1
Output Volatility in Russian Agriculture:
The Significance of Risk and Inefficiency
Raushan Bokusheva and Heinrich Hockmann*
1Introduction
The development of Russian agricultural production during transition has exhibited a remarkably incoherent character. In general, production declined for a number of years, while at the same time considerable output variation was observable (Figure 1). Several studies have been conducted in order to identify the possible causes of agricultural production’s decline in post-Soviet Russia (Serova, 2000, Macours and Swinnen, 2000, Liefert, 2002, Bezlepkina and Oude Lansink, 2003), with possible determinants being deterioration in the terms of trade, the elimination of producer and consumer subsidies, a weak institutional environment and undeveloped factor markets. However, so far the literature has paid little attention to high production variability and its effects on the evolution of production in Russian agriculture.
In this context, the objectives of this study are:
- To analyse whether Russian farms increased production by enhancing their productivity, and in particular their technical efficiency.
- To investigate whether production risk and technical efficiency affected the volatility of agricultural production.
The paper is organised as follows: Section 2 outlines the methodology employed to distinguish and assess two sources of production variability: production risk and technical inefficiency, and thus presents specification of the models used in the study. Data and estimation results with regard to the study objectives are discussed in Section 3. Conclusions are drawn in the final section. Panel data (from 1996 to 2001) of 443 large agricultural enterprises from three regions in central, southern and VolgaRussia are used.
2Theoretical background and model specification
This study employs stochastic frontier analysis (SFA). An important disadvantage of the traditional multiplicative stochastic specification of production technology is the implicit assumption that if any input has a positive effect on output, then a positive effect of this input on output variability is also imposed. Just and Pope (1978) showed that the effects of inputs on output should not be tied a priori to the effects of inputs on output variability. They proposed a more general stochastic specification that includes two general functions: one that specifies the effects of the input on the mean of the output and another that specifies the effect of input on the variance of the output. This approach was extended by Kumbhakar (2002) who introduced an additional function explaining technical inefficiency:
(1) .
Output and inputs are represented by y and x=[x1,...xJ], respectively. The mean production function is f(x; ) and q(x; ) represents the inefficiency function. The function g(xit,β)vit denotes the idiosyncratic component of production risk while systemic production risk is captured by vector D, which consists of dummy variables for the individual years (Greene, 2003). The stochastic term vit and uit are assumed to be i.i.d. N(0,1), and i.i.d. N+(0,u2), and , and are parameters to be estimated. The subscripts i = 1,..,N and t = 1,…,T denote farm and time, respectively. Mean production was approximated by a translog function. Constraints were introduced to impose monotonicity and the necessary conditions of quasi-concavity at the approximation point (Morey 1986). Risk and inefficiency were approximated by a Cobb-Douglas specification.
A single-step maximum likelihood (ML) procedure was employed to estimate the parameters of the specified models. Taking into consideration the distributional assumptions of ν and u, the likelihood function of TN observations is formulated as the product of the probability density functions f(zit) of TN single observations and the Jacobian |J| of the undertaken transformation (zfrom y):
(2), with and
where with .
The probability density function of zit is
(3)
with and Φ(·) being the distribution function of the standard normal random variable (Kumbhakar, 2002). The Jacobian in our case is a TN TN diagonal matrix with the elements .
Then, the log-likelihood function to be estimated is
(4).
The maximisation of the log-likelihood function in (13) provides the ML estimates of the parameters in f(x), g(x)and q(x), as well as of (Greene, 2003).
Technical inefficiency measures of individual producers in a particular year can be calculated from employing the result in Jondrow et al., (1982):
(5)
where and .
3Estimation and Empirical Results
3.1Data and estimation
The model is estimated using balanced panel data of 443 large agricultural enterprises from three Russian regions (70 farms from Oroel, 180 farms from Krasnodar and 193 from Samara). The data set covers the period from 1996 to 2001. To enable a more accurate assessment of the dependence of production on weather conditions, attention is focused on crop production. All enterprises included in the sample are large-scale farms with a total cropped area of more than 200 hectares that intensively grow grain for commercial use. The sample represents between 22 and 45 per cent of the total crop area in the individual regions. According to experts, Krasnodar and Samara are regions with a high exposure to natural hazards, and Samara and Oroel belong to a small group of Russian regions that have recently been very active in introducing Western production technologies (Schüle and Zimmermann, 2002).
Production output is measured as annual farm revenues from crop production, plus the value of unsold grain (Y)[1]. The mean output function is a function of the area of sown land (Land), labour, defined as the annual average number of employees involved in crop production (Labour), the value of depreciation, machinery maintenance and fuel costs in crop production as a proxy for capital (Capital), materials costs (Materials) and time trends (t and t2) as indicators of technical change.
Technical inefficiency depends on the same variables as mean production. Technical inefficiency might also be related to the farmers’ educational background and experience, but these data were not obtainable, and hence no corresponding indicators were introduced into the model. The quality of the management, however, does influence the allocation of inputs, and must therefore be partially reflected in the input usage decisions.
To enable a more precise assessment of various inputs’ effects on production risk, some components of materials costs, such as seed costs (Seed), fertiliser costs (Fertiliser) and other production costs (Supplemental input)[2], were considered individually in the production risk function. The variables Land, Labour and Capital specify this function as well.
All monetary data were measured in 1,000 roubles and adjusted to the year 2001 using the regional price indices for agricultural inputs and output as provided by Goskomstat (Goskomstat, 2002a and b, 2003a and b). However, for fertiliser and capital, these indices were not obtainable. In the case of fertiliser, values were adjusted by employing the fertiliser price index at the country level. To adjust capital costs, the country-level price index for machinery in crop production was applied. Maintenance and fuel costs were adjusted by the regional index for aggregated agricultural inputs.
3.2Estimation Results
Table 1 presents the coefficient estimates for 3 regional samples. Most parameter estimates are significant, with the exception of some cross-product variables in the mean production function, and some coefficients of the risk and inefficiency function.
Since the mean production function is of minor interest in our context, parameter estimates will be discussed only briefly. As documented by the production elasticites in Table 1, i.e., the coefficient for Land, Labour, Capital and Materials, the regions operate with significantly different technologies. Especially remarkable is the difference in production elasticity for land, ranging from 0.46 in Oreol to 0.30 and 0.34 in Krasnodar and Samara, respectively[3]. Since Oreol is one of the Russian regions with the most productive soils, this finding is consistent with the empirical evidence.
Table 1: Parameter estimates1)
Krasnodoar / Oreol / SamaraMean Production Function / constant Krasnodar / 1.08 / --- / ---
constant Oroel / --- / 1.20* / ---
constant Samara / --- / --- / 1.06***
Land / 0.30*** / 0.46*** / 0.34***
Labour / 0.14 / 0.11*** / 0.10***
Capital / 0.09*** / 0.13* / 0.20***
Materials / 0.47*** / 0.52*** / 0.53***
t / 0.05 / -0.11* / -0.13***
t2 / 0.00 / 0.01 / 0.03***
Land x t / -0.03 / -0.08*** / -0.04*
Labour x t / 0.03** / 0.02*** / -0.02***
Capital x t / 0.02** / 0.05*** / 0.02***
Materials x t / 0.00 / 0.02 / 0.03***
Land2 / 0.11* / 0.18*** / -0.02
Labor2 / 0.11*** / 0.11*** / 0.08***
Capital2 / 0.04*** / 0.05*** / 0.05***
Materials2 / 0.09*** / 0.16*** / 0.15***
Land x Labour / 0.05 / 0.04 / -0.08
Land x Capital / -0.04 / -0.07 / 0.10**
Land x Materials / -0.10** / -0.27*** / -0.13**
Labour x Capital / -0.08*** / -0.03 / -0.04*
Labour x Materials / -0.14*** / -0.09* / 0.03
Capital x Materials / 0.00 / -0.09*** / -0.12***
Production Risk Function / dummy 1996 / 0.37*** / 0.33*** / 0.50***
dummy 1997 / 0.32*** / 0.40*** / 0.60***
dummy 1998 / 0.32*** / 0.47*** / 0.29***
dummy 1999 / 0.40*** / 0.36*** / 0.36***
dummy 2000 / 0.30*** / 0.30*** / 0.40***
dummy 2001 / 0.25*** / 0.33*** / 0.36***
Land / 0.13 / 0.00 / 1.03
Labour / 0.03 / 0.39*** / -0.13***
Capital / 0.22*** / 0.34*** / -0.04***
Seed / 0.07 / -0.02 / 0.15***
Fertiliser / 0.46*** / 0.04*** / 0.03***
Supplemental input / 0.00 / 0.05*** / -0.01
TI Function / Land / -0.01 / 0.00 / -0.35***
Labour / 0.31 / 0.21 / 0.13
Capital / 0.04 / -0.12 / 0.36***
Materials / 0.38*** / 1.44*** / 1.15***
T / -0.07* / -0.45* / 0.11***
σu’ 2) / 0.42*** / 0.32** / 0.15***
1)*, ** and *** denote significantly different from zero at the 10, 5 and 1 % levels.
2)σu’ = constant*σu..
Source: own calculations
3.2.1Productivity
The estimates suggest a positive impact of technical change during the sample period for two of the three regions, Samara and Oreol (Figure 1). Only in Krasnodar is the impact of technical change not apparent. Moreover, the parameters for technical change are not significantly different from zero in this region. The most dynamic region was Samara, which corresponds to the observation that this region was very successful in the last decade in introducing minimum-tillage technologies.
With regard to the bias of technical change, there is evidence of both land-saving and capital-using technical change. In Samara, the capital-using impact was accompanied by labour-saving change, which indicates that the farms in this region would be expected to release redundant labour. A different process, though, seems to occur on the farms from the Krasnodar and Oreol regions, where the impact of technical change on capital use does not cause labour-saving. Under market conditions, the opposite should be expected.
3.2.2Efficiency
According to the estimates, input use influences technical efficiency. Material inputs are efficiency decreasing in all regions, while the other inputs show divergent patterns. In Krasnodar and Oreol they were not significant, whereas in Samara, land input had a positive influence on efficiency while capital had a negative impact, indicating the under-utilisation of this factor. In addition, there are indications that inefficiency evolved differently over time. The trends in farms' technical efficiency between 1996 and 2001 are presented in Table 2. The estimates suggest that technical efficiency increased in both Oreol and Krasnodar, while in Samara no improvement in farms' technical efficiency took place during that period.
Table 2:Changes in technical efficiency of the sample farms (1.0 = 100 percent efficiency)
Year / Annual average score*Krasnodar / Oroel / Samara
1996 / 0.97 / 0.71 / 0.88
1997 / 0.99 / 0.71 / 0.91
1998 / 0.99 / 0.79 / 0.87
1999 / 0.99 / 0.83 / 0.86
2000 / 1.00 / 0.96 / 0.87
2001 / 1.00 / 0.97 / 0.85
* Estimated by (5)
Source: authors’ calculations.
3.2.3Production Risk
Two sources of risk are considered in the model; systemic risk, represented by the dummy variables for the individual years, and idiosyncratic risk due to the firms’ input choice. The parameters of systemic risk are highly significant for all three regions. In addition, the values are relatively large, which implies that a considerable portion of the output variation can be attributed to systemic risk in the selected regions. Moreover, the estimates for Krasnodar and Oreol show a similar annual pattern, while Samara appears, conversely, to be affected in the individual years. This suggests that either weather conditions were relatively different in Samara, or that the production technologies are differently suited to capture risk affects.
Different effects of idiosyncratic risk were also estimated in Samara compared to the other two regions. This not only concerns the significance of the estimates but also the sign of the coefficients. Risk-reducing effects of inputs (negative signs) were only estimated in Samara for capital and labour. In the other two regions, these inputs were risk-increasing. Material inputs tend to have a risk-increasing affect in all regions. Moreover, the estimation results do not confirm the view that pesticides are a yield-stabilising factor (Quiggin and Chambers, 2003).
In addition, the variance of output defined as σ2 = {exp(Dt)g(x)}2+ q(x)2σu2 is explained mostly by variance due to production risk. For most farms in all regions q(x)2σu2 {exp(Dt)g(x)}2, i.e., according to the model estimates, production risk contributes considerably to the volatility of agricultural production[4].
4Conclusions
This study deals with the impact of production risk and technical inefficiency of agricultural producers in Russia. The results obtained from analysing the panel data of 443 farms from various Russian regions suggest that technical inefficiency enhances the variability of agricultural production in Russia. Moreover, according to the model estimates, production risk contributes considerably to the volatility of Russian agricultural production. For most farms, output variability is explained mainly by production risk. Thus, neglecting risk may cause incorrect technical efficiency estimates.
In Oreol and Krasnodar farm efficiency increased, while in Samara it remained stable at a relatively low level. However, the results suggest a correspondence between technical inefficiency and technical change. Under constant technology, it is reasonable to assume that firms learn from past experience and, thus, are on a path towards the best production practice. On the other hand, increases in factor productivity suggest a high degree of innovation adoption, which shifts the production possibility set outward. Thus, different farms can define the production frontier in subsequent years. This effect seems to be the case for the farms in Samara.
Additionally, the analysis demonstrates that farms respond only weakly to production risk: most production factors enhance farms' production volatility. This implies that the current factor endowment of Russian farms is not adjusted to production conditions. Hence, as production risk plays an important role in the development of agricultural production at this stage, farms have to search for options to improve their response to production risk, primarily with respect to the introduction of modern production technologies and practices which can reduce output volatility and enable more flexible factor use. Finally, further research is needed to analyse farmers' response to production risk. If agricultural producers do not exhibit risk-adjusting behaviour, the reasons behind this have to be analysed. In this regard, it would be necessary to model farmers' risk preferences and estimate their impact on input use explicitly.
References
Bezlepkina, I. and Oude Lansink, A.O. (2003). Liquidity and productivity in Russian agriculture: Farm data evidence. Proceedings of the 25th International Conference of Agricultural Economists (IAAE): 399-408. Durban, South Africa, 16-22 August.
Goskomstat of Russia (2003a). Agriculture in Russia. Moscow, Russia: Goskomstat, State Statistical Agency.
Goskomstat of Russia (2002a). Agriculture in Russia. Moscow, Russia: Goskomstat, State Statistical Agency.
Goskomstat of Russia (2002b). Russian statistical yearbook. Moscow, Russia: Goskomstat, State Statistical Agency.
Goskomstat of Russia (2003b). Russian regions. Moscow, Russia: Goskomstat, State Statistical Agency.
Jondrow, J., Lovell, C. A. K., Materov I.S. and Schmidt, P. (1982).On the estimation of technical inefficiency in the stochastic frontier production function model. Journal of Econometrics 19: 233-38.
Just, R. E. and Pope, R. D. (1978). Stochastic representation of production functions and econometric implications. Journal of Econometrics 7: 67-86.
Kumbhakar, S. C. (2002).Specification and estimation of production risk, risk preferences and technical efficiency. American Journal of Agricultural Economics 84: 8-22.
Liefert, W. (2002). Comparative (Dis?)Advantage in Russian Agriculture. American Journal of Agricultural Economics 84: 762-767.
Macours, K. and Swinnen, J. F. M. (2000). Causes of output decline in economic transition: the case of Central and Eastern European Agriculture. Journal of Comparative Economics 28: 172-206.
Morey, E.R. (1986). An introduction to checking, testing and imposing curvature properties: the true function and the estimated function. Canadian Journal of Economics 19: 207-235.
Quiggin, J. and Chambers, R. G. (2003). The state-contingent approach to modeling environmental risk management. In Babcock B. A., Fraser R. W. and Lekakis J.N. (eds) Risk Management and the Environment: Agriculture in Perspective: 11-28. Dordrecht: Kluwer Academic Publishers. [(12) please give place of publication]
Schüle, H. and Zimmermann J. (2002). Management information systems (MIS) for large agricultural enterprises in Russia. Proceedings of the International Congress of Farm Management Association (IFMA). Arnhem, The Netherlands, 7-12 July.
Serova, E. (2000). Russia’s agro-food sector. In Wehrheim, P., Froberg, K., Serova, E. and von Braun, J. (eds) Russia’s Agro-Food Sector, Kluwer Academic, Bonn: 81-106.
[1] The value of unsold grain (net inventory accumulation) was calculated as a difference between the farm's annual grain production and grain sales multiplied by the grain price in the year 2001.
[2] Whereas other costs are calculated as the difference between total crop production costs and costs of labour, seed, fertiliser, equipment and machine maintenance, and fuel. Usually they consist of costs of plant protection. Therefore, these costs could be considered as a proxy for pesticide and herbicide use.
[3] As all variables have been standardised by their geometric means, their logged values are zero at the sample means. Therefore, the partial production elasticities of each input at the sample means can be read from the Cobb-Douglas part of the mean production function.
[4] On average over the sample period, the production variance induced by production risk is greater than that induced by technical inefficiency in 113 (of 180) farms in Krasnodar, 62 (of 70) in Oroel and 165 (out of 193) in Samara. The average ratios of variance caused by production risk to whole output variance are 0.63, 0.87 and 0.85, respectively.