Unemployment, migration, and wages in Turkey 1962-85

Policy, Planning, and Research I WORKING PAPERS Poverty, Equity, and Growth The World Bank July 1989 WPS 230 Unemployment, Migration, and Wages in Turkey 1962-85 Bent Hansen Long-term voluntary unemployment of a more fully educated youth population is as much a factor in the long-term upward trend in unemployment in Turkey as involuntary unemployment because of stabilization. The Policy, Planning, and ResearchComplex disinbutes PPR W,od ngPaper odissensinate e fuindings of work in progress and to encourage th cxchange of ideas among Bank staff and all others intcresto in desrlopment Issues. These papers carry the namcs of the authors, reflect only their views, and should bc used and cited accordingly The findmgs, intcrprcuuons, and conclusions are the authors' own. They should not be attributed to the World Bank, its l3oard of Directors, its management, or any of its member countres  Plc, Planning, and Research Poverty, Eqult-1, and Growth Working with poor, incomplete data on Tur- bly as a result of th,. recession accompanying the key's labor market, Hansen analyzed the status stabilization policy of 1978. This helps to and development of Turkey's labor market since explain why rural unemployment which was 1962. He concludes that: lower than urban unemployment in 1973-74, but has increased much more than urban unemploy- The estimated steady upward trend in ment since then. unemployment since 1962 (to 12.7 percent in 1985) may be misleading. Cyclical short-term Hansen concludes that, aU things consid- changes have bcen important, unemployment ered, the long upward trend in unemployment in was probably declining in the early 1970s and Turkey is partly a matter of the vo.Jntary again after 1982, and unemployment increased unemployment of a better-educated population most betwecn 1978 and 1982, with the stabiliza- of youth, and partly a matter of involuntary tion program. At the same time, young, edu- unemployment related to the stabilization cated single people with relatively little work program. experience and long spells of unemployment dominated the unemployment picture. So long- Although the data are extremely unreliable, term unemploymcnt of youth was as much a Hansen believes that wages in manufacturing factor as unemployment because of stabilization. and mining for 1962-85 have been determined less by rrarket forces than by increasingly Rural-urban migration was substantial militant unions, govenunent intervention, and between 1965 and 1975 but, remarkably enough, the establishment of an incomes policy, includ- was reversed between 1975 and 1980 - proba- ing minimum wages. This paper is a product of the Political Economy of Poverty, Equity, and Growth study. Copies are available free from the World Bank, 1818 H Street NW, Washington DC 20433. Please contact Celina Bermudez, room 14-006, extension 39248 (74 pages with tables). The PPR Working Paper Series disseminates the findings of work under way in the Bank's Policy, Planning, and Research Complex. An objective of the series is to get these findings out quickly, even if presentations are less than fully polished. The findings, interpretations, and conclusions in these papers do not necessarily represent official policy of the Bank. Produced at the PPR Dissemination Center  Unemployment, Migration and Wages in Turkey, 1962-85 by Bent Hansen Content page Introduction ........................................... iii 1. LTVaemployment, 1962-85 .................................. 1 Notes to Section 1 .................................... 21 Tables 1.1 - 1.9 ...................................... 22 2. Internal and External Migracion ..32 Tables 2.1 - 2.4 .. .36 3. Nominal Wage Rates - Areas of Knowledge and Ignorance . 39 Notes to Section 3 .54 Tables 3.1 - 3.7 .56 4. Phillips R.lation for Turkey .63 NL.ces to Section 4 .70 Tables 4.1 - 4.3. 72 The author would like to acknowledge Ms. Jariya Charoenwattana for her computational and other-assistance.  Introduction Turkey has during the last decade gone through an extremely interesting process of policy transformation from an inward looking import substitution strategy with etatism the basic economic ideo- logy to an outward looking strategy of export led growth with some institutional change in the direction of privatisation. - After an IMF-IBRD-OECD sponsored stabilization program in 1978, a broad reform program was initiated in 1980 an4s still in progress. Developing countries may have much to learn ffom the Tturkish experience of policy transformation, its positive and negative aspects. Amongst the negative aspects is an apparent, strong and 3ersistent :- increase in the rate of unemployment with a sharp decline in real product wages. lf a fact, how and why has this happened? To answer these questions detailed labor market analysis is required. Unfortunately, statistics and other information about the Turkish labor market are not up to satisfactory standards, less so, of course, the more we go back in time. Data problems are therefore a primary concern of this paper. Equally important we have the problem of how to model the Turkish labor market. Given the speci- fic institutional and political environment and its development, it is by no means obvious that standard models should apply easily and without qualifications. Competitive equilibrium models with market clearing on the one hand and disequilibrium models with Phillips type dynamics on the other hand,are natural candidates. Given the institutional-political setting, however, market-power oriented modelling should not be ruled out. These are the kind of questions we are addressing, natural for research work growing out of a lar- ger project on "The Political Economy of Poverty, Equity and Growth". Given the poor data no definitive ar.swers should be expected. No definitive answers will be given but there are indications that market clearing may be the appropriate model for agricultural mar- kets for unskilled (uneducated) labor and that, for the period in question and for other parts of the labor market, Tqarket-power oriented modelling may be more appropriate. These conclusions, however, are little more than conjectures. -iii-  1: Unemployment, 1962-85 - The official estimates of national unemployment, numbers and rate. for the years 1962-86, prepared-by the SPO (State Planning Organization) and presented in Table 1.4 and Graph 1.1 give rise to serious questions. As reported to the World Bank, the SPO operates with the notion of "total civilian employment" to which is added "urban and rural unemployment excluding agricultural labor surplus" to arrive at the "domestic civilian labor force". Adding to the "urban and rural unemployment excluding agricultural labor surplus" something called "agricultural labor surplus at peak season", so-called "domestic labor surplug'"obtains. Adding further "labor stock abroad", the "total labor surplus" is finally arrived at. Adding "labor stock abroad" to "domestic civilian labor force", the "total civilian labor force" obtains. Measuring "urban and rural unemployment excluding agricultural laber surplus" on "domestic civilian labor force" an unemployment rate of 12.8 percent is found for 1984. Measuring the "domestic labor surplus" un the "domestic civilian labor force" a rate of 16.5 percent is found for the same year. Measuring, finally, "total labor surplus" on "total civilian labor force" a rate of 21.0 percent obtains. It is this conceptual proliferation that - 1-  gives rise to the confusing reporting of unemployment rates for Turkey ranging (for 1984) from 13 to 21 percent. We obviously have to make up our minds here about what we are asking for and that presumably must mean that we have to specify an analytical point-of-view. It is imporc..ant to notice that the SPO distinguishes between "unemployment" and "labor surplus". The notion of unemployment points in the direction of labor market analysis in terms of demand and supply where a crucial problem is to measure the degree of market clearing, so-called involuntary unemployment and its role in wage formation, in a Phillips relation, say, and the standard statistic here is that of labor "not employed and seeking employment at going market wage", typically obtained through special labor force surveys. The notion of labor surplus points in the direction of development theoretical analysis of potential production at full employment, defined either as a target or as labor supply under certain ideal (target) conditions. The problem is whether we are aiming at descriptive, analytical-empirical or prescriptive, normative analysis. While the SPO estimates of "'agricultural labor surplus at peak season" presumably are of the normative type it is not clear how the SPO defines and estimates urban and rural unemploympnt.1 I shall here concentrate upon the unemployment estimates. Although apparently now anchored upon the single national labor force survey of 1985 (Table 1.3), they may for other years be -2-  related to the regular quinquennial population censuses.2 The SPO estimates for 1965, 1975, and 1980, however, greatly exceed both level and increase of the explicit courts of unemployed, seeking employment, made by the censuses for these years and shown in Table 1.2 (compare with Graph 1.1). The censuses have an increase from 0.4 percent in 1965 to 3.6 percent in 1980 while the corresponding SPO estimates (chained) are 3.6 and 10.7 percent, respectively. The SPO estimates may perhaps also have considered the counts of economically active population with occupation "not adequately described and unknown" which increased from 5.1 percent in 1960 to 6.4 percent in 1965 and 10.9 percent in 1970 (age 15 and above) and those for "not adequately defined" which increased from 2.0 percent in 1970 to 4.1 percent in 1975 (age 12 and above). Be that as it may, it is generally acknowledged that population censuses are a poor source of information about unemployment and we should test the SPO estimates of unemployment against all other sources of information available. Other such sources are the special labor force surveys for urban areas, conducted for the years 1967-69 and 1982-85 and for rural areas 1973-74, with data for applications and placements, 1965-86, and for registered unemployed and new vacancies, 1962-86, all from the public labor placement offices. These series are presented in Graph 1.1 and/or Tables 1-.1, 3 and 4. The labor force surveys (1985, pp.vi-viii) are patterned upon the United States model. Assuming sampling and -3-  inflation to be in order, comparability is a problem. We shall later return to the question of the applicability of the U.S. model in the Turkish context. The starting point must be the only national survey, published so conducted in 1985. Like the urban places surveys it was conducted in October-November which is a slack season in agriculture. Unemployed are divided into "active" and "inactive" unemployed, the latter apparently so-called discouraged workers (ibid.p.ix)'I have included the "inactive" unemployed and operate also with a wider concept of unemployed, including in addition "unemployed not seeking employment" and "seasonal workers not seeking a job". Information about such unemployed has been collected by all the Turkish LFSs. It is not clear howkunemployed not seeking employmenO is defined. Indeed, strictly speaking this is nothing but the definition of labor not in the labor force. We may also here be encountering something corresponding to "discouraged workers". It may make sense to include both this category and seasonal workers not employed at the time of the survey even though they do not seek work just now.The measured national become unemployment rates for 1985 / then 12.3 and 12.7 percent, (see Table 1.3) respectively/. Comparing with the measured unemployment rates for urban places (see below) for 1985 we find, computing the difference between national and urban LFSs, implicit unemployment rates for rural areas of 10.3 and 10.7 percent, respectively, a surprisingly high rural unemployment rate, even considering the -4-  season. The national rate of 12.7 percent is the same as the rate reported to the OECD by the SPO for 1985(Table-1.4).. . For earlier years there are no national LFSs but only special surveys for urban and rural areas as mentioned above. For urban areas the LFSs for the years 1982-85 should be fully comparable. The coverage is limited to cities with more than 10,000 inhabitants (tabulations for cities with more than 20,000 inhabitants exist, ibid. pp. 75-88); in 1980 the total population in such cities actually exceeded slightly that in urban areas defined as provincial and district centers. 1968 and 1969 should also be mutually comparable, apparently covering all conglomerations. 1967, finally, covered only cities with 50,000 inhabitants and abov3 with a population about 55 percent of that of urban areas in the just mentioned sense. Assuming unemployment to increase with city size, not a self-evident assumption, 1967 might be upward biased as compared with 1982-85 which in turn might be upward biased as compared with 1968-69 which in so far as size is concerned should be unbiased. For 1967-69 the source refers to November as the time of the surveys. For 1982-85, October-November appear to be the time of the surveys. We have finally the problem that 1967-69 cover persons age 15 and above, 1982-85 persons age 12 and above. High youth unemployment could create an upward bias for 1982-85 as compared with 1967-69. What these possible biases add up to is hard to say. Assuming the biases not to be seriously distorting, and disregarding the -5-  seasonality problems the conclusion would be that urban unemployment may have increased from a level of about 3 percent in 1967-69 to 11-15 percent in 1982-85 with a tendency to increase within both of these periods. For rural areas explicit labor force surveys have only been conducted for 1973-74 and seasonality is a major problem. These surveys were conducted in June, October, and April, in this order, and thus do not cover the winter season where there may be little field work in agriculture and construction work may be down in parts of the country but, on the other hand, farm families may not be looking for employment outside the own farm and in any case unemployment is not easy to either define or measure. A detailed study by FAO and Turkish authorities gives a picture of agricultural seasonalities in the aggregate. The study is clearly of the "normative" type, "combining labor requirements for each major agricultural operation under different methods (power sources) of production to estimate labor force demand and supply in agriculture" and gave the results shown in Table 1.5. "Peak season surplus" is apparently something conceptually different from "seasonally removable labor surplus", the former estimated to be 990,000 in 1967 and 113,000 in 1972, the latter zero in both years. The peak season surplus for 1967 is identical with that appearing for 1967 in the SPO estimates of national labor force and employment for the period 1962-72. For 1972, however, -6-  the latter differs from the figure in Table 1.5, 850,000 against 113,000. 1967 apparently has by the SPO been used as a bench-mark year after which a slow downward trend is assumed, bringing the number down to 665,000 in 1984. More interesting from our point-of-view is, however, that the slack se&sons appear to be March-June and September-November, December-February being relatively busy (soil preparation for the winter wheat?) with July-August the peak (harvest). If that really is the case, an average of unemployed seeking jObs-in June,. October and April mig'-t overestimate annual average- unemployment in rural areas. If now the rural or, at least, the agricultural labor market in Turkey, as experience from other LDCs gives us some reason for believing clears almost instantaneously, certainly at the peak season and possibly at any time during the year, rural wages being sufficiently flexible, there should be no rural unemployed seeking work at any time or, more realistically, the number should at any time be small. People employed in rural areas or not being in the rural labor force might still be seeking employment in urban areas or abroad if they planned to migrate but such questions are not asked by the surveys which in this regard are deficient as is the '-lted States model. Rural unemployment according to the labor force surveys of 1973-74 amounted to 0.2 - 0.7 percent of the rural labor force in June, October and April. This is consistent with the assumption of -7-  market clearing and suggests that at the peak season in agriculture there was at that time practically speaking no unemployment in this sense, but, of course, does not exclude surplus labor in the normative sense even at the peak season. Granted market clearing around the year we should expect the same result from eventual winter surveys. Indeed:, winter surveys might disclose zero unemployment ;accompanied by huge surplus labor. Market clearing requires highly flexible nominal and real (product) wages and we should expect wages to be relatively low during the slack winter months. There is little information about rural wages in Turkey. The little we have is from the SII, Social Insurance Institute, seasonalities are averaged out, it covers only a few recent years, and it applies only to a few insuredworkers in in agriculturewho most certainly are not representative of landless farm hands. Let us return now to the LFS for 1985 and the impiicit unemployment estimate for rural areas, computed by me as the difference between national and urban (10,000 and above) LFS for that year. Table VI.Ap.1.6 shows the breakdown by urban and rural areas and by completed education. The latter breakdown w:ith a distribution of registered unemployed by age groups for 1979-86 (shown in Table 1.8)and distributions by age, marital status, work experience and length of unemploymentin 1985 (I?, table 1.9 help us somewhat to come to grips with the nature of unemployment in Turkey. As already mentioned the implicit unemployment rate for -8-  rural areas in 1985 appears remarkably high. The reason may be that non-agricultural, rural activities are included. Moreover, the national as well as urban LFS were taken in a slack agricultural season with possible implications for measured unemployment. From Tables 1.6-9 it is immediately seen that in 1985 the unemployed typically were educated, young singles with relatively little work experience and being long-term unemployed. The composition of the unemployed with a predominance of young educated singles, possibly living at home, explains why the welfare implications of the increase in unemployment may be minor despite the absence ofunemployment insurance, while the social stability aspects may be so much more serious. One of the most striking differences between rural and urban unemployment is the low unemployment rate of 5.8 percent for labor without formal education in rural areas and the much higher rate of 19.8 percent for such labor in urban areas. This could be the combined result of rural-urban migration with a high degree of wage flexibility for agricultural work which may not generally attract educated labor, except for specialized work. Table 1.8 thus does not seem to exclude the possibility of approximate clearing of the rural labor market for uneducated labor, provided that the rural labor market is sufficiently segmented to make this distinction meaningful. Labor with primary education has about average unemployment rates in both urban and -9  rural areas. High school and vocationally educated have considerably higher unemployment rates in rural than in urban areas. It is the educated labor that contributes the bulk of unemployed both in urban and rural areas, 87 percent in urban and 84 percent in rural areas. This is, of course, partly because educated labor now is the bulk of the labor force, 90 percent in urban and 71 percent in rural areas. Yet, even in rural areas with 29 percent of the labor force uneducated, these supply only 16 percent of the unemployed and at the national level the picture is much the same, the uneducated representing 21 percent of the labor force but only 14 percent of the unemployed. The general conclusion would seem to be that the labor markets for uneducated clear better than those for educated which again would mean that wages for uneducated labor are more flexible than those for educated labor. These findings immediately raise the question whether new entrants to the labor market, school leavers in particular, dominate amongst the unemployed so that we should expect relatively high unemployment rates simply because the educated population increases relatively fast. Of the Turkish LFSs two provides breakdown of unemployed by work experience. Of the unemployed seeking job 1969, 71 percent had had a job before, 29 percent were seeking job for the first time. For 1985 these figures were 64 and 36. Age distributions of unemployed registered at the placement offices are available back to 1979. - 10 -  As Table 1.8 shows the share of the age group 14-19 is not particularly high and did cend to decline from about 18 percent in 1979 to 13 percent in 1986 when unemployment according to all indicators increased strongly. The age group 14-24 remained approximately constant during this period at 55 percent. For the United States corresponding percentages declined from 25.3 percent in 1979 to 17.6 percent in 1984 (age 16-19) and from 48.8 percent in 1979 to 39.1 percent in 1984 (age 16-24). Considering the fact that U.S.-youngsters to a much larger extent go to college and thus enter . the labor market several years later than their Turkish counterparts, it would appear that new entrants from the educational system play a less important role in Turkey than in the United States. This does not mean, of course, that it may not be an important factor in Turkey, it might be argued that school leavers may not register and search for jobs through the placement offices and thus do not appear in the statistics. Of 654,000 unemployed seeking employment in 1982, only 32 percent relied upon the placement offices (Statistical Yearbook of Turkey, 1985, Table 152) while for unemployed with completed primary school education this percent was 29 percent and for those with completed high school education it was 39 percent. Breakdowns of the LFSs for 1985 by age and education support the new entrants hypothesis. In understanding, finally, the high measured rural unemployment rate in 1985 we have to point to the possibility - 11 -  that the standard questions of the American LFS-model may not be appropriately specified for rural Turkey during slack seasons in agriculture. When the leading question: are you employed, is answered in the negative, the follow-up question: are you seeking employment, may not be adequately specified as to both wage level at which and locality in which employment is sought. Even assuming that employment is sought at "going market wage rate", and already here the problems are legion, employment may be sought in rural areas (without migration or commuting) or in urban areas (with migration or commuting) or even abroad (with emigration). Going agricultural market wages during the slack season may be so low that even uneducated, unskilled labor might not be interested in such agricultural employment, yet be seeking urban or foreign employment at more attractive remuneration, everything censidered. Considering the undifferentiated questioning it is quite possible that agricultural labor markets may clear during the agricultural slack with the rural LFS recording large numbers of non-employed seeking employment - outside agriculture or rural areas. Adequate questioning should first ask the rural non-employed: are you seeking work here at going market wages here? After an affirmative answer the person would be counted as rural (and, of course, national) unemployed in the rural (and national) labor force. If the answer were negative, the person would be counted as not in either the rural or the national labor force. The person should then be asked: are - 12 -  you seeking work elsewhere (in urban areas) at going wages there? If the answer were affirmative the person would be counted as national unemployed in the national, but not in the rural labor force. For rural and urban unemployed (labor force) to add up to national unemployed (labor force) we would then have to include rural residents not employed, not seeking rural employment but seeking urban employment, in the category of urban unemployed (urban labor force). From a demand-supply analytical point-of-view this would make sense since such persons would exert pressure on urban but not on rural wages. Whether this interpretation of the Turkish LFS of 1985 holds water can only be settled through careful investigation into the questioning procedures applied by the LFSs. If corroborated, the interpretation makes the results of the 1985 LFSs compatible with the assumption of market clearing in rural labor markets and justifies the use of the national rate of unemployment in a Phillips relation for urban wage rates. - 13 -  For years before 1985 direct information about rural unemployment is nonex-itent - apart from the 1973-74 surveys. One way of approaching the problem is to accept both urban labor force surveys and SPO estimates at face value, assuming that both talk about unemployment in the same sense and -from them compute the implicit rural unemployment for the periods 1967-69 and 1982-85 when only urban LFSs were conducted and the implicit urban unemployment for 1973-74 when only rural LFSs were conducced. Since possibly the SPO notion of unemployment is wider than that of unemployed seeking employment, I have in Tables VI.Ap.1.1 and 2 computed adjusted unemployment rates on a wider definition, including some unemployed not seeking employment. Information about such unemployment has been collezted by all the Turkish LFSs.3 As already argued, it makes sense to count both categories as unemployed in the labor force and if rural LFSs were to be conducted for the winter season the latter category might take on some real importance although landless rural labor seems to be a relatively small group in Turkey.As it is, the adjusted unemployment rates of the LFSs are not spectacular'y higher than the unadjusted ones. Taking into account the coverage of the LFSs as measured by population, we find on this basis an implicit urban unemployment rate of 10.5 percent for 1973-74 and implicit rural unemployment rates of 5.3 percent for 1967-69 and 10.7 percent for 1982-85, average. The implicit urban rate for 1973-74 is simply not credible. The implicit rural rates are too - 14 -  high compared with the actual rates measured by the urban LFSs, especially for 1967-69. The SPO estimates for years earlier than 1985 thus do not appear easy to reconcile with the LFSs. One way of approaching the problem of the national unemployment rates-for years before 1985 would be to accept the national LFS for 1985 with the urban LFSs for 1982-85 and 1967-69 at face value and assume that the ratio between national and urban unemployment rates in the wide sense for all years were the same as in 1985. With this assumption we would actually tend to overestimate national unemployment rates the further we go back in time, assuming the riwral rate to be proportionately lower than than the urban one, because the urban labor force has increased relatively over time. This leads to the following conjectures for the national rate as compared with the SPO estimates: National unemployment rates conjectured SPO estimates pct index pct index 1985 12.7 389 12.7 259 1984 12.0 12.4. 1983 12.0 353 12.4 249 1982 10.6 11.8 - 15 -  1969 4.1 5.2 1968 2.9 100 4.9 100 1967 2.8 4.6 SPO estimates chained in 1972. Judging from these conjectures, the (chained) SPO estimates overestimate the national rate of unemployment by some