Trends in Total Factor Productivity in Indian Agriculture: State-level Evidence Using Non-parametric Sequential Malmquist Index

Recognizing the critical role of agricultural sector in the overall growth as well as development performance, this study estimates total factor productivity (TFP) in Indian agriculture at state-level. Using Index of Agricultural Production as the measure of output, changes in TFP are estimated using non-parametric Sequential Malmquist TFP index. The TFP change is decomposed into efficiency change and technical change. It is found that productivity improvements are marked in very few states, and so is technical change. The improvements in efficiency are observed to be low for most of the states and efficiency decline is observed in several states implying huge gains in production possible even with existing technology. In order to achieve higher productivity, it is essential to increase efficiency levels as well as achieve a more even spread of new technology. Section I: Introduction A rise in production can be attributed to a growth in inputs or growth in total factor productivity. The level of Total Factor Productivity (TFP) can be measured by dividing total output by total inputs. When all inputs in the production process are accounted for, TFP growth can be thought of as the amount of growth in real output that is not explained by growth in inputs. Productivity growth encompasses changes in efficiency as well as changes in the best practice. A firm is fully technically efficient if it is operating on the production frontier (i.e. it is achieving best practice), the production frontier being defined for a reference time period with reference to a particular set of firms. A rise in efficiency implies either more output is produced with the same amount of inputs or that less inputs are required to produce the same level of output. Equally, the outward shift of a production frontier implies productivity growth. There are several studies which point out decline in agricultural productivity in developing countries even in the years well-known for success of Green Revolution. The modified Malmquist TFP indexusing Sequential/ Long memory technology, as proposed by Forstner and Isaksson (2002) and Nin et al (2003), attempts to rectify the biases in computation of productivity growth arising from non-neutral technical change. This study uses non-parametric Sequential Malmquist TFP Index to estimate changes in total factor productivity in Indian agriculture at state-level. Section II: Literature Review Most of the studies that estimate agricultural total factor productivity in developing countries 1 The literature of getting technological regression in developing countries, even in those which are well-known for technical progress is quite vast for GDP but relatively less for agricultural TFP have found TFP to be declining even in the years which are well known for green revolution success arising primarily due to adoption of new and improved varieties of wheat and rice. Kawagoe et al. (1985), using data for 1960, 1970 and 1980 in 21 developed countries and 22 less developed countries, estimate cross-country production functions for 1970 and 1980. They find technological regression during both decades for the less developed countries, but technological progress in the developed countries. Kawagoe and Hayami (1985) use an indirect production function and find similar results in that data set. Fulginiti and Perrin (1993) estimate technical progress for LDCs for the period 1961-1985 using Cobb-Douglas production specification. The study reports technological regression for 14 of the 18 countries. It is possible, as suggested by the authors that interferences with the agricultural sector such as price policies had a depressing effect on incentives so as to stifle potential productivity gains. Fulginiti and Perrin (1998) use a parametric meta-production function and a non-parametric Malmquist index to examine the performance of the agricultural sectors in a set of 18 LDCs and find productivity regress in many of them. Trueblood (1996) uses non-parametric Malmquist index and also estimates CobbDouglas production function for 117 countries. The study also finds negative productivity growth in a significant number of developing countries. Arnade (1998) estimates agricultural productivity indices using non-parametric Malmquist index approach for 70 countries during the years 1961-1993. It is found that thirty six out of forty seven developing countries in the sample show negative rates of technical change. Kudaligama and Yanagida (2000), using deterministic and stochastic frontiers for 43 developed and developing countries over 1960, 1970 and 1980, indicate agricultural productivity for developing countries on a per farm basis deteriorated over the time period under consideration. Nin et al (2003) estimate TFP growth for 20 countries during 1961-1994 using non parametric Malmquist TFP index with an alternative definition of technologysequential technology and find that the earlier results reverse, and most of the developing countries experience productivity growth. Coelli et al (2003) estimate TFP for Bangladesh crop agriculture for the period 19611992 using stochastic frontier approach and find a decline in TFP over the period (mean TFP change = 0.9537). Rahman (2004) applies sequential Malmquist index approach to same dataset and finds TFP rising at the rate of 0.9% p.a and this growth is primarily led by those regions which have experienced high levels of Green revolution technology. Technical progress is found to be growing at 1.9% p.a that offsets declining efficiency at 1% p.a. Alene (2009) estimates TFP in African agriculture for the period 1970-2004 using both contemporaneous and sequential Malmquist TFP index. The study finds that while the conventional Malmquist method estimates aggregate TFP growth to be a modest 0.3% p.a (most of the stagnation of TFP growth is explained by technical regress), using sequential Malmquist approach the TFP is found to be rising at 1.8% p.a. There are a number of studies on the measurement of productivity that have been carried out for India as well. These studies pertain to agriculture sector and cropspecific analysis. There are few estimates available of TFP changes at state-level. A notable study in this regard is Fan, Hazell and Thorat (1998) which estimates TFP for agriculture at state-level using Tornqvist-Theil index for the period 1970-1994. The study finds that total factor productivity for India grew at an average annual rate of 0.69 percent between 1970 and 1995. In the 1970s, total factor productivity improved rapidly, growing at 1.44 percent per annum, grew faster in the 1980s at 1.99 percent per annum. But since 1990, total factor productivity growth in Indian agriculture has declined by 0.59 percent per annum. The study also reports state-level estimatesfor the whole period 1970 to 1994, the states with TFP growth rate in the range 0-1 percent per annum are Andhra Pradesh, Karnataka, Uttar Pradesh, Himachal Pradesh and Kerala; with TFP growth rate greater than one are Punjab, Bihar, Orissa, Maharashtra, West Bengal and J&K. The states with negative TFP growth are Haryana, Madhya Pradesh, Gujarat, Assam and Rajasthan. Kumar and Rosegrant (1994) estimate TFP growth for rice. They find that the TFP index has risen by around 1.85 per cent annually in the southern region (Andhra Pradesh, Tamil Nadu, Karnataka and Kerala), 0.76 per cent in the northern region (Haryana, Punjab and Uttar Pradesh) and 0.36 per cent in the eastern region (Assam, Bihar, Orissa and West Bengal). In the western region (Gujarat, Maharashtra, Madhya Pradesh and Rajasthan), the annual TFP growth was found to be negative but insignificant. Chand et al (2011) estimate crop-level TFP for the period 1986-2005 using DivisiaTornqvist index. They find highest TFP growth for wheat crop. Except wheat and groundnut, TFP growth during 1986-95 is found to be lower than 1975-1985 in all crops and for several crops during 1996-2005. The percentage of cropped area for different states is distributed as per TFP growth rates and they find that the states of Punjab, Gujarat and Andhra Pradesh have highest TFP growth with 90% or more of cropped area having TFP growth more than 1%. Tamil Nadu, Haryana, Uttar Pradesh, Maharashtra have cropped area distributed across all TFP growth categories 2 The TFP growth categories are formulated as follows: negative growth (less than zero), stagnant growth (00.5%), low growth (0.5-1%), moderate (1-2%) and high (greater than 2%). . The states of Madhya Pradesh, West Bengal, Bihar, Orissa, Karnataka, Kerala and Himachal Pradesh have larger percentage of cropped area reporting negative or stagnant TFP growth. Turning to studies on crop sector and crop-specific studies, Rosegrant and Evenson (1992) use Tornqvist-Theil index to estimate TFP change for Indian crop sector. They find rate of growth of TFP to be 1% for the entire period 1957-1985, 0.81% for the period 1957-1965, 1.22% during 1965-1975 and 0.98% during 1975-1985. Mukherjee and Kuroda (2001) use Törnqvist-Theil methodology to construct the TFP index for Indian agriculture in fourteen states from 1973 to 1993. They find TFP index to be 1.73 for 1973-79, 2.51 for 1980-89, 1.34 for 1990-1993 and 2.19 for entire period 1973-2003. Bosworth and Collins (2007) use growth accounting approach and estimate TFP growth in primary sector for India to be 0.8% during 1978-2004, 1% for the period 1978-1993 and 0.5% for the period 1993-2004. Murgai (1999) uses Tornqvist-Theil Index to estimate TFP growth in Punjab at district level during 1960-1993. TFP growth averaged 1.9 percent from 1960 to 1993. Productivity growth in Punjab is found to be lowest during the green revolution years, even as farmers moved from traditional varieties of wheat and rice to modern hybrid seed varieties and the agricultural sector experienced high growth rates in production. The study attributes most yield improvements to rapid factor accumulation, particularly that of fertilizers and capital. Contrary to widespread belief, the contribution of productivity growth to economic growth is found to be small. Rao (2005) uses Tornqvist-Theil index to estimate TFP changes for Andhra Pradesh across different crops for the period 1980-81 to 1999-2000. The study finds TFP growth rate for all the crops to be 0.23% in the pre-1990s period and -0.17% during the post-reform period. The corresponding percentages are found to be -0.02 and 0.91 for foodgrains and 0.41 and -1.06 for the non-foodgrains. Kumar and Mittal (2006) estimate TFP growth across different states for paddy and wheat. They find TFP of paddy has started showing deceleration in Haryana and Punjab but TFP of wheat is still growing in these two Green Revolution states. About 60 per cent of the area under coarse cereals is facing stagnated TFP. Similarly, the productivity gains which occurred for pulses and sugarcane during the early years of Green Revolution, have now exhausted their potential. Bhushan (2005) uses Data Envelopment Analysis to estimate Malmquist TFP index for major wheat producing states in IndiaPunjab, Haryana, Madhya Pradesh, Uttar Pradesh and Rajasthan. He finds TFP growth rate to be highest in Punjab and Haryana which is attributed to technical progress in these two states. Rajasthan (with no efficiency change) and Uttar Pradesh (with improvement in efficiency and negative growth in technological progress) have positive TFP growth rate while Madhya Pradesh (no change in efficiency and negative growth of technical progress) is reported to record negative TFP growth rate. As compared to 1980s, mean growth of TFP is found to be higher in 1990s and the primary source of TFP growth is technical progress and not efficiency improvements. Section III: Methodology This section begins by briefly describing the Malmquist TFP index and thereafter discusses the modified version of the index by using a different method to construct production frontier. Let the set theoretic representation of a production function that involves multiple outputs and inputs technology be described as the technology set S. Let x and y denote an N*1 input vector of non-negative real numbers and a non-negative M*1 output vector, respectively. The technology set is then defined as: S={(x,y): x can produce y} (1) This set consists of all input-output vectors (x,y) such that x can produce y. The piece-wise linear convex hull approach to estimate frontier was proposed by Farrell (1957) but the application of this methodology increased only after the term Data Envelopment Analysis was coined by Charnes, Cooper and Thodes (1978). Data Envelopment Analysis (DEA) is a non-parametric method of frontier estimation that makes use of linear programming. The approach constructs a “piece-wise surface (or frontier) over the data” (Coelli et al, 2005:162) such that the constructed frontier envelops all given data points, that is, all observed data points lie on or below the production frontier. The benchmark technology is hence constructed from among the observed input-output bundles of various production entities. “Efficiency measures are then calculated relative to this surface.” (Coelli et al 2005:162) A major advantage in the use of DEA in measuring productivity growth is that this method does not require any price data. This is a distinct advantage, because in general, agricultural input price data are seldom available and such prices could be distorted due to government intervention DEA uses Distance Functions that allow us to describe a multi-input, multi-output production technology without any specification of a behavioural objective (such as . The DEA seems to be a much more powerful tool for measurement of productivity since it also makes the least number of restrictive assumptions (no requirement of functional form of production function / distribution form of inefficiency) and at the same time permits decomposition of TFP change into two components of efficiency change and technical change that would help in gaining insights into the sources of growth of TFP. However, the disadvantage of DEA is that it does not account for noise (all noise is grouped into inefficiency) and the usual econometric tests of hypotheses and significance cannot be carried out. 3 Most of the literature mentions about the price distortions only in developing countries because of government intervention. However, this is true even for developed countries where, in fact, the quantitative levels of support by the government to the farmers are extremely high as compared to those provided by governments of developing countries. The deadlock in WTO over the issue of opening up agricultural markets and reducing government support is an evidence in point here. Hence the problem of obtaining reliable/ undistorted price data for agricultural sector is true for both developed as well as developing countries. cost-minimization or profit-maximization). The concept of distance function is closely associated with production frontiers. Distance functions can be outputoriented or input-oriented. An output distance function considers the maximum proportional expansion of the output vector corresponding to a given input vector. It measures the distance of a firm from its production frontierhow close a particular level of output is to the maximum attainable level of output that could be obtained from the same level of inputs if production is technically efficient. Fare, Grosskopf, Norris and Zhang (1994) define an output distance function at time t as