported in 70-71. The results show a consistent pattern. Asset levels, and saving 1 rates, increase with measured income for the group in 1970-71. The S definition shows significantly higher rates than the S definition Y-C dw 2 of savings. Regarding the HYV- - non-HYV distinction, small samples prevent a clear assessment, but the general tendency is for HYV house- holds to save at a higher rate. Table 6 has only been presented for descriptive purposes, and the results are not analyzed in this section. Various parts of the paper discuss the subjects touched here. Part III incorporates asset levels and their composition into a definition of permanent income; Section IV.1 assesses the impact of HYV cultivation (investment opportunities) on savings and Section IV.2 discusses the effect of sources of income on saving rates. If transitory incomes can be assumed to' cancel out for each group, then the results indicate that saving rates increase with permanent income. 2See Section 11.3 for a discussion of these differences. 35 Section III.1 - The Savings - Income Connection The key question to be examined in this part of the paper is the rela- tionship between saving rates and income - are the two independent or do savings rates increase with income? A related question is the identification of what is the appropriate reference period for a household - does it base its consumption behavior on current income or on some assessment of lifetime or permanent income? The answers to these questions are crucial for an understanding of household saving behavior and for the design of policies on growth and income distribution. If current income is the prime determinant of consurption behavior, then 1 saving rates and current income are expected to be positiwely related. On this view (variously referred to as the Keynesian consumpton hypothesis or the absolute income hypothesis), the marginal propensity to. save (MPS) may be constant, but the average propensity to save (APS) increases with current in- come. Though this hypothesis appears to be consistent wimA cross-section data for the U.S., Kuznets' finding that the long run savfmgs ratio for the U.S. had stayed relatively constant (despite a secular increase in income) brought into doubt the validity of the Keynesian consumption function which would have predicted an increase in the savings ratio. Ne theories of the determinants of consumption were formulated in the post-ar period and their primary purpose was to reconcile the stylized facts of saving; i.e., a constant long run savings ratio and cross-section evidemce that the rich This notion receives support from both Fisher "in gmneral, it may be said that, other things being equal, the smaller the income, the higher the preference for present over future income," (1939, p. 72), and Keynes "men are disposed , as a rule and on 'the average, to increase tteir consumption as their income increases, but not by as much as the increase in their income" (1936, p. 96). 36 save more. Amongst the most prominent of the new theories were the perma- nent income hypothesis (PIH) of Friedman (1957) and the life-cycle hypothesis (LCH) of Ando-Modigliani-Brumberg. (See Modigliani-Brumberg (1954), M4odi- gliani-Ando (1965)). What relevance do these theories have to the economies of the less de- veloped countries (LDC's)? Insofar as economic theory is universally appli- cable, there is no reason why theories developed to explain consumption be- havior in the West should also not apply to households in rural India. Re- garding the LCH, Modigliani demurs on the issue of applicability by stating that "the life cycle model does not purport to represent a universal theory of individual and aggregate saving formation and wealth holding, but is in- stead basically designed to apply to private capitalistic economies in which at least the bulk of income, consumption, and accumulation transactions oc- cur through markets" (.1966, p. 215). The caveat notwithstanding, Modigliani and others do test the LCH with aggregate data from LDC's.1 By contrast, Friedman is more direct about the connection: "Acceptance of the permanent income hypothesis necessarily has implications for any problem of economic understanding or policy in which the determination of savings plays a signi- ficant role" (1957, p. 233) and again, "acceptance of the permanent income hypothesis removes both the direct and this particular indirect connection between low real income and a low savings ratio (of the LDC's). According to it, the savings ratio is independent of the level of income" (ibid., p. 234). Though differences exist in the structures of the two theories, they are nevertheless alike in (a) using a multi-period framework to postulate that houqeholds do not determine their consumption on the basis of current Kelley-Williamson (1968) and Ramanathan (1969) use household data from Indonesia and India to test the'life cycle hypothesis. 37 income, but rather adjust their consumption pattern to a long run or . permanent" income stream and (b) predicting that the household saving rate is essentially independent of the level of (permanent) income. The proposition that household saving rates are independent of the level of permanent income, is an important result of both the life-cycle and permanent income theories of consumer behavior. It shall hereafter be re- ferred to as the "independence proposition". The development literature has emphasized the low level of income of some LDC's as a cause for their low savings rate. The independence proposition explicitly removes any connection between the two; a low level of income might imply a low savings level, but not a low savings rate. At the macro level, the proposition has strong im- plications for policies related to the redistribution of income. If the savings rate is independent of the level of permanent income, then permanent income transfers within the population should not affect the overall savings rate. The testing of the independence proposition is therefore important from both a theoretical and policy point of view. Before this proposition is tested for households in rural India, its 1 theoretical derivation needs to be examined. The result follows from the assumption of inter-temporal utility maximization on the part of a consuming 2 unit. Essentially, the consumer maximizes utility, U, subject to the con- straint that discounted life-time consumption, C, be equal to wealth, W. Though bequests, K, are not present in the strict version of the life cycle model, they can easily be incorporated. Formally, if U(c) and U(k) repre- sent the utility from consumption and bequests, C(t) is consumption at a period of time, p the time preference discount rate and r the rate of in- 1The two theories - LCH and PIH - use different empirical methods but employ a common theoretical model of consumer behavior. 2 The following discussion is based on Blinder (1974, 1976), who dis- cusses the theories in detail. o........ . . 38 terest, then the problem is to maximize (1) U(C(t))e-ptdt + U(K), subject to T -rt -rt (2) SC(t)e + Kt e = W. The formal solution to this problem is the subject of optimal control 1 theory, and not presented here. Maximization yields the result that the 2 time rate of change of consumption, C, is given by (3) C (r -P) U, (c) /-U" (c).- Further results about the pattern of consumption can be obtained if an explicit form for the utility function is assumed. As Blinder has shown, the Friedman, Ando-Modigliani result that consumption is proportional to lifetime wealth is based on a special formulation of the utility function: T -6 -Pt 1-a C4) U CCt ) 1+bK, (4 ) U, = e b > 0 1-6 1-6 where 6 is the elasticity of the marginal utility of consum,.tion, a the elasticity of the marginal utility of bequests, and b a taste parameter. If bequests are allowed, then for proportionality to hold, 6 must be equal to 3. In the general case, the optimal consumption pattern is given by the fol- lowing equations: (5) C(t) = C e(r-p)t/6 or C = egt g (t P) o o 6 -rT (6) C = 0 (r,p,6,T)(W - Ke ) o 1 See Dorfman (1969), Blinder (1974). 2If it is assumed that U'(c) > 0,.U"(c) < 0, then this result shows that for consumption to grow through time, r > p. If r < p, the consumer can increase his utility by borrowing trom the future. Given a budget constraint, this implies that consumption will have to decline with time. 3Note that for 6=1 this utility function reduces to a popular form in the consumption literature; namely, U(c) = log C. 39 (7) 0 e dt] = a constant; (8) K = (berT)l/5 C 0 O' If bequests are ignored (K = 0), then equation (5) reduces to C(t) OW.eg' (9) or C(t) = O'W, and 0' = f(r, p, 6, T) Equation (9) is the Friedman and Ando-Modigliani resmilt that consumption at any instant in time is proportional to the stock of wealth of a household, W. Further, the constant of proportionality is independent of W; hence, the "independence proposition". Equation (9) can readily be c=nverted into a re- lationship involving consumption and permanent income, Y by noting that Y = rW where r is the appropriate rate of interest. Henme, p C(t) = O'W = 0'rW = qrW = kY (10) or C = kY p P which is the Friedman formulation with C the permanent flow of consumption, P_ and! k the average propensity to consume. (k, of course, depends on factors like taste, interest rate, age, but not Y ). P The preceding results have been derived under the assumption that bequests, K, are zero. If bequests are allowed to be positive, then consumption will only be proportional to wealth (or the average propensity to consume (APC) in- 1 dependent of wealth) if 6 = S . Thus, the independence proposition is ne- gated if 6 # a. The interpretation of the condition 6 # a is that if 5 > 6 (or the elasticity of the marginal utility from bequests eceeds the elasti- A plausible rationaie lor this condiLion is the assumption that a hos- hold views the consumption of its heirs as its "own".. Em other words, th time horizon for a consumption unit is infinity. 40 city of the marginal utility of consump-tion), consumption is a luxury good and the propensity to consume increases with wealth. If 8 < 6, and bequests are the luxury good, the propensity to consume declines with wealth. Theoretical consideration of bequests and the form of the utility function provide one reason for not expecting APC to be independent of wealth (or Y ). A "practical" reason for questioning the independence proposition is offered by the presence of households living under subsistence conditions. The existence of households at or near a subsistence level of living (biological or social) means that these households are constrained to consume all their permanent income, i.e., APC = k = 1. The Friedman formulation that k is independent of Y may hold beyond P subsistence levels. If subsistence households are included, consumption analysis should show that the rate of consumption increases from some level k to I as permanent incomes are steadily reduced to a subsistence level. If a subsistence level is identified, then the consumption behavior of non-subsistence households can be tested. High income households were oversampled in the NCAER survey; thus, the data contains an adequate number of households across a wide range of income. The problems of definition and measurement aside, the data are thus rich in their potential to test the independence proposition. Given that there is a positive overall savings rate for the rural population, the prior expectation is that there Zellner (1960) discusses this weakness, and Musgrove (1974) incorporates these considerations into a modified permanent income model. 41 are zero or negative savings (or consumption is equal to income) at the subsistence level, with perhaps a unitary elasticity of savings (and consumption) for high income households. But before the hypothesis can be tested, there is the difficult question of the measurement of permanent income/wealth. This question is taken up in the next section, and Section 111.3 uses the constructed permanent income measures to test their 1 relationship with saving rates. The life-cycle theory of consumer behavior is not directly tested in this paper (i.e., testing for relationships between age, earnings, family size and saving rates). This is because the conditions necessary for a "strict" interpretation of the life cycle model are not present in rural India. There is a suspected lack of an age pattern to-earnings in unskilled jobs and the existence of joint families (rather than the nuclear families of the LCH household) probably means a different saving pattern (and composition) than that dictated by considerations of "saving for retirement". 42 Section III.Z - Permanent Income - Definition and Measurement The previous section outlined the theoretical basis for the permanent income hypothiesis. Briefly, it was shown that, under certain conditions, permanent consumption, C , was proportional to wealth, p W, and that the factor of proportionality was independent of W; i.e., C = qW, where q is a function of tastes, interest rates, physical p assets, etc. but not W. If W were known, then the proportionality hypothesis could be tested. Though direct information on the physical assets owned by a household may be known, the perceived capitalized value of future returns to labor is not. The determination of an explicit measure for permanent income, Y , essentially boils down to the problem of arriving at an approximation of the future labor earnings of a household. This section is concerned with the estimation of Y . The longitudinal p nature of the NCAER data is exploited in this section to yield two different estimates of permanent income: (i) a weighted average of incomes and (ii) an earnings function modified to allow for permanent unobserved differences amongst individuals. It is hoped that these measures are more accurate in their representation of permanent income than would have been possible with one year data III. 2(a) - Permanent Income - A Weighted Average If future incomes of a household were known, then its wealth at time 0 is (11) 1 Y ert dt, t where r is the discount rate and Y t is income at time t. Future incomes 43 are, however, unmeasurable. The. NCAER data has information on incomes for the 'pst three years. With. certain heroic assumptions, a measure of permanent income can be extracted from the survey data. A generalized formula for any method that uses past incomes to construct a measure for permanent income is (12). Y = W Y + W Y + W Y + ...+ W Y, p 0o 1 -1 2 -2 n n where W are the weights and Y the observed incomes. How are these weights specified? And how does this formulation of Y relate to the P, theoretical measure Y = rW, where W is as defined in equation 11? The next few pages are concerned with answering these questions. In his original formulation of the PIH, Friedman (1957) used aggregate time series data for the U.S. to estimate permanent income. Using an income expectations approach, Friedman derived the following formula for estimating (13) Y T e(a)(t-T)Y (t) dt, where a is the trend rate of growth in past incomes, Y (t) is measured M income at time t, and a is the adjustment coefficient in the equation dy (14) =tp (Yn(t) - y (t)) The weighting pattern yielded by this method is at (a-a) t (15) W= e or Wt =e As Friedman readily admits, this method of computing Y requires the p. interpretation that Y is the."ixpected or predicted value of current P income' (1957, p. 143) rather than an estimate of permanent income. In his later article, (1963), Friedman rejects the income expectations 44 model:and offers an alternative. rationale for constructing weights. This method, which Friedman states- is applicable to bothiindividual and aggregate data, is to "regard individuals as taking their past experience, adjusted for trend, as the best single estimate of their likely future experience ..... ( these assumptions, the estimate of current permanent income can be regarded as -made in two steps. First, each past measured income is adjusted to the value it would have had if wealth then had been equal to its wealth now. On the assumptions just made, the adjusted a (T-TO) value of ym(T') is ea yM (T'). Second, these adjusted values are used as estimates of the future receipts to be expected, the discounted value of which constitutes present wealth," (1963, p. 22-23, italics mine). The estimate of Y yielded by this method (16) y = 1 rp a)_Tym(t)dt, pp which is exactly the same as the one yielded by the incC=a expectations model, equation 13, with a replaced by the discount rate r. Though novel in its approach, there are two questionable assumptions about the construction of this "direct" measure of permanent income. One is the assumption that the best estimate of income t' in the future is one obtained t' in the past. This is particularly questionable for households whose major source of income is labor earnings. It is doubtful whether'past experience is a good indicator of future income for new entrants into the labor force. Explicit consideration of human capital accumulation on the part of these households will yield a pattern of expected future income which.is radically different from past incomes. Nevertheless, the Friedman assumption, though questionable for urban U.S. households, is quite reasonable for farm households in rural India. 45 For these households, the- best predictor of f.ture incanes. is likely to be some function of past:incomes. One can of course argue about the*best function, but the'fact'remains that any function will have to be somewhat arbitrary. Thus, the simple assumption that incomes in the future are a mirror image of the past, adjusted for trend, is acceptable. However, there is a more serious problem associated with Friedman's result-equation 16. It is that, given his assumptions, his calculation of weights is wrong. If it is assumed that the trend rate of growth a represents both the growth experienced in the past and the growth expected in the future (a reasonable and almost 'mandatoryt assumption), the the computation of wealth is as follows: (a) Expected income in the future, given that past incomes have been growing at the rate a is ym(+T') = ea (T )y (-T (b) If expected growth in the future is a', then ym (+T) = a'(T-T') m(+T') = (a+a)(T-T') Y(-T) M If '= a, then ym(+T') = e2a(T-T') (17) ...and permanent income, Yp = r e(r-2a)(t)ym(t) dt. This equation differs from Friedman's formula in that a is replaced by 2a. Thus, the "new" weighting pattern given by Friedman (equation 16) does not reduce to the pattern yielded by an income expectations model (equation 13). The only situation for which the results are identical is if it is (unreasonably) assumed that income in the future is not expected to grow, but that income.in the past has been growing at a rate a. 46 It was asserted earlier that the mirror image formula for imputing future incomes, adjusted for trend, was reasonable. for farm houeholds in India. Consequently, the discrete version of equation 17 is used in this paper to estimate permanent income. (18 r 2iI+) .i~o i (1+r) The NCAER data has incomes for only three years, whereas the above formula needs estimates for the lifetime of the household. As an approximation, the raw weights, unadjusted for trend, are normalized to equal 1. Thus, the equation used is 2 22 i=-0 Y .(1+a ) (1) (1-+r ) p2 l 1r i=0 Two problems still remain-what value of a, and what value of r, should be used to compute Y ? p The high yielding varieties of foodgrains were introduced in Indian agriculture in the -mid 60's. The NCAER data covers the period 1968-69 to 1970-71. The average rate of growth of incomes during the sample period was 3.5%. If the rate of growth in the future is expected to be that realized in the late 60's, then a common average a of 3.5% can be assumed.Obviously, the experienced and expected growth in income for individual households is expected to vary radically from a; however, lack of a reliable indicator of individual a dictates that a common a be assumed. 47' The discount rate r can also vary between households. Again, lack of prior information dictates that a common r be assumed. Table 7 shows the weights which emerge if r is allowed to vary between 10 and 90%. The elasticity of consumption with. respect to income is estimated for r=10, 35 and 75% (Section III.3Ca)). The results show that the choice of weights makes very little difference to the consumption elasticity. Thus, only the permanent income measure yielded by the assumption r=351, Yp35, is used in the rest of the paper. III.2(b) - Permanent Income - A Modified Earnings Function The 'earnings function' method is often used with cross-section data to estimate the permanent income of a household. Essentially, this method requires that the determinants of income (X) be identified and a regression of the form (20) Y X$ + e be estimated. (Y is measured income, X the factors determining income, and e an error term). The predicted values (21) Y=X0:....... 1 are then taken to be estimates of permanent income, Y, and Y-Y is taken to be an estimate of transitory income. Though plausible, there is a serious problem with. this approach.. Deviations from 'mean' levels, rather than identifying the assumed transitory components of income might reflect just the opposite; i.e., differences in permanent levels.2 'Friend-Corckett (1973), who use-this method, call Y "normal" income. 2 P An analogous problem is encountered when one uses education as one of the variables in an earnings function. Unless one has a valid measure for ability in such an equation, the residuals of the regression will reflect ability differences which are permanent in nature. 刊 49 Thus, the-residuals from an ear, -,ings functi.on may not bwe- distributed randomly around the-mean (permanent) 1-evels. If the:permanent component of tFLese residuals could be isolated, then one would have a -more. accurate representation of Y than that P yielded by(21). If incomes data is available for an ind:Evidual for more than one year (as in NCAER) and other longitudinal data, then, under certain assumptions, the systematic part of the error can be separated from its transitory component. A 'simple variance components' model described in the next few pages, is used to isolate unobserved permanent differences in the earnings capacity of individuals. Though variants of this method have been used before in the estimation of wealth 1 (permanent income), it is believed that this is the first time this procedure has been used to test the savings/consumption behavior of individuals. In its most general (linear) form, the income of an individual in year t may be expressed as k (22) Y X. it + E: i + E t + C it Z 3 it J=l where Y is measured income of ivdividual i in year t, the O's are it parameters, the X's the determinants of income, and s' E + E + e it t i it is a composite error term. E is an error specific to the time period, t e* is an error specific to each individual and sit' is an error with zero expectation. .... ...... 1 See Gordon (1976), and Lilliard (1975). 50 It is the presence of e which.causes traditional earnings function methods to give unreliable estimates of permanent income. The sit are determinants of income specific to an individual; -moreover, they are assumed to be constant with time. In other words, e -most likely captures factors like education, ability, inherent soil quality of land, etc. If the i can be identified, =i individual effects), they are no longer a part of the error term, and equation (2) becomes k (23) Y = i + Z . X. + s + S. . j=l 3 J t it Individual time effects can also be incorporated into the estimation by use of dummy terms for the individual years. Thus, equations (23) becomes k (24) Y = i + .Z a.. + d + s . . -t .j=1 3J3it t it How can the y 's be estimated? If equation (24) is rewritten for the average values, T and 3, for the three years, and if it is further assumed that the average impact of dt is zero for the three years, equation (24) becomes k (25)i i it . . . . . . . The term yi can be eliminated, and the 's identified, by estimating an equation like (26), k (26) Yi - = 1 -i + d + dlD1 d D +( -C itJlj jit ji o 1 1 Z it and assuming that the error term is randomly distributed with zero mean. (d is a constant term in the regression., and D and D are dummy variables representing years 1 and 2, respectively). 51 The.consistent estimates*of.S., obtained frPm equation C261, can now be inserted.in equation (25), to obtain consistent estimates of yi; i.e., ' = Y i E$ I . This value of y can now Ea inserted in equation 23 to obtain an estimate of the permanent income of the household; i.e., (27) Y i+EaX This estimate of Y is not without some strong assumptions. In particular Rx it assumes that all sources of permanent income (X's) have been identified, and that the pattern of X's observed for year t are expected to prevail in the future; i.e., no "windfall" gains can be had by just changing the composition of assets. Estimation of Y for Farm Households pi What sources of permanent income can be identified? Income is a return to labor and capital assets, and the latter can be decomposed into land and other assets. Thus, as an approximation, (28) Y = rH + rKK + wL where H, K and L represent land, capital and labor and r., rK and w the returns from these assets. Information on some physical assets owned by a household (irrigation, farm equipment, other farm assets and livestock) are available for the beginning of the third year of the survey, 1970-71. Since information is available on the investment in these assets for the first 2 years, the value of assets owned for these years can be approximated. Thus, 1Equation (27) differs from the traditional earnings function (equation (22)) in that individual effects, yi, have explicitly been allowed. 52 K' = Value of CIrrigation Equipment + Farm Equipment + Other .Farm Assets + Liyestock.) 1,2 can be estimated for eacfh of te.three years. Estimates for the value of land were obtained as follows: R P xA + P .x A t i i,t ui ui,t where P is PU represent prices of irrigated and unirrigated land 3 prevailing in the village , and A, A . are the amounts of irrigated i,t' ui,t and unirrigated land owned by the household.4'5 A proxy for the earning units in a household, L, is slightly more difficult. The NCAER survey supplies information on the number of "earners" in a household; however, this definition of earners is restrictive. A household member was defined as an earner if he received any outside eai:nings in an agricultural year. Even very temporary employment would therefore classify a person as an earner. This is a crude index of potential earnings and certainly not reliable for large farms which are unlikely to have members with outside employment. The non-earners Some assets which also affect income have obviously been ignored by this procedure. This should not affect the estimates of Y obtained if it is assumed that the value of "ignored" assets is proportional to Kt, and that they affect incomes in an identical manner. 2 Irrigation equipment does not have a value independent of the value of land. If the latter is also being used to estimate permanent income, theti, double counting is avoided by ignoring irrigation equipmant in the calculation of Kt. This procedure- was followed in this paper. (Inclusion of irrigation equipment in Kt did not make much difference to the values of Y ). It should be noted that if the'level of irrigation'asseats are a proxy for the quality of irrigation, then such. differences afe already incorporated in the model via Y.. 3 It is assumed that the prices of identical quality land were not changing during the three years of the survey. 4 Obviously, this procedure innora diff"rences in the qualit"M of iirigation. Lack of appropriate data meant that these differences could not be explicitly incorporated into the analysis. If the quality of irrigation does not change through time, then these differences should be captured by the "individual effects" term, y . It should be mentioned that the land ownership figures for the first year (1968-69) are suspect. This information was made available 53 in a gamily were classified.as family workers if they worked for for any period on the-farm. On-farm work is also likely to vary with farm size. ThougF variables representing earners and family workers could be entered separately, into the analysis, these units are not homogeneous in earning capacty-. Furthermore, these variables ignore the potential earnings of young family members not presently in the labor force. One alternative to either earners or family workers is to use family size, F, as a proxy for L. Its advantage is that variations in it may indeed capture differences in earning members across families (The simple correlation coefficient between family size and income for 1970-71 was .403). The most important drawback with the use of F is that it gives equal earning weight to all members of the family. Thus, any proxy for L-earners, family workers, and family size-has its drawbacks. Given the constraints of data availability, the disadvantages associated with family size seem to be the least. Consequently, it was used as a proxy for L in the estimation of Y . p in a private communication from NCAER. -Rather than eliminate any house- holds with inconsistent land ownership data, the ownership figure for the third year of the survey, 1970-71, was imposed for the other two years. This procedure was based on NCAER's assertion that the figures for land ownership are most accurate for 1970-71. Since only households that had zero investment in land for each of the three years were selected for analysis (see Appendix I) they must have had the same amount of land for all the years. Consequently, the: imputation of land ownership figures of the third year for 1968-69 and 1969-70 were justified. The net area irrigated for the first and second years had also to be accordingly adjusted. The procedure was as. follows: (i) the proportion of land reported to be irrigated in the first (and second) year was imputed to the land owned in the third year; and (ii) if this procedure resulted in a greater acreage being irrigated in the first two years, then the irrigated acreage of the third year was assumed. 54 Variables representing weather conditions, Mt, and the new teach- nology, were also used in. the..detexmination of Y.. -M was introduced in binary form - 0 if conditions.were average, and 1, Below- average. The percentage of area cultivated witb-the new- technology, BYV,, was used -1 to represent the technology level for each household. The estimate of Y obtained by using the HYV level of 1970-71 represents a strong P, assumption, namely, that households expect to stay at the 1970-71 technology level. The X's having been identified, equation 24 can now be written as, (30) Yit i 1 2K + a3L + BYVt + A + e. Equation 30 is estimated for the entire sample of farm households (1980) that were selected for analysis.2 ThQugh some may find the eStimption of an all-India earnings function questionable, it should be pointed out that if major determinants of income have been adequately identified, estimation of(30)is defensible. In addition, an objection to(30)is made even less relevant by the presence of y - permanent differences amongst individuals and constant regional differences are captured by this term, as well as differences in soil quality, quality of irrigation etc. The results for equation (30), estimated with, and without, the individual effects term for 1970 -71 data, are shown in Column 2 and 3 of Table 8. Column 1 represents the results for equation (26), and 1 Again, lackof data prevented adjustment for the sophistication of adoption. 2 This method of estimating Y resulted in negative values of Y for p P a few households. Rather than eliminate these households, average three year income was substituted forY P. 55 TABLE 8 ESTIMATION OF PERMANENT INCOME MODIFIED EARNINGS FUNCT'I=ON, 1970-71 DEVIATION FROM MODIFIED EARNINGS TRADITIONAL EARNINGS VARIABLE MEANS FUNCTION FUNCTION 1970-1971 1970-1971 Constant -49.2 87.6 615.0 (1.0) (.85) (4.1) Land Value, H .18 .175 .09 (11.2) (67.3) (23.8) Capital Assets, K .33 .37 .295 (8.1) (29.2) (15.6) Family Labor, L 188.1 168.7 270.6 (family size) (12.3) (12.0) (152) Technology, HYV 5.6 4.4 14.9 (5.3) (2.5) (5.7) Weather, M -28.8 -198.9 -245.2 (.37) (1.9) (1.5) D (=1, first year) 125.0 (1.7) D2(=1, second yearl 22.7 (.34) -2 .006 .858 .560 Standard Error .2026 .1950 .2890 No. of Observa- tions 5949 1980 1980 Notes: 1) For definition of variables; see p. 51 -54. cof text. 56 columns C2) and (3) show the nature of. changei the I'~ -wheindvua effects are. compounded in thie- error term. Even without the inclusion of the- model is able to exlain 56% of the. variance- in annual incomes-a large fraction for cross sect:Lon data. Te -inclusion of Yi. (colulmn 2) increases radically the. explanatory~ power of the-model from 56% to 86%. Thus, identification of Y.. not only yields unbiased 3- and more efficient estimates for ~,but also allows one to measure transitory income, Y-Y1, with greater accuracy than would have been possible with the traditional model. M~easures of Permanent Income - Concluding Remarks This section has outlined two conceptually different approaches to the estimation of permanent income-- (i) Yp,a weighted average of past incomes, with i reflecting the discount rate used by the household, and (ii) Y,an earnings function estimate with inclusion of individual differences. Table 9 shows the correlations amongst the different measures of permanent income and the two measures of savings - S dWand S YC. The correlations amongst the Y.measures is very high (>.99), and that between Y and Y~ ranges from .98 to .993. These correlations suggest that the definition of Y P,is unlikely to make a difference in the results regarding consumption (savings) behavior. The next section compares consumption elasticities for three different measures of Y i-10, 33 and 75%, and Y .All Y .-measures yield similar results, Given the PX P3. correlations reported in'Tabie 9 and the. results of the. next section, it 'was.decided to use. only Y p5and Y RXin the. subsequent analysis. The discount rate r=35% was chosen.because (a) it is intermediate between low (10%) and high (75%) rates of discount and (b) it is close to a discount rate which is popular in the consumption literature - r =33 1/3%. 57 TABLE 9 CORRELATION MATRIX OF SAVINGS AND PERMANENT INCOME MEASURES, CULTIVATORS, 1970-71 S Sy-C YpCQ y y y Ypx dW plo P35 p75 SdW 1.00 Sy-C .344 1.00 YpCG .714 .903 1.00 y .607 .700 .887 1.00 plo y .632 .731 .915 .997 1.00 p 35 y .657 .763 ,943 .988 .996 1.00 p75 Ypx .620 .819 .894 .993 .992 .985 1.00 Notes: 1)-SdW represents change in net worth and Sy-C the residual estimate of savings. See p. 13-20. of the text for a dis- cussion and comparison of these.measures of savings. 2) - Subscripts to Y indicate the discount rate used in esti- p mating the permanent income of a household. See p. 42-57 of the text for a definition and discussion of these measures. Yp- is the measured household income for the 1970- 71 agricultural year. 58 SECTION III. 3 Permanent Income. Measures - Consumption Function Results In this section, the. measures of permanent income developed in the last few pages are. used to test Yriedman's.contention that the. propensity to consume (save) is independent of the level of permanent income. It will be. convenient at this stage. to express formally the statistical elements of Friedman's hypothesis. It asserts that there is a systematic relationship between the permanent components of consumption (Cp) and income (Yp) (Equation 31); that both measured consumption and measured income can be broken up into their permanent and transitory components CT and Y (Equations 32 and 33), and that these transitory components are not correlated with each other or the permanent com- ponents (equation 34). (31) C = KY . . . . . . . . . . . . . . . . . . . P p (32) YY= 7 +Y7 . ................ (33) C =C + C p T. . . . . . . . . . . . . . . . . . (34) cov(YP,Y ) cov(C T,C ) cov(Y ,CT) 0 If these assumptions are satisfied, and if equations of the form leg C = a + b. log Y p. are estimated for the different measures of Y then one should obtain p 2 the "independence propostion" or "unitary elasticity" result that b=1. The variables in these equations can represent arithmetic or logarthirmic values. Both. types of relationships are explored by Friedman. 2 The popular result that the cross-section elasticity of consumption with respect to measured income, b,. is less than 1 follows immediately if one notes that when measured income is used, Yp is measured with error, Y t. 59 The results of Table 10 CColumn 11 indipate- that, :regardless of the-measure. of permanent income, the consumption elasticity, T, is significantly les than 1.. This would seem to reject Yriedman"s contention that consumption is-.proportional to permane=t income. But is an aggregative relationship as estimated in column I justified for rural households? It was pointed out earlier that subsistence level considerations argue against such a relationship - sbistence households are constrained to consume all their income. Thus thee is a dependence between k and Yp. Another argument against an aggregative relationship is the (presumed) dependence between the interest rate on borrowed funds, the rate of return on investment and permanent income. The assumptions of the PIK imply that households can borrvw and lend freely at given interest rates; these rates, according to the Iypothesis, may vary but not systematically with. the level of pexmanent income. It is doubtful that these conditions are satisfied in rural India, and Bhalla's (1975) analysis shows that these rates vary systematically with farm size and adoption of technology, i.e. factors related to permanent income. Thus, an aggregatiYe relationship, as estimated in column 1, is not within the "spirit" of.the PIK. A valid test of the PIH can still be constructed if the dependence between the average propensity to consume, k, and Yp is "purged" from the data. If households are- grouped according to the capital market and investment conditions that they face, th consumption elasticities could be estimatal for each.group. Any attempt at such-a grouping is bound to be imperfect; consequently, a simple (and reasonable) criterion was adopted - households were classified by the average level of per 60 1 capita income in the preceding three years, Y The groups chosen apc. 2' for analysis are suhsistence Rs. 500, intermediate Rs. 500 --Rs. 1500 and rich.> Rs. 1500 Cincomes in per capital termsi. The consumption function results for the different groups are shown in columns 2-4, Table- 10. All Yp -measures yield qualitatively similar results. The- consumption elasticity is not significantly different than 1 for the rich households. At subsistence levels, where households are constrained to consume their entire income, one would expect b to be close to 1. Though higher than the elasticity for the intermediate range, b is less than 1 for the poor households. These results indicate that rather than the income 6lasticity of consumption being constant and equal to 1, it appears to follow a non-linear pattern. It is close to 1 for incomes near the subsistence level, decreases for intermediate incomes and then rises to 1 for rich households. Thus, the hypothesis of a constant, unitary elasticity is rejected by the data. lY a is defined to be EY /EF , where Y is measured income and apc t t t Ft the family size. An acceptable figure for the pove'rty (subsistence) level in Indian literature is Rs. 240 per capita annually, at 1960-61 prices. This is equal to Rs. 440 in 1970-71 prices. The. figure chosen in this paper Rs. 500 (or $70) can be taken to be a conservative definition of the subsistence level. 3Apart from classification of households by capital markat conditions, use of Y as a grouping variable has the additional advantage that the ape results of the consumption function can alternatively be. interpreted as a piecewise linear approximation to an inherently non linear relationship between C and Yp. (This under the assumption that Y is proportional to permanent income.) ape 4 It should be noted that even if the elasticity were equal to 1 for hold qroups, the aggregate elasticity may still be less than 1. This all house- would result if saving rates did depend on the level of income, an assumption explicitly rejected-by the PIR. 61 TABLE 10 ELASTICITIES OF CONSUMPTION WITH RESPECT TO PERMANENT INCOME - CULTIVATORS, 1970-71 Measure of Permanent Income All Yapc<500 500<Yapc<1500 Yapc>1500 Measured Income, Yp,= .80 .85 .79 .73 (114.1) (59.2) (49.6) (12.6) Permanent Income, Ypx .75 .81 .71 .84 (65.9) (27.4) (21.1) (7.3) Y 10 .75 .76 .69 .85 (66.0) (27.8) (20.9) (7.3) Y .77 .82 .78 .93 p35 (72.2) (31.9) (25.0) (8.6) Y .80 .89 .86 .98* p75 (80.6) (37.5) (30.5) (10.0) # of Observations 1980 915 940 125 Notes: 1) - The equations have been estimated according to the log formulation of the permanent income hypothesis i.e. log C a + b.log Ypi, where C is the "residual" estimate of consumption (Y-S ) per capita and Ypi is the parti6ular formulation of permanent income per capita used in the regression. See notes 1-2 of Table 8 for a definition of the variables. 2) - Yapc - Average per capita income of the households during the three years of the survey i.e. Yapc = ZY/Z , where Yt t represent aggregate income and family size during the three years of the survey. 3) - Figures in parentheses represent the t-statistics of the co- efficients. * refers to consumption elastidities no being different from 1 at the 5% level of confidence. 62 The. results of this section (along with. the result of Table 9 that correlation amongst the Yp.measures is very high - .9-91 indicate (a) that the consumption elasticity is significantly different from 1 for a large segment of tFue rural population; (b1 that earnings function and the weighted average method of constructing permanent income give similar results and Cc) that undue concern with selecting the proper discount rate may be unnecessary since differences in discount rates do not make much difference to the results. The rest of the paper is concerned with the saving behavior of rural households. Only the arithmetic form of Friedman's model can be used since savings often take on zero and negative values. Given the similarity amongst the different measures of Yp, only two measures of permanent income are used - Y and Y . . p35 px The results for Y 10 Y35 and Y are even more similar for pl0 p35p75 the arithmetic version of the. permanent income model.. The propensities to save yielded by the three measures are virtually identical. 63 SECTION III. 4(a) Savings Model - A Derivation In this section, the two measures of permanent income, Yp35 and Y are used to estimate a savings model. If the arithmetic version. px of Friedman's model is used, then the savings (S) of a household is derived as follows: C = kY by assumption, P p or C + C kY + C p T p T or C = kY + C p T or Y-C = S = Y-kY - C =Y + Y -kY -C p T p T p T (36) or S (1-k) Y + Y C In equation (36), (1-k) = k' is the average saving rate, which is equal to the marginal saving rate. The propensity to save out of tran- sitory income (Yt) is equal to 1. CT is transitory consumption, which is t usually part of the error term since information about its magnitude is usually not available. The "basic" savings function, estimated as in (36'), (36') S= + 1 p 2t +e can be used to test elements of the PIH, as well as to compare the PIH with the Keynesian theory of saving behavior. In particular, the result that a < o would support the Keynesian hypothesis and reject the contention of the PIH that saving ,rates are independent of the level of income. (0 M 0 corresponds to the result that the consumption elasticity is unity). 0 The magnitude of 2 indicates the proportion of transitory income saved; it should, according to PIH, be equal to 1. The Keynesian hypothesis, by 64 postulating the dependence of savings on current income, as.serts that the distinction between the sources of income. (permanent and transitoryl is irrelevant; thus, 0 should be eual.to . A "eaiht result of the PIL is that 52 is greater than 1 A variable that is often mentioned in theoretical analysis but usually ignored in empirical studies is transitory consumption. For an analysis based on grouped or aggregate data, the transitory com- ponent of consumption can be expected to average out to zero and the variable justifiably ignored. For an individual household, however, the transitory component of consumption (C T) can be expected to significantly affect its savings. If the assumption is made that households attempt to maintain a planned level of consumption (CP) then differences in C should be reflected fully in differences in T savings i.e., an X rupee iicrease in CT should lead to a X rupee decline in savings. Identification of C is a difficult matter. Surveys in which only income and savings are ascertained do not provide any information on this variable; and expenditure surveys have to explicitly ask about "accidental" or "transient" consumptionto identify this component. The NCAER data is unique in that it gathered data on savings, income and consumption expenditures. One question of the household survey recorded expenditures on "Marriages, religious functions, funerals and unexpected travel". In a book published by NCAER to accompany the data tapes, this question is listed as "large unexpected expenditures". It is doubtful that all the items listed above are "unexpected". The apparent discrepancy was cleared up in private conversation with NCAER. The item was meant to include large, transient expenditures, i.e., those expenditures which the household did not indulge in on a regular basis--hence, the term "unexpected expenditures". 6.5 The level of these expenditures, UT can be used as an approximation to C, though the objection can be raised that U' contains in it elements T T of planned consumption. An approximation to the permanent component of C in U is the average level of U for the past three years, UA if T T A this assumption is accepted, then C can be defined to be equal to T U - UA i.e., the difference between third year unexpected expenditures and average unexpected expenditures for the three years. Though plausible, use of this definition of C implies that it is likely to be measured with error. Thus, incorporation of CT in equation 36' yields (37) S= + 8 Y + a2 t+SC + u where part of the error, u, is the measurement error in transitory consumption. Estimation of equation (37) will yield inconsistent estimates for the coefficients; however, the efficiency of the estimates is likely to improve since more information is being incorporated into the model. The gain in efficiency was felt to be worth the price of inconsistency, and accordingly, equation (37) was used as the "basic" 1 savings model. A level regression of savings, estimated as in equation (37), results in errors whose variance increases with permanent income. If the introduction of C is found to be questionable, it should be mentioned that none of the results are qualitatively affected by its exclusion, though the variance in savings explained, -2, is always increased significantly by its inclusion. 66 If the assumption is made that the variance of the residaals is propor- 1 tional to the square of income , then the heteroscedast:brity present in 2 equation (37) can be corrected by estimating an equatiom of the form (38) S 80 Y C (38) + + T + 1 2Y Y 3Y P p p p The equation actually estimates was not (38) but aw-ariant of it which allows one to identify significant differences between the MPS Y p (81) and the MPS (82), namely, Y C t S o0 Y CT (39) - + 8+ B --+ a - (39 1 2Y 3Y P p p p In this equation, Y is measured income (= Y + Y ) and tfhe t-statistic on 1 indicates whether the marginal propensities to save are significantly different. This occurs under the assumption that transitory consumption (e in equation 36') is propOrtional to Y p 2It is this transformation that accounts for the relatively low R2 reported in Table 11. When these same equations were estimated in level form, the R2 were considerably increased. This should be kept in min if the appropriateness of the model is being judged by the level of R. What is being explained by (38) are not variations in level but rather the variations in saving rates. In an evaluation of cross-country data, Singh (1972) found that controlling for heteroscedasticity lowered the R2 from .95 to .57 for a simple Keynesian savings function. For the NCAER data, the corresponding figures for R2 are .59 and .19. 67 Section III. 4(b) Savings Model - Propensities to Savp- Tables 11 and 12 present the results for equation (39) for three different measures of permeanent income Yp and Y Measured income Y is equal to Y O for an infinite discount ratel*. TEw- results regarding the proportionality hypothesis are similar to the ones obtained with the logarithmic version of the model. Saving rates are not independent of the level of income at the aggregate level and for the low and medium ranges of income; but, for the rich households, (and the Yp35 definition of permanent income), the t-statistic on 3 is 1.9 - not significantly 0 different from zero at the 5% level of confidence. The MPS out of transitory income is significantly higher than the MPS out of permanent income for the aggregate sample; and for subsistence households. In neither case, however, is this propensity 2 equal to 1, as required by the PIH. For the intermediate group, 02 > 0 for only the Y definition. 2 An additibnal term, S Y , is introduced into the equations to test 4 P for non-linearities in the marginal propensity to save permanent income. This coefficient is significant and positive for the aggregate regression, and for households in the intermediate range of incomes. The sign of the quadratic term in income is "perverse" for the rich group - it is negative. Though significant at only the 10% level of confidence, 5 is significantly different from zero for the Y definition of 0 permanent income. However, the t-statistic - 2.4 - is smaller than for the other groups (8.0 for subsistence levels and 9.9 for intermediate incomes). 2This result 2 1 but B2 1 is in accord with most empirical studies of savings behavior. See Mayer (1972). -68 TABLE 11 SAVINGS MDDEL WITH ALTERNATIVE DEFINITIONS OF PERMANENT INCOME - CULTIVATORS, 1970-71; SAMPLE - ALL OBSERVATIONS: (N=1980 OBS.) Linear Version Q adratic Form Variable Y (=Y) Y p35 Y Ypx p P (=Y) Yp35 Ypx Constant -73.9 -74.1 -77.3 -48.7 -41.4 -45.4 (Bo) (24.6) (22.5) (23.2) (13.2) (9.5) (10.3) Permanent Income, Yp .22 .23 .23 .10 .07 .09 (B1 + B2) (27.4) (6.2) (4.2) (7.2) (11.7) (10.3) Transitory Income, YT .34 .30 .34 .30 (B2) (20.5) (20.9) (20.8) (21.5) Transitory Consumption, CT -.55 -.61 -.60 -.56 -.62 -.60 (15.7) (16.4) (15.9) (16.5) (17.1) (16.5) (B 3 2 -3 Yp X10 .094 .111 .104 (B4) (11.1) (11.1) (10.7) A2.3 .29 .34 .34 .33 .38 .37 Standard Error .1995 .1925 .1911 .1936 .1868 .1859 Notes: 1) The Linear version of the model is: S dW= Bo + BY + BY + B3C T+e where S w,Y , Y, C represent per capita savings (net worth esti- mate), permanent income, measured income and transitory consumption, respectively. (The2quadratic version of the model includes an ad- ditional term, B4Y ). 2) In order to correct for heteroscedasticity, the assumption was made that the variance of the residuals is proportional to the sqaure of permanent income. Consequently, the estimating form used was: SdW Bo0 B1 B 2Y B Y Y Y Ypi 3) The reported coefficient is (Bl+B2), and the reported t-statistic is the one associated with Bl. Note that in the linear model this t-statistic represents the significance of the difference in the marginal propensities to save permanent and transitory income. 69 [The MPS permanent income is given by (B1+B2), and B2 represents the MPS transitory income; Y=Y +Y T. 4) Figures in parentheses represent the absolute value of the t-statistics. 70 Table 12 Piecewise Approximation to Savings Mode'. - Cultivators, 197-71 Sample 0 < Yapc < 500 (N = 915 Obs,.) Linear Ver.--;-n QuadratLc Form Variable Ypgo Yp35 Ypx Yp. Tp35 Ypx Constant -53.5 -39.4 -41.6 -46.3 -26.4 -39.9 (12.3) (7.7) (8.0) (5.9) (2.4) (3.2) Permanent Income, Yp .15 .11 .12 .10 .02 .11 2) (9.7) (6.4) (3.6) (1.9) (3.0) (1.28) Transitory Income, Yt .25 .21 .25 .21 (9.8) (10.2) (9.8) (10.2) Transitory Consumption, C -.66 -.69 -.61 -.66 -.69 -.61 t 3 (11.7) (11,7) (10.7) (11.7) (11. 7) (10.7) 2 -4 Ypi x10 .083 .140 .20 (1.1) (1.33) (0.15) .22 .22 .21 .22 .22 .21 Standard Error .1993 .1864 .1818 .1993 .1863 .1819 Sample 500 < Yapc < 1500 (N = 940 Obs.) Constant -125.6 -131.6 -150.3 -51.2 -56.9 -95.7 (no) (13.6) (9.2) (9.9) (3.0) (1.6) (2.4) Permanent Income, Yp (01+2) .28 .30 .32 .05 .11 .18 (19.1) (2.6) (0.74) (1.1) (3.1) (1.7) Transitory Income, Yt .37 .34 .38 .34 (S2) (17.0) (17.3) (17.1) (17.3) Trans±tory Consumption, -.50 -.56 -.58 -.50 -.57 -.59 ct3) (11.5) (12.2) (12.2) (11.7) (12.3) (12.3) 2 -4 Ypi X10 .147 .108 .074 4) (5.18) (2.4) (1.5) R .23 .33 .33 .25 .33 .33 Standard Error .1872 .1817 .1824 .1847 .1812 .1823 Table 12 (cont.) Sample: Yapc > 1500 (N = 125 Obs.) Linear Version Quadratic Form Variable Yp Yp35 Ypx YpCG Yp35 Ypx Constant -363.4 -246.7 -311.6 -367.3 -473.3 -638.6 (So) (6.2) (1.9) (2.4) (4.4) (1.6) (2.1) Permanent Income, .51 .46 .49 .51 .66 .79 Yp (14.0) (1.6) (.99) (6.2) (.18) (.83) Transitory Income, .61 .57 .62 .58 yt (10.2) (11.0) (10.2) (11.1) Transitory Consump- -.63 -.85 -.86 -.64 -.86 -.85 tion, Ct (3.8) (4.7) (4.5) (3.G) (4.7) (4.4) 2 -4 Yp X10 -.001 -.037 -.057 ($4) (.06) (.84) (1.2) -2 R Standard Error .25 .49 .51 .24 .50 .51 .1848 .1864 .1891 .1856 .1867 .1887 Notes: 1) See notes, Table 11 for definition of variables, estimating form, and interpretation of coefficients. 2) Observations have been selected according to the vulue of Yapc- average income per capita. For definition, see note 2, Table 9. 72 (Y definLtion) the result is nevertheless interesting in that it is contrary to most assumptions about sayings behavior and different than the revealed tendency for the< Rs. 1500 group. Te.negative sign for a implies that the. marginal propensity to save decreases with.7.increases in permanent income. Is this result plausible? Actually, there are a number of reasons why one migE.t expect this "strange" result to occur. The households in the rich group (top 5% of the sample, and possibly top 1% of the rural population) might be indulging in conspicuous consumption, thus causing their rate of saving to fall. Another reason is the interaction of subsidized credit and investment opportunities which- causes the rate of consumption to increase for the rich group. (See Section on investment opportunities). Uzawa (1968) discusses consumption behavior in terms of changing time preferences and obtains the result that the MPS declines with income; and another is the result meLtioned in part III.1 - namely, that a > 6, or that consumption (rather than bequests) is the luxury good. Though differences occur between the two definitions of permanent income, the general tenor of the results appears to be the following: (a) the MPS out of transitory is greater than out of permanent income, but always less than 1; (b) that neither the average, nor the marginal, propensity to save is constant and (c) that the APS rises from a low level to a constant (30 0) asymptotic level at the high_ ranges of o 1 income. These results regarding savings behavior are used in Section 111.5 to postulate a "new" savings function; the next section, 11I.4(c) discusses the results pertaining to transitory consumption. 1The "perverse" result -that the NTS declines with. increases in income is (mathematically) required at the high. ranges of income if it is postulated that the APS rises from a low level to an asymptotic level... The latter is a desirable property of any, savings function; otherwise, the model will become explosive at high levels of-income. - - - - - - 73 III.4c) - Savings Model - Effect of Transitory ConsuMtion According to the-model of savings outlined in th-previous section, the marginal propensity to save: transitory, consumptiom sshould Be -1, i.e. stochastic shocks in consumption are absorBed by, cEangos in savings. The results of Tables 11 and 12 indicate that the magnfitude of 3 (equation 39) is not different from -1 for rich households and significantly greater than -1 for other households. Can anything be :nterpreted from this result and the observation that the coefficient af CT declines systematically with increases in permanent income - -., (subsistence households) to -.9 (rich households)? The result that transitory consumption needs are completely absorbed by declines in current savings is based on the assumptMon that households can borrow andlend freely at given interest rates. If liquidity 2 constraints are present2, a household may not be able Cor find it convenient) to borrow the amount required to finance ,. In such an instance, it might'reduce its permanent (planned) level of consumption rather than enter into debt or sale of assets. CT would therefore not be fully reflected in savings i.e. its coefficient would be less than 1 in absolute terms. This result is what is observed. for the groups most likely to face capital market problems - the poor and middle income groups. That liquidity constraints might be responsible for this result is It is interesting to note. that the estimated maitude of a is not at all sensitive to the. aeasures of Y used. 3 2These liquidity constraints can either take the.form of an increasing cost of borrowing or a large difference in the prices of assets bought and sold. 74 also suggested by the. estimated' -value of 3 for the rich group-it is equal to-1. This group is unlikely to face capital -market problems; indeed, it is likely to o5tain favorable terms- of borrowing to finance its expenditures, transient or otfierwise. Support for these results is obtained from a recent series of articles on the effects of famine or scarcity conditions in rural India. These studies indicate that like the result above, it is the preservation of sources of income streams rather than consumption levels that is of crucial importance in determining savings behavior. Reaction to shocks in income levels (or in consumption needs) is more in the form of an alteration in planned consumption than might be expected from an inhabitant of a perfect world. Jodha (1975) concludes from an analysis of households living in scarcity conditions in Rajasthan: "The fact that current consumption is the first thing to be curtailed during the. scarcity period indicates that protection of a given consumption level is not the final goal of the adjustment mechanism during the scarcity period. On the other hand, postponement of the sale or mortgage of assets rather than selling them to maintain current cons=mption, etc., indicate that protection of future streams of income is the main goal of the adjustment mechanism." (p. 1614) The preceding analysis ignores the interdependeny between tran- sient consumption needs and reported consumption.levels. The reported level of transient consumption is an ex-post one; to the extent there is any discretion in transient expenditures, a household witb- a liquidity constraint is already forced to a level of expenditures commensurate with its other (permanent) consumption needs. However, if all transient expenditures are discretionary, then one should get the result of B3=-l. 75 Section 111.5 - Towards a New' Savinas Function The consumption. and savings function results ofSe=tion 11.3 (a) and III.4(b) indicated the -following relationship betment savings and permanent income - an average propensity, to save whichts close to zero for subsistence households but one which increases wit3h permanent income towards an asymptotic value. Te geometrical pattern'f or the sggested functional relationship between S and Y is as indicated in Fig. I. The region OA incor- P. porates subsistence level behavior; rather than a level, there is now a range of permanent incomes along which there is essentially r saving. Region AB is the middle income range which incorporates the Keynesian contention that saving rates increase with income. Region BC has a savings rate which is constant and independent of the level of permaxn income; 1 this region of rich households behaves according to the NTH. The transition from AB to BC contains the region where theSES declines and approaches an asymptotic rate equal to the APS. The savings-income relationship of Figure 1 has sour precedents in the development literature. It was first supplied by Landau (1971) who, in an analysis based on a cross-section of developimg. countries, contended on theoretical grounds that the savings functin should look like Figure 1. (His analysis, however, dealt with the -u;ings behavior of a country through time). In a simulation model of raistribution and its effects on growth, Ahluwalia-Chenery (1974) offer a savings function which in its piecewise linear form looks like tf Landau model - the household saving rate is assumed to increase asymptaUcally to a 1 Note that a non-asymptotic level would imply a cotiinuously in- creasing savings rate and this would be inconsistent with time series behavior. 76 Savings, S Asymptotic Saving rate Permanent Income, Y p O Subsistence Middle Income Rich Households C Figure 1 - An 'exponential' model of savings behavior 77 constant savings rate at high levels of income. Using d2ita from several Latin American cities, Musgrove (1974b) suggests that the relationship shown in the figure accurately reflects the savings behav.ior of house- holds in developing countries - and though it is not tested explicitly, Musgrove posits the following algebraic form for the rela=ionship: S q) (30) S =k (1 -exp (-xY)) Y where k - asymptotic saving rate reached at high levels of income x - parameter which indicates the point of transition towards the asymptotic rate q - parameter determining the speed of the transition. A non-linear dependence of savings on income can be introduced in aysF other than the manner shown in equation (20). As mentioned before, straightforward method is to introduce a quadratic term for permanent 2 (31) S= a + by + CY APS a + b + cY; MPS = b + 2c Y = APS -a + cY Y Y Ari- Aternative way of introducing a non-linear relationship between savings and income is the method of Klein (1954), who suggests the following 'Values of q = 1 and q = 2 were both experimented with, but only Che results for q = 2 are presented in the next section. The latter gave dfsistently better results, both in terms of standard error, and the asymptotic values of k. The q 2 form will be referred to as the etponential form. 78 relationship (32) S =aY + bY log Y APS= a + b log Y ;MPS= a + b + b log Y APS + b. Though convenient from an estimation point of view, both (31) and (3 2) suffer .from the serious drawback that they disallow any points of inflexion in the S-Y plane. This characteristic prevents these functions from incorporating both the Friedman and Keynesian contentions about savings behavior; thus, the savings pattern observed in piece-wise linear form for rural India cannot be adequately represented by these models. If the non-linear term in these models is different from zero, it is so at all levels of income. In other words, these models do not allow their functions to change curves. This would not necessarily be a shortcoming of these functions if it were not for the fact that their predicted savings behavior becomes explosive at high levels of income within the sample range (Table 14). The search for a form which would permit flexibility in saving behavior led Singh to postulate the following savings function: (33) S = aY + bY / (logY)2 + cY / (logY) 2 4 APS = a + b/ (logY) + c / (logY) 3' 5 MPS = APS - 3b/ (logY) - 5c (log Y) If b < o and c > o, then this function has the property that the MPS first increases and then converges to the long run APS - a property 79 shared by the exponential form. The APS is bounded from above by the value of a; the APS is not, however, defined for the region Y approaching zero. These four functional forms were estimated and the results are 1 presented in Table 13. If "fit" were the sole criteria for acceptance (and for forms ap different as these, this is a poor criterion) then the quadratic savings function would be accepted. It, however, be- comes explosive very soon and yields estimates of APS-39% and MPS=72% at a per capita income level of Rs. 3000; and at Rs. 4000 these values become 50 and 92 respectively. The Klein form does not perform as well as the quadratic form in terms of fit, and its estimates Are only slightly better at the high ranges of income. In both these forms, the marginal andaverage savings rates are always increasing. The exponential and Singh forms perform equally vall in terms of fit, and provide similar estimates for the marginal and average propen- sities to save. The difference in the two functions is in their implica- tions for low and high incomes. The exponential form yields reasonable estimates along the entire spectrum of income. The estimated asymptotic savings rate is 38% and the maximum MPS (54%) is reached at a per capita income level of Rs. 2000. (Table 14). The Singh form, by contrast, does not perform well at either end of the income distribution. At low incomes (<Rs. 200) the predicted Note that the estimates for the coefficients of transitory income and consumption are not affected by the particular form chosen to reflect the relationship between S and Y . p 80 TABLE 13 COMPARISON OF SAVINGS FUNCTIONS - CULTIVATORS, 1970-71 Equation Form: Quadratic Klein Singh Exponential Parameter b -45.4 -1.02 1.06 .38 (10.3) (24.7) (13.2) (15.2) -6 b .093 .17 -56.4 -.49 x 10 (6.0) (26.6) (9.2) (8.6) b .0001 665.6 (10.7) (5.9) b .30 .30 .30 .28 (21.5) (2.7) (21.4) (20.2) b -.60 -.60 -.59 -.59 (16.5) (16.4) (16.3) (16.0) -2 2 .37 .38 .38 .36 Standard Error .1849 .1851 .1856 .1883 Q Notes: 1) The functional form for the equations are: 2 Quadratic: S =b + bY + b3Y + b YT + b5cT Klein: S = bY + b2Y log Y + bYT + bT Singh: S = b Y + bY/(log Y)2 + bY/(log Y) + b Y + b.c ng.~4 r 234 T 2 Exponential: S = b Y (1 - eb2 ) + b YT+ b5T 2) Variäbl1äe2)+,Yyby 2) Variablés , Y, YT, C are per capita savings, permanent income (YpX definition), trasitory income and transitory consumption, retgectively. 3) Al. equations were estimated in ratio form, e.g. S 4) Absolute values of t statistics are in parentheses. 5) The results for permanent income (YP35 defiition> are virtually identical to Ypx and are therefore not presented. 81 TABLE 14 FUNCTIONAL FORMS AND PREDICTED SAVING INCOME QUADRATIC KLEIN SINGH EPONENTIAL LEVEL MPS APS MPS APS MPS AFS MPS APS PER CAPITA 200 13.4 -11.4 6.2 -11.0 23.7 -10.5 2.2 0.7 500 19.7 5.4 22.0 4.8 39.3 4.6 12.7 4.4 1000 30.1 15.1 33.9 16.7 47.3 17.1 37.8 14.8 1500 40.5 21.8 40.9 23.7 51.3 23.9 53.6 25.6 2000 50.9 27.8 45.8 28.6 53.8 28.4 54.0 32.3 5000 113.3 60.4 61.6 44.4 61.0 41.0 38.3 38.3 7000 154.9 81.4 67.4 .50.2 63.3 45.0 38.3 38.3 Notes: 1) The saving rates are calculated using the estimates shown in Table 13. APS and MPS are the average and marginal pro- pensities to save. 82 MPS.is 24%, and at high incomes, (Rs. 5000), the predicted APS is 41%, and MPS is 61%. Also, the asymptotic savings rate yielded by the Singh form is abnormally high (100%) and the coefficients for 1/(1ogY)2 and 1/(logY) are difficult to interpret. These consider- ations lead one to reject the Singh form in favor of the exponential form, though it should be mentioned that the latter requires non-linear methods and so is computationally more difficult to use. In summary, it appears that the exponential form best describes household savings behavior with respect to household permanent income. This conclusion is supported by results obtained from a piecewise linear approximation to the S-Y relationship, and by a comparison with other non-linear forms. The asymptotic savings rate yielded by this equation is 38%, which though on the high side, is nevertheless consider- ably lower than the estimates yielded by the other equations. Savings and Income Distribution The results of this section (Part III) are supportive of a non-linear savings function, with the marginal and average rates of saving increasing with income and converging to an asymptotic rate. The fact that this pattern emerges regardless of the definition of income used--current or permanent income--increases its validity. These results imply that both the level and the distribution of permanent income are important deter- minants of aggregate household savings; in particular, that policies of income redistribution will cause a reduction in aggregate savings, The fact that the exponential form yields a decline in the MPS is not of much consequence since the inflexion point in the savings function occurs at Rs. 1800 per capita, or near the top 5% of income (cultivator 83 population). Thus, only if wealth were redistributed amongst the upper classes (an unlikely policy) would aggregate savings not be reduced. Thus, a "strong" result of this study is that, contrary to the permanent income hypothesis, redistribution policies will result in a decline in the supply of household savings. This completes the 'general' analysis of the savings behavior of farmers in rural India. The next section deals with two 'special' topics in saving behavior - the impact of investment opportunities and sources of income on savings, and Section V contains the conclusions derived from the analysis presented in this and the other sections. 84 SECTION IV.1 - Investment Opportunities. and Savings Th his classic study, on, traditional agriculture, Schultz (1964) contended that the cause. of a low savings rate amongst rural households was not necessarily their low incomes but ratRer that there were few investment outlets for potential savers, and the investment opportunities (10) that were available promised only a small rate of return. This relationship between investment opportunities and savings was discussed in qualitative terms by Schultz who concluded that "although there has been a long standing concern about the effects of the level of per family income upon percentage of income that is saved, there has been no comparable concern about-the effects of difference in relative prices of new income streams upon savings and investment." (1964, p. 74). An alternative explanation for a low rural savings rate, but one in the same spirit as Schultz, has been provided by MeKinnon (1973). His contention is that capital is likely to be severely misallocated in the rural sector due to a relative absence of financial intermediaries. This has prevented capital from reaching its highest level--a saver is forced to invest in his own enterprise for a low rate of return whilst investors of high return projects face a shortage of funds. Thus, potential and actual rural savings might be low for two reasons: (a) few investment opportunities and (b) a misallocation of funds due to the underdeveloped nature of the capital market. Adam (1973) has also commented extensively on the -McKinnon phenomenon and believes strongly that there are large amounts of voluntary savings that can be mobilized in the rural secror. 85 These, and other studies haye contended that investment opportunities have a positive effect on sayings. .Theoretically, hoyer, the. effect of an increase in yield (rate of return on investmentj on savings is ambiguous. Like the change in the price. of any normal good, changes in the interest rate (price" of savings) result in income and substitution effects whose resolution is ambiguous on an a priori 1 basis. The historical absence of financial institutions prevents the testing of any relationship between interest rates and savings.2 However, households do receive a return on their savings through invest- ments on the farm. If evidence were available on household responses to differences in these rates of return, the one could possibly generalize to a statement on the yield sensitivity of households; e.g., if savings are inelastic to rates of return on investments (deposits), then the goveznment need not concern itself unduly with deposit rate policy. A direct test of the hypothesis that investment opportunities lead to an increase in savings has (seemingly) never been conducted. Indirect confirmation of the hypothesis is deducedFfrom differences in propensities to save amongst land owninjgroups, agricultural 3 income groups, etc. This study is different from other studies in at least two respects: (a) it explicitly discusses the relationship of investment 1Specification of the- matiyes for saving may remove some of the ambiguity .from the results - a target saving model, for instance, will imply a decline in savings with. an increase in the interest rate. 2Researchers Csee Wright (1967), Weber.-.970)1 have tried to test the interest rate effect on an aggregate basis for the U.S. economy, but have generally met with.inconclusive results. See Section IV.2 (Sources of Income and Savings' for a critical discussion of such studies. 86 opportunities to savings in rural areas unde different assumptions about the nature of the capital -market and (5) a premlim-nary, but direct measure of investment opportunity is developed amd empirically tested. The attributes of a proper 10 index, and the dlifficulties involved in constructing this measure, are also extensiwely discussed. The term "investment opportunity" was first used lay Irving Fisher. He defined it as follows: "The concept of investment apportunity rests on that of an 'option.' An option is any possible income stream open to an individual by utilizing his resources, capital, Labor, land, money, to produce or secure said income stream. An investment opportunity is the opportunity to shift from one such option, or optiamal income stream, to another." (1934, p. 151). For a farm household, an investment oppor- tunity can also be defined to be a perceived shift in its production possibilities. The green revolution years of the late sixties prorvide an ideal representation of shifts in production possibilities. The introduction of high yielding varieties afforded farmers the chance to increase significantly their rates of return from investment; this investment opportunity meant increased profits from both old capital (irrigation) and from new investments. But did this improvement in the return to existing and additional capital result in more savings On the part of farm households? 87 Before the results are discussed, it might be useful to review the theoretical relationship between investment opportuniitties and savings. This is best illustrated by use of diagrams which show the optimal con- 1 sumption and investment decisions on the part of a farm Emusehold. These decisions are dependent upon the transactions in the capital market and Figures 2 (a) and 2 (b) represent the polar cases of (a) a perfect capital market and (b) no capital market or a Robinson Crusoe 2 economy. In both figures, PP and P P' represent the oldi and new production possibility frontiers, C05 C1 represent present and future consumption and AB the rate of exchange between the two per.iods. In the case of a perfect capital market, the shift in PP implies an increase in wealth (AB to A'B') with relative prices Eld constant. If consumption in each period is a superior good, the household will, (in response to a wealth increase), increase consumption- in both periods. Thus, first period consumption will increase, and savings decrease. For the Robinson Crusoe economy, the effect of a shift in PP on savings is ambiguous. This is due to the presence of'bot:h wealth and 1 The theory of inter-temporal choice (optimal investment decision) was developed by Fisher and elaborated by Hirshleifer (1958, 1970). The following discussion follows closely the detailed analysis provided by Hirshleifer (1970). 2 The qualitative results for the case of increasing marginal costs to borrowing are similar to the Robinson Crusoe economy. 88 cl A 0 0 Figure 2(a) - Investment Opportunities and Savings - Perfect Capital Market CI Figure 2(b) - Investment Op portunities and Savings - No Capital Market or Robinson Crusoe economy 89 substitution effects. The. shift to P'P' implies that dhe returns to investment will increase; howeyez, the. amount (sayed). -ay increase. or decrease- These results are based on exogenous shifts in them,PP frontier. It is more realistic to assume that some own investment: (land improve- ment, irrigation, etc.) is necessary to move from PP toP'P'. In this instance, Fig. 2(b) is no longer the appropriate representation of consumer behavior for the Robinson Crusoe economy. CTHe results for the. perfect capital market case are not qualitatively aEfected by the assumption of an endogenous shift in production possibiDlities). The effect of investment opportunities on savings, rather than being ambiguous, is now positive. The explanation is straighttforward - funds for additional investment have to come from additional savings. The theoretical results indicate that the effect Mf IO on savings is not unambiguous but is rather dependent on assumptiams about the capital market. The polar cases of no capital market aand perfect capital market lead to opposite conclusions - the formear causes an increase in savings (in the short run) and the latter as decrease. In the case of an imperfect capital market, the effect us ambiguous, a priori, and dependent on assumptions about the relati've costs of borrowing and returns from investment. How relevant are these results for an analysis of! rural Indian households? Capital markets exist in rural India'and are imperfect. Some farmers might indeed be facing Robinson Crusoe decisions, but a large number of them do borrow from money lenders, lanllords, banks, cooperatives, government, etc. The imperfect nature off the market means that identical households face different interest rates. Evidence for 90 this, on the basis of NCAER data, is contained in Bhalla (1975). Part of this evidence is presented in Tables 15 and 16 which show that (a) large landholders (richer farmers) face lower rates of interest and (b) HYV households can borrow at significantly lower rates of interest than non-HYV households. Given the presence of these differences, no unambiguous statement can be made about the response of savings to an increase in investment opportunities. However, if households can be classified according to the capital markets that they, face, then one should be able to test for differences in saving behavior. Investment Opportunity-Construction of Index and Empirical Results The households covered by the NCAER survey presumably face different marginal rates of return on any new investment. This likelihood is enhanced by the fact that the period covered by the survey -- 1968-69 to 1970-71 -- encompasses the years when the new technology was being adopted by farms in India. However, the construction of an index which will reflect differences in the perceived rate of return on investments is not an easy task. An attempt to construct such a measure is fraught with difficulties, for the following sorts of reasons: (a) The adoption status (HYV cultivation/non-RYV cultivation) of the household should be indicative of its perceived rates of return. The new technology, if adopted properly, can significantly increase the returns from cultivation. (b) In addition to the adoption status, a proper index of invest- ment opportunities should also be weighted by the presence, type and extent of irrigation. The new technology is heavily dependent for its success on the availability and controlled supply of water. A tube TABLE 15 AVERAGE RATES OF INTEREST ON BORROWINGS FROM DIFFERENT SOURCES BY SIZE OF HOLDING 1970-71 (CULTIVATORS ONLY) (Interest rates in per cent per annum) Size of Holding Government Co-operatives Commercial Money- Friends and All (acres) Banks lenders Relatives Sources etc 0 - 5 8.9 8.9 7.2 22.5 0.0 16.0 5 - 10 9.0 9.1 8.1 20.9 0.0 14.3 10 - 15 8.3 9.0 9.4 23.3 0.0 14.9 15 and above 8.4 9.0 9,2 16.3 0.0 9.0 All Holdings 8.8 9.0 8.4 21.8 0.0 14.8 *including landlords and others Source: NCAER, Credit Requirements for Agriculture, 1974. 92 TABLE 16 COST OF CAPITAL BY FARM SIZE AND HYV CLASSIFICATION Farm Size HYV Growers Non-HYV Growers All (acres) % % 0-5 13.9 18.6 17.3 5-15 12.3 14.8 13.8 15-25 11.6 12.7 12.2 > 25 9.7 13.0 11.8 All 12.4 15.8 14.5 Notes: (1) In these calculations, all farms which had a marginal cost of borrowing equal to 0% because of borrowing from "friends and relatives" are excluded. Since most (78%) borrowing from "friends and relatives" is con- centrated in the large size groups, over 15 acres, this means that the estimates presented are upper bounds for the group above 15 acres. Source: Calculated from NCAER survey. See Bhalla (1.975). 93 well investment may therefore be expected to yield diffment returns than a canal irrigation system. (e) The regional location of a household also may have an effect on its perception. One obvious reason is that differences in land quality may affect the profitability of certain investments; another important reason is that the perceptions of a farmer in a successfuZ area of the green revolution (e.g. Punjab) probably differ systematically from those of an identical farmer in an unsuccessful region (e.g. BThar). (d) Distinction by crops is also necessary for a valid investment opportunities index. The cultivation of wheat has proved to be more profitable than the cultivation of rice. Identical on-farm investments, when used in the production of different crops are likely to yield dif- ferent returns. On the assumption that households perceive these differences accurately, this too will affect the estimate of the return to be earned on a particular capital outlay. (e) The availability and proper application of fertilizer can also make a substantial difference to the returns from cultivation. Also, access to working capital may be crucial for the success of the adoption of the new technology. The above list of relevant factors affecting an investment oppor- tunity index is necessarily incomplete; it, nevertheless, points out the difficulties involved in the construction of such an index. A proper definition of an index notwithstanding, the purpose of this section remains the testing of an investment opportunity effect - do households facing a higher rate on their investments save more? The previous discussion has made clear that the adoption status of a house- hold should be included in any assessment of the rates of return I _._._......_._._._. .__________..____ __________.____ P 94 perceived by a household. Thus it might seem appropriate to C classify households according to whether they have, or have not, adopted the ne/w technology. This "index", however, has several draw- backs. It attributes equal opportunities to an adopter regardless of the crop grown, or its regional location. Most importantly, it attributes no investment opportunity to a farmer who might very well be on the verge of adoption. An alternative index is the adoption status of the household is the adoption rate of the district in which a household resides. Soil quality, pattern of crops produced (wheat, rice, etc.) presence of extension programmers, credit availability (government, co-operatives etc.) are all variables which vary more amongst districts than amongst households within a district. Differences in profitability of invest- ment between different.regions should be reflected in differences in district adoption rates. Moreover, use of this index asigns the same 10 to all household had actually adopted the technology. Comparing non-adopters, it is likely that a farm household in Punjab district faces a greater investment opportunity than a household in a Binar district, and differences in district adoption rates should reflect this perception. Though not perfect, the district adoption rate comes closest to a desired index of i4vestment opportunities. The NCAER data contained 2952 cultivator households in 1970-7). These households were aggregated into 100 districts, and a weighted percentage of adopter households was estimated fcr each district. This percentage was then assigned as an TO index to each household within the district. Since the analysis of savings is for the third year of the survey, perceived opportunities 95 of a household are likely to he.hased on past rates of adoption. Consequently, district adoption 'rates- for the second yea of the survey were chosen for analysis. The classification of households by per capita income Csubsistence < Rs. 1500, intermediate Rs. 50G-Rs. 1500 and rich > Rs. 1500) should effectively capture the differences in the capital markets faced by the household. It is likely that the subsistence group corresponds to the 'no capital market' situation and that the rich group (top 5% of income) face a 'perfect' capital market. The hypothesis that investment opportunities affect savings can now be tested - equation (34) allows one to test for differences in the 1 propensities to save: (34) S= a1 + blYp + b2p 10 + b3YT + b4CT where Y = measure of permanent income Y3. or Y ) pR P9 px Y = transitory income T, C T transitory consumption IO investment opportunities index = weighted average of adopters of new technology on a district basis, 1969-70 The results for equation 34, for all household groups., are presented in Table 17. For the suhsistence group CRobinson Cursoe economy) the coefficient b2 is. positive and significant. Thus, holding constant the level of permanent income, it is seen that households 1Again, in order to control for heteroscedasticity, the equation was estimated in ratio- form. 96 TABLE 17 THE EFFECT OF INVESTMENT OPPORTUNITIES ON SAVINGS - CULTIVATORS, 1970-71 RESULTS FOR PERMANENT INCOME, Y px All 500< Yapc Variable . Observations Yapc<500 -157b0 . Yapc>1500 Constant, a -77.0 -41.4 -155.2 -3045 (22.7) (8.0) (10.2) (2.3) MPS, b .23 .11 .34 .51 (24.7) (6.0) (15.8) (7.8) MPS, b2 .009 .05 -.06 -.05 (0.6) (1.8) (3.0) (1.1) Transitory Income,b3 .30 .21 .34 .58 (20.9) (10.4) (17.3) (11.0) Transitory Consump- -.59 -.61 -.57 -.86 4 (15.9) (10.7) (12.0) (4.5) -E2 .34 .21 .34 .51 Standard Error .1912 .1816 .1816 .1890 Notes: 1) Results are based on equation (34) of text. 2) Figures in parentheses represent the absolute value of the t-statistic. 97 facing greater investment opportuniti.es save more-. But do households facing a perfect capital market save lass in response to an increase in investment opportunities? Coefficient 52 is negacive, But not significant (t-statistic is equal to -1.1) for the rich group. 7or the intermediate group, however, the. coefficient is negative and significant.1 Thus, the general nature of the results is encouraging. As emphasized earlier, the results support the contention that consideration of the capital market is crucial for any analysis of savings and investment opportunities. The poorest group, and the one most likely to face problems in the capital market increases its savings in response to investment opportinities. The richest (and intermediate) groups with access to a "perfect" capital market, decrease their savings in response to investment opportunities. Though supportive of theory, these results should be viewed with caution. The measures of permanent income and index of investment opportunities are highly imperfect, and it is by no means clear that the same results would be obtained with perfect measures. 'The results for the Yp35 measure. of permanent income are virtually identical to the Y* measure in terms of signs and significance of the pX estimated coefficients. Consequently, these results are not presented. 98 SECTION IV.2 --Sources of Income and Savings The propensity to save, but of different sources of income. has re- ceived considerable attention in both. the growth- thebry and economic development literature. Identification of tHe source of income (profits, wages, agricultural income, non-agricultural income etc.) is not relevant per se for a study of savings behavior. A rupee is a rupee and presumably the household does not determine its Behavior on the basis of the source of income. Rather, its importance is derived from the assumption that such a classification allows one to stratify households according to differences in the economic environment. Thus, sources of income become a proxy for economic unobservables, and conclusions about the effect of the latter can be drawn from observations on the former. . A traditional method of analysis is to divide income on the basis of occupation into two sources--profits (entrepreneurs) and wages (workers). It is then observed that there is a higher (marginal and average) propensity to save out of profits than out of wages. This result' has been used to explain the process of growth and also as evidence to support the notion that a redistribution of income (from capitalist to worker) will adversely affect the savings rate. This result (see Houthakker, 1965) may be due to a spurious correlation.. If profits are correlated with. levels of income, then the higher MPS out of profits may be due to higher levels of income rather than an occupation of 'source- effect. The results of Part III strongly suggest that the former effect may be important and dominant. Other than differences in level of income, there are alternative explanations for the higher observed propensity to save of capitalists. The permanent income hypothesis offers one explanation. If it is 99 assued that entrepreneurial income is inhrently more- yariahle. than salary income, then at any given level of income, transitory income will form a larger part of total income for the self-employed. T.us, any cross-section data will show- a higher saving rate for the entrepreneur. (This requires the assumption that tha MPS out of transitory income is greater than the MPS out of permanent income - an assumption supported by mos.t analysis of savings behavior, and Part III.) Another explanation for the higher saving ratio of entrepreneurs proceeds from the assumption that investment opportunities may exist within the firm (thus inducing capitalists to save more) or that capital market imperfections drastically reduce outlets for savings. The above explanations for the higher prcpensity to save on the part of the capitalists are not mutually exclusive. Thus, there is no way to choose between the different explanations - is it the higher level of permanent income, or the greater- variability, or profitable outlets for saving or just tastes that cause entrepeneurs to save at a higher rate? The longitudinal nature of the NCAER data allows one to be more precise about the different causes of savings. In particular, a relatively rigorous test of Friedman' s contention, that it is the variability in income which accounts for higher observed propensities, is conducted in the latter half of this section. But first a 'traditional' analysis of savings and. sources of income is presented. Sources of Income - Research- with Farm Data The easy- distinction Fetween a capitalist and workier is lost when one attempts to study the sources of income hypothesis for farmers. The income from a farm is' a return to both labor and capital., and cannot be 100 eas.ily separated. into its components. Farm incomes can, howeyer, be divided into their on-farm and off-farm components, and thus a divison of income into separate. sources can. be. achieed. Though. this differentiation does not follow the traditional capitalist/worker dichotomy, it can be asserted that a household whici derives all of its income from a farm systematically differs-from one 'fiich. derives only fifty percent of its income in a likemanner. The former is likely to have greater control over assets, a higher ratic of profits/income and perhaps greater investment opportunities. That a meaningful differentiation of income by source can be achieved for a farmer by dividing his income into two categories (agricultural income and non-agricultural income) is essentially the view taken by Kelly- Williamson for Indonesia, Ong'et. al., for Taiwan and Mizoguchi (1970) -and Noda (1970) for Japzn. If this dichotomy is accepted, then one can easily test for differences in the propensity to save by comparing coefficients b and c in equationi (35) S a + bY + cYo, where Y agricultural -income, and a Y non-agricultural income. 0 Table 18 presents the results for equation 35 estimated for six different groups - households grouped by the share of agricultural income- in total income, (S ) (5 50 percent, 50-75 percent and > 75 percent), ag and land owning categories C< 5 acres, 5-15 acres, and > 15 acres). The classifications were chosen to create homogeneous groups - it is implicitly assumed that the- size of farm owned and/or the. share of agricultural income are- indicative of asset holdings, ratio of profits TABLE 18 SOURCES OF INCOME AND SAVINGS - CULTIVATORS, 1970-71 Land Categories Percentage Income - Agriculture All <5 5-15 >15 Observations Acres Acres 'Acres 50% 50-75% 75% MPS, Agricultural Income .21 .10 .22 .32 .03 .16 .26 (B 1+B2) (2.8) (3.2) (2.5) (.26) (2.2) (.15) (.93) MPS, Non-Agricultural Income .26 .18 .31 .34 .23 .14 .33 (B) 2 (13.1) (7.4) (8.6) (4.8) (6.2) (1.4) (4.6) PS, Total Income (%) .21 .12 .23 .33 .16 .15 .26 ApS, sdW/ (%) 14.1 4.5 12.6 21.4 8.9 9.4 15.8 Income, Y 4959 2838 5099" 8716 3322 3773 5766 Percentage of -78.6 63.9 86.4 930 33.2 63.5 96.0 Income from Agriculture Land (acres) 10.8 2.6 9.3 28.8 4.3 6.82; 13.8 No. of Observations 1980 810 743 427 360 365 1255 Notes: 1) The estimating form used was: S B B Y B Y e dW 0 _+ i a 2 + y Y where SdW' Y, Y are per capita savings (change in net worth estimate), measured Income, and agricultural income. 2) * - The reported coefficient is (D +B ) and the reported t-statistic is the one associated with B . This ?-statistic indicates the sig- nificance of the difference ýetween the marginal propensities to save from the different sources of income. 3) Absolute value of the t-statistic is in parentheses. 102 to wages, investment opportunities etc. The results show that except for one group CS =5-75%}, the- . --ag - WPS out of non-agricultural income, c, is always higher than the. MKPS out of agricultural income,'b. 'Te. difference. is- signifficant for all households, land owning categories < 5 acres-, 5-15 acres and S < 50%. ag For the 'exception" group, Sag 50-75%, c is less than 5, but tfie difference is not significant. These results are contrary to expectations. If it is assumed that Ya roughly corresponds to profits and Y to wages, them one should have observed that b was greater than c. Analogously, the sfze of Y could be a a proxy for investment opportunities - again, prior belli,ef would indicate that b>c. Different conclusions about the effect sources of Lncome have on savings are drawn by Kelley-Williamson and Ong et. 'aL. Part of the explanation for these conclusions is that the studies employ a different method of estimation. Households are first <classified as per the ratio of income accruing from agriculture, (S ). :Separate savings ag regressions of the form, (36) S a' + b'(Ya + Y 0) are then estimated for each group. This procedure forces the. propensities to save to be the same within groups; thus, it is impossible to test whether propensities to save differ by source of income- Both authors 1Kelley-Williamson employ a lightly different testng -nethod. They estimate. equations which reflect the. cumulative proportions of income that are derived from agriculture.-11 percent, 21 percent ..............91 per- cent. 103 * find that the average- and marginal savings rates incr-se aa the ratio of agricultural incbme..to.total income increases. rThis,esult is .then interpreted to support the thebry that entrepreneurs, or farms with. greater control over assets* save-more. Ong et.-al. add that this might be due to the. higher investment opportunities of the full time farmers. There may be a serious ,.roblLm associated with the testing procedures employed in these two studies. An increasing ratio of agricultural income/total income may be positively correlated with increasing total income. (The simple correlation between the two is .24 in the NCAER data). If savings rates are associated with level of income, then what one observes by running regressi'ons of ratio groups is simply the effect of higher incomes rather than the effect of "control over assets," nentrepreneurial" income or "investment opportunities." If the purpose is to test propensities to save from different income sources, then equation(35)is the appropriate form. An equation, estimated as in(36), may only indicate correlation, rather than causation. How than are the within group results, that the MPS out of non- agricultural income is higher than the I-TS out of agricultural income, to be interpreted? This result, though contrary to other hypotheses, is entirely consistent with the interpretation that sources of income merely reflect the composition of income i.e. its permanent/transitory nature. Non-agricultural income is mainly composed of wages and salaries from outside employment. Small farmers supplement their annual income with outside work since their own farms are not large enough to keep them fully employed. (Outside income formed only 7 percent of total income for households owning more than fifteen acres). Outside income, however, is likely to be more uncertain than on-farm income, since it is dependent 104 on the probability of obtaining a job. (Apart f-om its regular component, outside work.is also resorted to under "special" (and transient) conditions of the: family). Thus, Y' is likely to have 0 a greater transitory component than Ya, and if the MPS out of transitory income is higher (and Part III suggests that this is the. case), then the observed propensities to save (b and c in equation (35)) can be explained by reference to permanent and transitory components. Though plausible, the above interpretation remains conjectural in scope. An empirical verification of the 'variability of income' hypothesis is attempted below. Let b represent the constant propensity to save out of permanent p income, and b the propensity to save out of transitory income. Let t Ya' and Y 0 represent the permanent components, and Y "'Y Y " the a. a o transitory components of the two sources of income. The "true" model of savings behavior is then (37) S ==a =+ b (X + J ) + b (Y + Y ") + e, p a o. ta o rather than the equation actually estimated, S=-a+b (Y Y ") c (Y'+Y")+e +0 0 The transitory components are assumed (by definition) to be uncorrelated with each other, or the permanent components, i.e (38a) cov(Y', Y "1 = cov ( y "') cov (Y , Y "I coyY Y ") a a o o. a o : o a Coy (Y , Y r1) = 0 a 0O In addition, let it be further assumed (temporarily) that there is'no correlation between the permanent components of income, i.e. 105 (38b) cov(Y Y =. *a* o· It can now be shown (see Appendix II), that a a (39a) b =b .a + a , and pvarY var Y a var Y " vrY (39b) c= b 0 + 0 . var Y var Y o 0 If it is now assumed that b > b , then c >B only if Y has a greater t p 0 transitory component i.e. (40) var Y "/var Y > var Y "/var Y 0 0 a a Thus, if var Y " and var Y " can be identified, the hypothesis that b and c differ due to the "transitory" nature of Y and Y can easily be tested. If a non-zero correlation between the permanent components is permitted, then an unamhiguous test of the above assertion cannot be conducted. Specifically, (see Appendix 2), (41a) b =- b (var Y var Y +-var Y var Y )+ D p a o a o b cov(Y ', Y ') (var Y " - cov(Y ', Y ')) + p a o- 0 a 0 h var Y "yar Y -b var Y 0"cov CY ', Y 0 and t a o: t o a· o (41b) 1 - ,pCyar Y.< var Y) † b covCY a, Y 1 Cyar Y - cov (Y ', Y ')) jp o. a p a· o aa o +b Cyar Y "'var Y¯) -B cov (yt ) t 0. a t a· 0 2 and D =var Y var Y -cov(Y ,Y ) . n i o a o Again, if b > b then c > b, if t 106 (42) (bh, b) (yar Y "yar Y - yar Y" yar Y). t p a ao o a The first term of equation 42 reduces to equation 40 if cov(Y ,Y 0 L0 is zero. If this last assumption is not accepted, then the second term in equation 42 if positive, can cause the, expression to be greater than zero and hence c > b. If this is the case, it can no longer be interpreted that c > b because Y has a greater transitory component than Y . If .o a the second term is negative, and still c > b, then of course one has the "strong" result that c > b because var Y "/var Y > var Y "/var Y o o a a Since very restrictive assumptions are necessary to make the second term negative, a "general" test of the hypothesis cannot be conducted. In the empirical analysis below, the 'biases' caused by the assumption cov(Y Y ') # 0 will be noted. a U Identification of the permanent/transitory component of income variance, (var Y ', var Y " etc.) is not possible with cross-section data. a a However, the longitudinal nature of the NCAER data can be exploited to yield estimates of these variances. The only assumptions necessary arg (a) that transitory components of income have zero mean and they they are not correlated for the first and third year of the survey; (b) that permanent components of income are related by a constant factor of proportionality and (c) that this factor of proportionality is given by the ratio of incomes in the two years. In symbols, (43a) E (Y i") E= i 1, 3 (43b) cov(Yi", 3Y" covy ", Y") 0 ai a3 oi o3 107 C43c1 Y g .=gY' Y gY" a'l a.a- 3 'o* Q .03 (43d) g =E fYfEC3) a g = R CYr01)/ECYQ31 These assumptions, (and equation 38a), are enough to Isolate estimates of var Y , var Y " etc., Themethod is as follows, if onae uses covariances a3 a3 for incomes in the first and tird year, tFLen (44a) cov(Y ,aY> = cov(Y '+ y "t, y '+ " al a3 al- al a3 a3 = cov(g Y +J + Y""l Y3+Y3) ga var Y' Similarly, a a3. (44b) cov(Y , Y0)= g var Y0' Equation 44 yields estimates of var Ya' var Y0'. By using the relatioaship varY = var Y + var Y , i=a, o, estimates of the transitory components i3 ~ 3 1 can be obtained. The hypothesis being tested is that differences in propensities to save from different sources are observed because of differences in variability of income. Support for this hypothesis would be obtained if the result c > b is associated with the result var Y"/var Y > 0 0 var Y"/var Y (Equation 40). a a Table 19 presents the results for the estimated proportions of income variance, by source and type of income. Estimates of b and c, reported in Table 18, are also presented. The results confirm, in a rather striking manner,the hypothesis that variability in income accounts for differerces in observed saving rates. In every cas,a that c > b, var Y'/var Y > var Y"/var Y. In the one case that b > c, the opposite (and consistent) result that var Y"/var Y < var Y"/var Y o o a a 108 TAIBLE 19- Vatiability of Income. in Different Sources of rmcome. *., ayY 3 . var Y" aY -ýPSY 14PSY - ...... -var .Y a b Land Groups < 5 acres .39 .37 .18 .10 5-15 acres .53 .43 .31 .22 > 15 aczes .70 .50 .34 .32 Share of agricultural income < 50% .31 .14 .23. .03 50-75% .08 .17 .14. .16 > 75% .85 .41 .33' .26 All householdE .52 .38 .26 .21 'Notes: 1) Estimates for b and c (equation 35) are frim Table. 18. 2) Y ", Y '' represent the transitory componemrfes of agricultural and non-agricultural income, respectively- 109 is obtained. Though encouraging, the.esillts are not general for the reason noted earlier i.e. the non-zero nature of covcf , 7 1.. For two a - . groups of households, (1 5 acres; > 15 acres) the correlation between sources of income is not significantly different from zero. The above results are valid for these two groups; for the other groups, the results are consistent with the stated hypothesis. In summary, the results of this section indicate that saving rates increase with the. level of land ownership and the share of income (S ) derived from agriculture. Other authors (Felley-Williamson, ,ag Ong et. al., Noda) have interpreted this finding to be in support of the hypothesis that land ownership and Sag, as proxies for entre- preneurial income and/or investment opportunities,have a positive effect on the propensities to sare. Though plausible, this tconclusion' was not accepted for the NCAER households for two reasons: (a) mean income levels also increase with the saving rates of the groups, and therefore might be a major cause of the observed result - analysis of Part III indicates that saving rates increase with permanent and/or measured income; and (b) within group r.gressions resulted in propensities to save being higher out of non-agricultural income. A simple explanation for this 'contrary' result was offered - namely, that differences in variability in the sources of income were causing differences in the observed propensities to saye. The panel nature- of NCAMR data was used to estimate the permanent and transitory variabiliMy of each. source of income. The results were consistent with (and for sme groups, strongly in support of) the variability of income hypothesis. 110 Section V - Summary and Conclusions Empirical evidence, on the determinants of household savings in developing countries is sparse, and is especially- lacing for the rural areas of the developing countries --areas which. mf ten account for 60-80% of the total population. In this paper an attempt was made to analyze the -avings behavior of rural households. As its basis, it used the NCAER Eousehold panel data for rural India, 1968-69 to 1970-71. The study- was conducted at two levels - descriptive (Part II) and analytical CParts III and IV). The important "descriptive" results of the study ware: (a) The nature of the NCAER data allowed the astimation of two definitions of savings - SdW Cchange in net worth),, ,and SY-C (income minus consumption). It was found that the S - .&efinition dW. yielded estimates which were more consistent with natiamal accounts estimates. The national accounts estimate for all households (urban and rural) in India was 10.1% during 1970-71. Te S dW estimate, for the rural (and poorer) households of Ind±a was 5.4%, and the S estimate was 16.3%; (b) Disaggregation of the rural sector into its cultivating and non-cultivating population revealed that the former group had higher incomes and a higher savings rate--Rs.. 3300 and 27.8% vs. Rs. 1768 and 0.4% respectively. Thus, in terms, of savfUgs potential, there seems to be none for the poQrest sector of the pqpulation, i.e., non-cultivators; (c) Regardless of the manner in which- households were grouped, the data consistently revealed that- household saving rz!r-es increased with measured group income. If transitory income amongst groups is 111 presumed to. cancel out, then these. figures indicate. tkAt saying rates. increase iith- permanent income.. .This inteXretation vas supported by the regression results contained in Part III; (d) The NCAER survey is one.of the very few surveys with information on the non-monetized investment of farm households. Problems of measurement were discussed, and the pattern of non-monetized investment explored, in Section 11.4(b). It was observed that for farms of all sizes, a sizeable- proportion (20-25%) had positive investment of a non-monetized nature. This component, though. small for the aggregate sample,'was found to be a significant component of investment (15%) for the small farmers (<5 acres). Since these house- holds form a major proportion (60%) of the cultivating population, the estimates of global savings that ignore this component are likely to underestimate significantly the "gross" savings of farm households. Parts III and IV of the paper contain theoretical and empirical analysis of the savings behavi-)r of farm households, 1970-71. A major goal of the study was to rigorously test the contention that the saving rate is independent of the level of permanent inccme - a major result of both the life cycle and the Iermanent income hypothesis (PIH) of saving behavior. Toward this end, the longitudinal nature of the data was exploited to yield two conceptually di2ferent estimates of permanent income - on-, based on a weighted average of past incomes, Yp,, (with weights derived from considerations of subjective discount rates and expected incomel', and the. other based on the. assats" owned by a household. The. latter was a.modified earnings funrtion method 112 of estimating pe;rmanent income -,modified because it explicitly allowed for (permanent) individual differences in the estimation of permanent income. The importance of thes-e "individual effects" was indicated by the fact that tb.eir inclusion increased radically the variance explained in measured income -from 56% to 86%. These measures of permanent income, Y, along with.measured P income, Y, were tested in ,a model of savings Behavior. One of the major conclusions of this analysis was the observation that differences in subjective discount rates, i, did not make much qualitative different to the results regarding saving behavior. The discount rates considered covered a wide range - 10%, 35% and 75%. These results indicate that undue concern with. selecting the "proper" discount rate may be unnecessary. In addition, the results of the earnings function method, Y (an.estimate based on very different px assumptions), were virtually identical to the results obtained with Y35. Thus, the results obtained are likely to be robust, and not sensitive to a particular definition of permanent income. In order to control for differences in capital market conditions (and to obtain a piecewise linear savings function) households were grouped according to the level of their average per capita income - subsistence (<Rs. 500 per capita), intermediate (Rs. 500 - Rs. 1500 p.c.1 and rich. (Rs. 1500 p.c.). Separate savings functions were estimated for each-group. The major results were: (a) The marginal propensity to save out of transitory income was found be significantly higher than the propensity to save out of permanent income for the aggregatE. sample, these propensities were 30% and 23% (Y definition). Only for the rich group were these propensities not px significantly different. These results reject the Keynesian hypothesis 113 which. stipulates that the. two propens.ities are identic=al. This, result also contradicts. Qne- element of .the'PIR in that the.MPS transitory income was considerably less than 1. (5) Saving rates out of permanent income, -marinal and average, were found to increase with.the level of permanent inexeme. This result holds for all measures of Y Thus, the proposition tHiat saving rates are independent of the level of permanent income is strongly rejected by the data. Thus, results Ca) and 1(b) together. rejec=t-both the Keynesian and the permanent income hypothesis of szvings behavior. (c) Classification of households by levels of ;average per capita income allowed one to determine the pattern of savings behavior. It was observed that the elasticity of savings with respect to permanent income was close to unity at subsistence levels, was less than unity at intermediate levels, and increased to umty at high levels of permanent income. An attempt was.made to iorporate these results into an aggregate savings function. An 'exponential' form for such a function was proposed and successfully tested ira this paper. This form has desirable properties at all ranges of income - it allows for a savings rate which rises from zero at.subsistence lels to an asymptotic (constant) savings rate at high income levets. This function also permits the savings rate to depend on levels of parmanent income, and in particular, to increase with. them. The asymptotic saving rate yielded by this form was 38% and this level is reachedi .at an income level of approximately Ra. 4000 per capita. (d) In the introduction, it was pointed out tdhat a major policy issue was the tradeoff between growth. and income distribution. According to the accepted theoriesof savings behavior- (life cycle and permanent income hypothesis), the saving rate is indaendent of the 114 leyel of permanent-income, and therefore, .income redistribution cannot affect the overall supply of sayings.. 'The results of this. paper strongly reject this conclusion. Saving rates do increase with the level of permanent income and only at very high. levels Ctop 5% of the cultivation population) do they reach. a constant level. Thus, any practical redistribution policy is likely to have a negative effect on the supply of household savings. In part IV of this paper, the effect of investment opportunities and sources of income on savings, was studied. The theoretical relation- ship between investment opportunities and savings was discussed and it was established that saving rat&s-could either increase or decrease with an increase in these opportunities. The result depended critically on the possibilities available to the household for transactions in the capital market. Two polar cases were studied: Ci) perfect capital market and (ii) no capital market. Theory predicts opposite effects of investment opportunities on savings for the two groups - savings should decline in the former case and increase in the latter. The conditions necessary for a proper investment opportunities index was extensively discussed, and an index proposed and tested. The empirical results supported the theoretical predictions; investment opportunities increased savings, ceteris paribus, for the poor group of households (no capital market) and had a negative effect for the intermediate and rich group (perfect capital market). The sources of income effect on savings was studied by-decomposig income into its agricultural and non-agricultural components. The distinction followed the popular dichotomy in the literature'between 115 the sayings behavior of capitalists and workers. The peneral presump- tion is that the propensity to.saye is higher out of agricultural incomes since these incomes represent an "entrepreneurial" or an "investment opportunity" effect. It was observed that (a) such a distinction was improper for thefarming population since these house- holds derive income from both-capital and 'pure' labor Cb) that the relationship observed by other researchers-may be spurious since ratios of agricultural income tend to be positively related with levels of total income and (c) that a useful way of interpreting the different propensities was in terms of the permanent and transitory components of the. respective sources. The NCAER data revealed that the propensity to save from non-agricultural income was higher than the propensity to save from agricultural income. An 'alternative' explanation for this result is provided by the variability of income or permanent income hypothesis. According to it, the MPS out of non-agricultural income will be higher if such incomes have a greater transitory component, and if the MPS out of transitory income is higher than that out of permanent income. The latter condition was supported by the results in Part III. Regarding the former, the longitudinal nature of the data was used to isolate the permanent and transitory components of income variance for each.source of income. The results consistently supported the 'variability' hypothesis i.e. MPS out of non-agricultural income was higher when its transitory component was larger. 116 APPENDIX I The NCAER panel survey of rural Indian households, 1968-69 to 1970-71, has data for 4,118 households. The analysis of savings be- havior, reported in Parts III and IV of this apper, is based on a sample of 1,980 observations. This selection of observations was guided by three principles: (a) the data be relatively homogeneous in the occupational sttuc- ture of the households so that occupational effects on savings behavior are adequately controlled for; (b) the data be logically consistent in its information on sav- ings and income, the two major variables for analysis, and Cc) the data be devoid, to the extent possible, of transcription and/or measurement error. Occupational structure: Only households that described themselves as cultivators for all three years of the survey were selected for analy- sis. Further, to improve homogeneity, households which either bought. or sold land or cultivated less than .05 acres during any of the three years were excluded. (The latter.requirement was imposed to eliminate households which were described as cultivators because they had an ex- tremely small ("garden" plot) of land). Logical consistency: Some households had reported savings, SdW (change in net worth estimate),as being greater than income (Y). Since this is theoretically impossible (according to the definitions used by NCAER), these households were excluded. Moreover, any household with negative income for any of the three years was omitted. Though negative incomes are possible, their presence does cause problems in estimating equa- tions in ratio form, e.g. a household with income of Rs.-1000 and sav- ings of Rs.-2000 will show a savings rate of 200 percent. 1 The savings behavior of cultivators is analyzed in detail (Parts II-IV), with only summary information provided for the non-cultivators (Part II). The latter group was excluded from analysis because (a) they form the poorest sector of society and have little saving capac- ity (b) almost 50 percent of them reported zero savings, S estimate, dw during 1970-71 and (c) no information on assets or credit use is avail- able for them. The asymmetry in the collection of data is due to the fact that extra surveys on fertilizer and credit use were conducted on the cultivator households. 117 Transcription/measurement errors: (i) Households that had a saving rate less than -450 percent or greater than 75 percent for any of the three years were excluded. Sup- plementary information on consumption and income supports the contention that for a large number of households, measurement error is responsible for the observed extreme values of savings. (ii) Households with exceedingly low incomes (less than Rs.500). during the first year of the survey were also excluded. These house- holds revealed exorbitant increases in income from the first to the second year of the survey. (For households earning less than Rs.500, the percentage change in average income, 1968-69 to 1969-70, was 529 percent, for others, 2.2 percent). In 1968-69, there were 54 house- holds earning less than Rs.500; in 1970-71, only one household was present in this category. This leads one to suspect either under re- porting or unaccounting of income for low income households in the 1 first year of the survey. Since first-year incomes are-used in con- structing a measure of permanent income, these households (2.5 percent of the cultivator sample) were excluded from analysis. Procedures (a), (b), and (c) reduce the cultivator population from 2,459 to 1,980 observations. 1 The general suspicion with survey data is that the rich under- state their income. This may also be occurring with the rich households canvassed by NCAER. However, what one is concerned with in the case of low income households are the changes in understatement of income. These (low income) households did not show substantially different in- creases in income from the second to the third year. Independent estimates of savings and consumption for the first year also confirmed the tendency of under reporting of income for these households. 118 .APPENDIX II Sources of 'Income Equation -- Derivation of Formulae A model of the following form is estimated S= a + bY + cY (1) a o where Y = agricultural income, and a. Y = non-agricultural income o The' "true"' model of savings behavior is S =a + b (Y' + Y )+b (Y"+Y") (2) p a o t a o where b (bt) are the propensities to consume permanent (transitory) income; p -t Ya Y o are the permanent components, and a o* Ya ." are the transitory components of each source of income. Assumption 1: The marginal propensity to save out of transitory income is greater than the propensity to save out of permanent income i.e. b > b t p Assumption 2: The transitory components of income are not correlated with the permanent components or with each other i.e. cov (Y Y " G. 0 where z = a, o z z If equation (1) is estimated, then b. Yar(Y ) coy(Y Y ) - coS a ao a e co(Y ,Y 1 var CY 1 covCS,Y ) or at -covCY ,Y )(covCS,Y) o) ar aY a)/ 1 oa o ) -cov(Y ,Y var Y cov(S,Y ) (3) a a o 119 2 where D var Y var Y- cov (Y ,Y) a o a o Given the assumptions, one can derive expressions for cov(S,Y ) and cov(S,Y 0) from equation 2: cov(S,Ya') = b (varY ' + cov(Y , Y ')) cov(S,Y ") = b var'Y " a t a cov(S,Y') = b (var Y ' + cov(Y ',Y ') o p o a o cov(S,Y ") =b var Y " 0 t 0 .'. cov(S,Y ) = b (var Y ',+ cov(Y ,Y ')) + b var Y a p a a o t a and cov(S,Y ) b (var Y ' + cov(Y ',Y ')) + b var Y o p o a o t o Also, cov(Y Y) = cov(Ya' + Y " Y o'') = cov(Y ' o a0a a 0 a 0 Substituting in expression (3'), one obtains b =-{b var Y (var Y ' + cov(Y ',Y ')) -b cov(Y ',Y ')(var Y ' + cov(Y ,Y 1)) D p o a a o p a o o a o + b var,Y . var Y "- b var Y " cQv(Y ',Y '1)} t o a t o a o I1 and c = - {b var Y (var Y ' + cov(Y ',Y ')) - b cov(Y ',Y ')(varY ' + cov(Y ',Y. D p a o a o p a o a a o + b var Y var Y " - b var Y " cov(Y ',Y ')} t a o t a a o The conditions under which the propensity to save out of non- agricultural income is higher than out of agricultural income is c > b or c-b > 0. 120 Therefore, 1 c-b = [b {(var Y var Y ' -var Y var Y) + D p a o a o cov(Ya ,yo ')(var Y - var Ya' - var Y + var Y' + cov aY. t 0 a a o a 0 0 a [b { (var Y vac-Y"+ var Y var Y - v Ya- var Ya va Y D p0 a 0 0a 0 + cov(Y ',Y )(var Y " - var Y ")} a 0 a 0 + b {(var Y "var Y ' + y -var Ya''varY' -arYa + cov(Y ',Y ')(var Y - var Ya")}] a oo a or c -b =-[(b - b )(var Y "var Y '-var Y "var- ')+ D t p 0 a a (b - b ) cov(Y ',Y ')(var Y " - var Y ")i t p a. 0 0 a If it is now assumed that cov(Y ',Y ') 0 and h > b a 0 t p then c > b only if var Y " var Y ' - var Y " var Y ' > 0 o a a o or var Y0"(var Y -var Y ") -var Y "(var Y -var' ") > 0 o a a a 0 0 or var Y "var Y -var Y "var'Y > 0 o a a 0 or var Y "var Y > var Y "var Y o a a 0 var Y, var Y" 0 a or Y > var Y which is equation (40) of t2he text. 0 a 121 REFERENCES Adams, Dale W., "The Case for Voluntary Savings Mobilization: Why Rural Capital Markets Flounder," Spring Review, Vol. 19, June 1973. Ahluwalia, M. S. and H. Chenery, "A Model of Distribution and Growth," in H. Chenery.and others (ed.), Redistribution with Growth, Oxford University Press, 1974. Alamgir, Mohiuddin, "Aspects of Rural Savings, and Investment in Developing Countries," mimeo, Development Economics - Department, World Bank, 1975. ---, "Structure and Characteristics of Private Savings in Participating Developing Countries," mimeo, ECAFE, September 1974. Betancourt, Roger R., "The Normal Income Hypothesis in Chile," Journal of American Statistical Association, June 1971. Bhalla, Surjit S., "Farm Size, Productivity and Technical Change in Irdian Agriculture," chapter in R. Albert Berry and William R. Cline, Farm Size. Factor Productivity and Technical Chance in Developing Countries, report submitted to International Labor Organization, 1975. 3.1inder, Alan, Toward An Economic Theory of Income Distribution, X.I.T. Press, 1974. - ----, "Distribution Effects and the Aggregate Consumption Function," Journal of Political Economy, June 1976. Blumenthal, Tuvia, Savings in Post-War Japan, East Asian Mono- graph No. 35, Harvard University Press, 1970. Crockett, J..,and I. Friend, "Consumer Investment Behavior," in Determinants of Investment Behavior, National Bureau of Economic Research, 1967. ----, "Consumption and Saving in Economic Development," in R.-.Ferber (ed.), Consumption, Prices and Economic Develop- ment'(to be published by Brookings Institution), mimeo 1973. Dorfman, Robert, "'An Economic Interpretation of Optimal Control Theory," American Economic Review, December 1969. Fisher, Irving, The Theory of Interest, Macmillan Publishing Co., 1930. 122 Friedman, Milton, A Theory of the Consumption Functibn, Princeton: National Bureau of Economic Research, 1957. "Windfalls, the 'Horizon', and Related Concepts in the Permanent-Income Hypothesis," in Carl Christ (ed.), Measurement in Economics and Econometrics: Essays in Memory of Yehuda Grunfeld, Stanford University Press, 1963. Friend, I., "Propensity to Save in India," in P. S. Lokanathan .ed.), Economic Development: Issues and Policies, 1966. Gordon, eoger H., "Essays on the causes and Equitable Treatment of Differences in Earnings anU Avility", unpubished Ph.D. dissertation, M.I.T., June 1976. Hirshleifer, Jack, "On the Theory of Optimal Investment Deci- sions," Journal of Political Economy, August 1958. ---------, Investment, Interest and Capital, Prentice Hall, 1-970. Holbrook, Robert, "The Three Year Horizon: An Analysis of the Evidence," Journal of Political Economy, October 1967. Houthakker, H.'S., "on Some Determinants of Savings in Developed and Underdeveloped Countries," in E.A.G. Robinson (ed.), Problems in Economic Development, London, 1965. Jodha, N. S., "Famine and Famine Policies--Some Empirical Evir- dence," Ecoaomic and Political Weekly, October 11, 1975. Kelley, Allen E. and Jeffrey Williamson, "Household Saving Behavior in the Developing Economies," Economic Development and Cultural Change, April 1968. Keynes, J. M., The General Theory of Emvloyment, Interest and. Money, New York: Harcourt & Brace, 1936. Klein, L. R., "Statistical Estimation of Fconomic Relations from Survey Data," in Contributions of Survey Methods to Economics L.R. Klein (ed.), Columbia University, 1954. -------.and J. Morgan, "Results of Alternate Statistical Treatment of Sample Survey Data," Journal of American Statistical Associ- a ation, December 1951. Landau, L., "Determinants of Savings in Latin America," in H. Chenery (ed.), Studies in Development Planning, Harvard University Press, 1971. Lewis, W. A., "Economic Development with Unlimited Supplies of Labor," Manchester School of Economic and Social Studies, May 1954. Lilliard, Lee, A., "Inequality: Earnings vs. Human Wealth", Working P7per #80, National Bureau of Economic Research, April 75. 123 Liviatan, Nissan, "Tests of the Permanent Income Hypothesis Based on a Reinterview Saving Survey," in Carl Christ (ed.), :Measure- ment in Economics and Econometrics: Essays in Memory of Yehuda Gzunfeld, Stanford University Press, 1963. ------, "Multiperiod Future Consumption as an Aggregate," Ameri- can Economic Review, September 1966. Mayer, Thomas, Permanent Income, Wealth and Consumption, University of California Press, 1972. McKinnon, Ronald I., Money and Capital in Economic Develoament, Brookings Institution, 1973. Merton, Robert, "Lifetime Portfolio Selection Under Uncertainty: The ContinuousmTime Case," Review of Economics and Statistics, August 1969. Mikesell, Raymond F. and James E. Zinser, "The Nature of the Savings Function in Developing Countries: A Survey of the Theoretical and Empirical Literature," Journal of Econamic Literature, March 1973. Mizoguchi, T., Personal Savings & Consumption in Post-War JaDan, Tokyo, 1970. Modigliani, Franco and R. Brumberg, "Utility Analysis ahd the Consumption Function: An Interpretation of Cross-Section Data," in K. Kurihara (ed.), Post Keynesian Economics, 1954. .and A. Ando, "Life Cycle Hypothesis of Savings: Lggre- gate Implications and Tests," American Economic Review, March 1963. --, "The Life Cycle Hypothesis of Saving, the Demand for Wealth and the Supply of Capital," Social Research, Summer 1966. , "The Life Cycle Hypothesis of Saving and Intercom=try Differences in the Saving Ratio," in J. Wolfe et al (ed.), Induction, Growth and Trade, Essays in Honor of Sir Rov Harrod, Oxford: Clarendon Press, 1970. Musgrove, P., Determination and Distribution of Permanent House- hold Income in Urban South America, unpublished Ph.D. disserta- tion, M.I.T., June 1974a. ------- $"Determinants of Urban Household Consumption and Savings in Latin America: A Summary of Evidence from the ECIEL Surveys," mimeo, Brookings Institution, September 1974b. 124 National Council 6dr Applied Economic Research, Rural Bausehold Saving Survey, New Delhi, 1962. Credit'Requirements for Agriculture, New Delhi, 1974. ---, Additional Rural Incomes Survey, Final Report, New Delhi, 1974. Noda, T., "Savings of Farm Households," in Agriculture and Economic Growth: Japan's Experience, K. Ohkawa et al (ed.), Princeton University Press, 1970. Ong, Marcia L., D. Adams, and I. J. Singh, "Voluntary Rural Savings Capacities in Taiwan 1960 to 1970," Economics and Sociology paper No. 175, Ohio State University, 1973. Panikar, P. G. K., "Rural Savings in India," Economic Develop- ment and Cultural Change, October 1961. Ramanathan, R., "An Econometric Exploration of Indian Saving Behavior," Journal of American Statistical Association, March 1969. Schultze, T.,..Transforming Traditional Agriculture, Yale University Press, 1964. Singh, S. K., "Determinants of Aggregate Savings," Econadc Working Paper No. 127, World Bank, 1972. Snyder, Donald W., "Econometric Studies of Household Saving Behavior in Developing Countries: A Survey," Journal of Development Studies, January 1974. Tobin, J., "Consumer Debt and Spending: Some Evidence from Analysis of a Survey," in Consumer Installment Credit, Tart II, Vol. 1, National Bureau of Economic Research, 1957. Uzawa, Hirofumi, "Time Preference, the Consumption Function, and Optimum Asset Holdings," in Value, Capital and Growth: Papers in Honour of Sir John Hicks, J.N. Wolfe (ed.), Edinburgh University Press, Weber, W. E., "Effect of Interest Rates on Aggregate Consmption," American Economic Review, September 1970. Wright, C., "Some Evidence on the Interest Elasticity of Con- sumption," American Economic Review, September 1967.. 125 yaari, Menahem E., "On the Consumers Lifetime Allocation P=ocess," Internationål Economic Review, September 1964. Zellner, Arnold, "Tests of Some Basic Propositions in the TMheory of Consumption," American Economic Review, May 1960. rr z71 11... . DOMETSTIC FINANCE STUDIl?,S S8. Government Finance and Resource Xobilization in Pakistan. April 1975. W.S.. K-e. # 9. An Examination of Change in the Retail Price and, Taxation of Petroleum Products (July 1973-July,r 1974). September 1975. K. 'Saito. # 10. Fiscal Decentralization and Economic Development. November 1975. W.S. Kee. 11. Guidelines for Incomes Policy in Developing Countries. January 1976. V.V. Bhatt. $ 12. Centre states Financial Relations in the Context of Planned Davelopment. January 1976. V.V. Bhatt. # 13. Review of Tax Incentives In Selected Asian Countries. January 1976. W.S. Kee. f 14. Development Blanking: Structure and Functio'n. January 1976. V.V. Bhatt. # 15. Economic Development: An Analytic - listorical Approach. February 1976. V.V. Bhatt. 16. Subsidies of Major Malaysian Public Utilities. February 1976. J. Meerman- 4 17. A Simple Model of Budget Incidence. February 1976. J. Meerman. 18. The Definition of Public Overhead Expenditures and the Distribution by Household. February 1976. J. Meerman. i . Consolidated Accounts of Non-finiancial Public Sector of Jamaica. March 197G. K. Artus 20. Food Distribution Program of Sri Lanka. June 1976. K. Artus. 21. Aspects of Rural Savings and Investment. .June 1976 M. Alamgir. 22. On Development Problem and Strategy. June 1976. V.V. Bhatt. *23. Resource Mobilization in Developing ountries: Financial Institutions and Polic±es. . Septemaber 1976. V.V. Bhatt. 24. The OPEC Countries and the 'Absorption of 01 Revenues: An Essay. October 1976. S. Bery. 5 2. Public Sector Enterprises in India: A Descriptive, Survey. October 1976. S. Berv. i 26. Revenue Policy for Thailand: income, Sales and PropertyTa=. October 1976. W.S. Kee. 27. An Estimation of the rlousehold Sector Savings of Sri Lanka. August 1976. K. SaIto. 28. The Unorganized Financial Sector in Korea, 1945-75.. November 1976. Y.C. park. 29. Contractual Savin2s and the WhilisPtioen of Resources. Novermber 1976. P. Sh=me. 30. Tax Evasion and Avoidance: A Crucial but Neg.ected Aspect of Taxation. December 1976. J. N. Sharma f31. Aspects of Savings Behavior in Rural India. December 1976. Surjit Bhalla.