^{1}

^{2}

The authors have declared that no competing interests exist.

Conceived and designed the experiments: XAR YHE. Performed the experiments: XAR YHE. Analyzed the data: XAR. Contributed reagents/materials/analysis tools: XAR YHE. Wrote the paper: XAR.

The main objective of this paper is to decompose the productivity growth of Egyptian cotton production. We employ the stochastic frontier approach and decompose the changes in total factor productivity (CTFP) growth into four components: technical progress (TP), changes in scale component (CSC), changes in allocative efficiency (CAE), and changes in technical efficiency (CTE). Considering a situation of scarce statistical information, we propose four alternative empirical models, with the purpose of looking for convergence in the results. The results provide evidence that in this production system total productivity does not increase, which is mainly due to the negative average contributions of CAE and TP. Policy implications are offered in light of the results.

The agricultural sector is considered as one of the major sectors in Egypt’s economy, where its contribution to the country’s GDP is currently around 14% [

The production of cotton has great importance in the Egyptian agricultural sector. First, because cotton production is a very intensive activity in the use of labor and it is carried out principally by small family exploitations. Additionally, it generates new opportunities of employment in the subsequent processes (e.g. ginning, transportation, commercialization and the local textile industry). Second, cotton production is important in Egypt because it helps to guarantee the food security for population [

Considering the importance of cotton production in Egypt, two issues are surprising. First, there are not empirical studies on this sector that analyze the evolution of total factor productivity (TFP) from its determinants. Second, a substantial decline was observed in cotton production (period 2004–2008), while the production of other agricultural products, such as sugarbeet, fruits, vegetables, rice and wheat, increased in Egypt in a remarkable way [

Some studies on cotton in Egypt, such as [

In view of the above considerations, the main objective of this study is to carry out a disaggregate analysis (by provinces) of TFP in Egyptian cotton production. In addition, it aims to analyze the evolution of TFP from its determinants (including the efficiency). On the basis of these results, this paper aims to provide evidence on the possible problems affecting the cotton production system in Egypt. One of the peculiarities of this study is the use of scarce data. To overcome this limitation, we propose four alternative empirical models to compare the results obtained. If the results obtained by the different models are similar, they can be considered to be more consistent and meaningful.

To meet our objectives the rest of the paper is organized as follows. In the subsequent section we present a brief review of our methodology. Section three deals with the relationship between the statistical information available and the empirical models proposed. Section four presents the results. The final section presents the main conclusions and policy implications.

Solow [

There exist two main types of approaches that can be applied to estimate efficiency and decomposing productivity growth: data envelopment analysis (DEA) and stochastic frontier analysis (SFA) [

Following previous literature in the agricultural field (e. g., [_{it}_{it}_{it}_{it}_{it}

In ECM, the inefficiency term _{it}_{i}

In TEEM, the non-negative random variables (_{it}_{it}

Differentiating the production technology in _{j} = ∂_{j} is the elasticity of output with respect to the

Differentiating the log of Q in

In

To examine the effects of technical progress and changes in efficiency on TFP growth, the traditional definition for productivity total growth is used; that is, TFP is taken as the residual resulting from the output growth unexplained by input growth:
_{j}

By substituting _{j}_{j}

In _{j}ẋ_{j}

The interpretation of the four components in

In this study we use panel data at the province level covering the period 1990–2008. The dataset was obtained from “Agricultural Statistics”, a publication by the Ministry of Agriculture and Land Reclamation [

The most important problem from the empirical point of view is that, in this publication, the only real and direct data provided are the production (Q) and the cotton area (X_{A}). The other inputs—labor input (X_{L}), capital input (X_{K}) and materials input (X_{M})—can be calculated according to the input rate per hectare for every year (also published by the same source):
_{t}, k_{t} and m_{t}

Additionally, there is a lack of province-level statistical information on specific variables such as education, credit use, infrastructures, land quality, average size of plantations, composition of labor or characteristics of machinery, which may have a significant impact on the behavior of efficiency and productivity. There is also no data at the farm level, such as age, farm income, off-farm income, womens’ participation, experience or farm size.

Production (Q) | Tons (thousands) | 3.01 | 221.14 | 68.30 | 45.00 |

Cotton Area (X_{A}) |
Hectares (thousands) | 1.54 | 71.56 | 28.40 | 17.78 |

Labor (X_{L}) |
Workers (thousands) | 25.91 | 1176.35 | 386.55 | 253.24 |

Capital (X_{K}) |
Hours (thousands) | 100.14 | 5195.77 | 1778.59 | 1209.67 |

Materials (X_{M}) |
Tons (thousands) | 0.34 | 15.67 | 6.22 | 3.89 |

Given that the quantities of the productive factors (labor, capital and materials) are estimated from the data on the cotton area, one possibility to verify that this procedure of estimation does not affect the final results significantly consists in using two alternative stochastic frontier production functions (instead of only one with four inputs). One of the stochastic frontier production functions includes only the input which is measured directly, X_{A}:
_{A}:

From the two alternative stochastic frontier production functions (9 and 10), we estimate four alternative models:

Model A1: It uses the frontier production function (A) and the ECM specification (

Model A2: It uses the frontier production function (A) and the TEEM specification (

Model B1: It uses the frontier production function (B) and the ECM specification (

Model B2: It uses the frontier production function (B) and the TEEM specification (

In order to estimate the four models and decompose TFP growth (

Since in this study the statistical information on province-specific variables is not available, we create dummy variables (_{i}_{i} being equal to 1 if the province is

From the estimation of the translog stochastic frontier production function (

The Maximum Likelihood estimates for the parameters of the four alternatives models can be obtained by using the FRONTIER 4.1 program [

Constant | 0.249289 | (0.114530)** | 0.318254 | (0.1267098)** | 0.289005 | (0.068000) |
0.339955 | (0.074938) |

ln(XA) | 0.563986 | (0.072296) |
0.699767 | (0.037490) |
||||

ln(XL) | -0.015121 | (0.191079) | -0.035219 | (0.160326) | ||||

ln(XK) | 0.090003 | (0.262512) | -0.267552 | (0.228417) | ||||

ln(XM) | 0.444902 | (0.426944) | 1.042037 | (0.367086) |
||||

T | -0.051286 | (0.014198) |
-0.065920 | (0.017786) |
-0.056645 | (0.009873) |
-0.032985 | (0.013500)** |

½[ln(XA)]² | 0.056628 | (0.065249) | 0.107206 | (0.050583)** | ||||

½[ln(XL)]² | -1.326062 | (1.328960) | -2.713082 | (1.109566)** | ||||

½[ln(XK)]² | -0.188999 | (1.253958) | -1.498155 | (1.117122) | ||||

½[ln(XM)]² | -2.452777 | (2.239771) | -5.212243 | (2.014632) |
||||

½[t]² | 0.009530 | (0.002870) |
0.017803 | (0.004239) |
0.010833 | (0.002081) |
0.014033 | (0.003458) |

ln(XL)ln(XK) | -0.058979 | (0.609509) | -0.398795 | (0.579186) | ||||

ln(XL)ln(XM) | 1.348156 | (1.448512) | 3.092126 | (1.210854)** | ||||

ln(XK)ln(XM) | 0.683495 | (1.567905) | 2.067456 | (1.380572) | ||||

ln(XA)(t) | 0.062670 | (0.008544) |
0.054666 | (0.005415) |
||||

ln(XL)(t) | -0.118374 | (0.059259)** | -0.157438 | (0.056790) |
||||

ln(XK)(t) | -0.025466 | (0.076149) | -0.005993 | (0.061207) | ||||

ln(XM)(t) | 0.195844 | (0.122160) | 0.208574 | (0.107433)* | ||||

Constant | -0.370452 | (0.375941) | -1.063039 | (0.601952)* | ||||

D2 | -0.298510 | (0.342997) | -1.005989 | (0.564093)* | ||||

D3 | -0.386331 | (0.365466) | -1.055583 | (0.579813)* | ||||

D4 | -0.285047 | (0.350853) | -0.981394 | (0.572738)* | ||||

D5 | 0.310457 | (0.240579) | -0.147981 | (0.362236) | ||||

D6 | -0.338407 | (0.298452) | -0.858379 | (0.477231)* | ||||

D7 | 0.149254 | (0.271432) | -0.339536 | (0.399204) | ||||

D8 | 0.421313 | (0.225923)* | 0.005871 | (0.332894) | ||||

D9 | -0.077183 | (0.321359) | -0.634279 | (0.464208) | ||||

T | -0.120472 | (0.026703) |
-0.154287 | (0.045678) |
||||

Sigma-squared | 0.0705182 | (0.009734) |
0.055808 | (0.006899) |
0.154840 | (0.040440) |
0.212099 | (0.067683) |

Gamma | 0.062285 | (0.088807) | 0.052890 | (0.067910) | 0.888835 | (0.037669) |
0.926244 | (0.029062) |

Mu | 0.038352 | (0.093079) | 0.047633 | (0.055249) | ||||

Eta | 0.134596 | (0.030282) |
0.151787 | (0.030046) |
||||

Log likelihood function | -23.082432 | -6.236776 | -0.725175 | 15.178791 | ||||

LR test of the one-sided error | 24.227818 | 32.570919 | 68.942331 | 75.402053 | ||||

Total number of observations | 171 | 171 | 171 | 171 |

Note: ***, ** and * indicate significance at 1, 5 and 10% level, respectively

All variables appearing in natural logarithms were divided by their geometric mean prior to estimation. The time trend was at zero in 1999. As a result, the coefficients of the first-order terms of the variables in natural logarithms can be interpreted as production elasticities in that year evaluated at the geometric mean of the explanatory variables.

The first-order coefficients, _{j}

The technical progress coefficient, _{t}

The dummy variable coefficients of some provinces are statistically significant at the 10% level for model B2, which confirms that there are some province-specific effects. The negative and statistically significant coefficient for the time variable (

1990 | 0.519 | 0.438 | 0.378 | 0.468 | 0.378 | 0.519 | 0.451 |

1991 | 0.557 | 0.482 | 0.420 | 0.565 | 0.420 | 0.565 | 0.506 |

1992 | 0.594 | 0.526 | 0.575 | 0.547 | 0.526 | 0.594 | 0.561 |

1993 | 0.629 | 0.569 | 0.710 | 0.745 | 0.569 | 0.745 | 0.663 |

1994 | 0.663 | 0.611 | 0.478 | 0.730 | 0.478 | 0.730 | 0.621 |

1995 | 0.695 | 0.650 | 0.491 | 0.517 | 0.491 | 0.695 | 0.588 |

1996 | 0.725 | 0.687 | 0.721 | 0.825 | 0.687 | 0.825 | 0.740 |

1997 | 0.752 | 0.722 | 0.767 | 0.808 | 0.722 | 0.808 | 0.762 |

1998 | 0.778 | 0.754 | 0.634 | 0.786 | 0.634 | 0.786 | 0.738 |

1999 | 0.802 | 0.783 | 0.749 | 0.853 | 0.749 | 0.853 | 0.797 |

2000 | 0.823 | 0.809 | 0.788 | 0.848 | 0.788 | 0.848 | 0.817 |

2001 | 0.843 | 0.832 | 0.801 | 0.830 | 0.801 | 0.843 | 0.827 |

2002 | 0.860 | 0.854 | 0.879 | 0.898 | 0.854 | 0.898 | 0.873 |

2003 | 0.876 | 0.872 | 0.903 | 0.909 | 0.872 | 0.909 | 0.890 |

2004 | 0.891 | 0.889 | 0.882 | 0.882 | 0.882 | 0.891 | 0.886 |

2005 | 0.903 | 0.903 | 0.905 | 0.899 | 0.899 | 0.905 | 0.903 |

2006 | 0.915 | 0.916 | 0.900 | 0.920 | 0.900 | 0.920 | 0.913 |

2007 | 0.925 | 0.927 | 0.879 | 0.901 | 0.879 | 0.927 | 0.908 |

2008 | 0.934 | 0.937 | 0.895 | 0.914 | 0.895 | 0.937 | 0.920 |

Mean(1990–2008) | 0.773 | 0.745 | 0.724 | 0.781 | 0.724 | 0.781 | 0.756 |

Rate^{a} |
3.318 | 4.315 | 4.905 | 3.789 | 3.318 | 4.905 | 4.082 |

^{(a)}Annual average percentage growth rate (1990–2008)

For the four models, the estimated annual levels of technical efficiency are similar. With respect to the average level of efficiency of the period 1990–2008, the estimates vary from a minimum level of 0.724 (model A2) to a maximum level of 0.781 (model B2), and the average of the four models is 0.756.

The four models consistently show that technical efficiency improves during the considered period. The average annual percentage growth rate is also similar for the four models, varying from a minimum rate of 3.318% (model A1) to a maximum rate of 4.905% (model A2), and with the average annual growth rate across the four models being 4.082%.

Results of TFP change decomposition for the four models are reported in

TP
-0.0477
-0.0617
-0.0522
-0.0267
CSC
0.0346
0.0330
0.0248
0.0263
CAE
0.0000
-0.0114
0.0000
-0.0357
CTE
0.0230
0.0239
0.0287
0.0248
CTFP
0.0100
-0.0163
0.0013
-0.0113

^{(a)}Mean changes for the total sample (1990–2008)

The four models offer similar results in the sense that the four models indicate the negative contribution of the technical progress component and the positive contributions of technical efficiency and scale components. In addition, models B1 and B2 coincide in the negative contribution of the allocative efficiency component.

The four models have similar results for the average magnitude of the contribution of the different components. The average negative contribution of TP varies from a minimum of -2.67% (model B2) to a maximum of -6.17% (model B1). The average positive contribution of CSC varies from a minimum of 2.48% (model A2) to a maximum of 3.46% (model A1). The average positive contribution of CTE is practically identical in the four models and around 2.5%. The average negative contribution of CAE is also similar for models B1 and B2, and it is higher for the model that considered the specific effects of inefficiency (-3.57% for model B2 compared to -1.14% for model B1).

The main difference in the results is due to the fact that models A1 and A2 are more restricted. These models only include one productive factor and therefore they do not allow to quantify the behavior of allocative efficiency. In these models the allocative efficiency is assumed not to change, and in this case a slight average productivity growth is estimated (1% for model A1 and 0.13% for model A2). On the other hand, taking into account the possible changes in allocative efficiency (models B1 and B2) the estimated average change in TFP is negative (-1.63% for model B1 and -1.13% for model B2).

The negative contribution of the technical progress component must be considered in light of its definition as a residual. A negative value of this component must be interpreted as a downward movement of the production frontier. But this does not mean that in average there was no technical progress in cotton production; what this result probably indicates is that its contribution was small, and that there are possibly other factors (which have not been explicitly considered in the model) whose negative effect on production outweighs the small positive effect of the possible technical advances (e.g. land quality, climate changes, plant diseases, plant viruses, etc).

The positive value for the changes in the scale component means that the cotton production sector took advantage of the economies of scale. With the statistical information available it is not possible to identify the specific factors responsible for this improvement in scale economies.

The changes in allocative efficiency exerted a negative effect on TFP growth (models B1 and B2). The presence of allocative inefficiency shows that, during the period of analysis, input prices were not equal to the value of their marginal product and thus these inputs were not allocated in the correct proportions; i.e., the input combination that minimizes the cost of production was not chosen.

The positive value for the changes in technical efficiency shows that the gap between the production frontier and the actual cotton production was squeezed throughout the analyzed period.

Finally, if we consider the more complete models (B1 and B2), it is possible to conclude that the progress in technical efficiency and economies of scale does not compensate for the negative effects on productivity caused by the misallocation of the productive factors (that is, the presence of allocative inefficiency) and for the negative effects of the other factors included in the technical progress component.

1991
-0.1042
0.0338
-0.1270
0.0974
-0.1000
1992
-0.1387
0.0676
-0.2629
-0.0180
-0.3520
1993
-0.1045
0.0010
-0.0243
0.1977
0.0699
1994
-0.0424
-0.0433
0.3500
-0.0149
0.2494
1995
-0.0566
-0.0061
-0.3830
-0.2137
-0.6593
1996
-0.0237
0.0269
-0.0591
0.3083
0.2524
1997
-0.0698
0.0117
-0.1928
-0.0168
-0.2677
1998
0.0271
0.0098
0.2508
-0.0216
0.2661
1999
-0.0684
0.1160
-0.3081
0.0667
-0.1938
2000
-0.0814
-0.0104
0.2772
-0.0054
0.1800
2001
-0.0379
-0.0753
-0.0953
-0.0176
-0.2260
2002
0.0910
0.0214
0.0717
0.0680
0.2521
2003
0.0315
0.1027
-0.2203
0.0110
-0.0751
2004
0.0199
-0.0861
0.0816
-0.0274
-0.0119
2005
0.0341
0.1204
-0.0900
0.0168
0.0813
2006
0.0256
-0.0608
0.0593
0.0215
0.0455
2007
0.0128
0.2256
-0.0689
-0.0189
0.1506
2008
0.0054
0.0192
0.0978
0.0129
0.1353
Mean (1991–2008)
-0.0267
0.0263
-0.0357
0.0248
-0.0113

^{(a)}Changes for the total sample

Although during the analyzed period the changes in TFP and its components are not high in average, the annual variations are much higher. This is largely due to the fact that the variability of the output and the inputs is also high (

The evolution of the changes in productivity and in its components does not show a clear trend, but in the last years a certain stabilization and a slight progress in the evolution of TFP can be observed.

TP is the component that experienced the least variability. This component presents a clear change of trend from the year 2002, when it started to contribute positively to the growth of productivity.

CTFP is the indicator that offers the highest annual variability, followed by the allocative efficiency component. The latter is also the component that has the greatest relative importance in determining the behavior of TFP.

Results of TFP growth decomposition by provinces are reported in

Dakahlia | -0.0054 | 0.0148 | -0.0216 | 0.0151 | 0.0028 |

Sharkia | -0.0214 | 0.0230 | -0.0329 | 0.0234 | -0.0079 |

Kafr Elshikh | 0.0058 | 0.0056 | -0.0310 | 0.0195 | -0.0002 |

Gharbia | -0.0297 | 0.0286 | -0.0402 | 0.0250 | -0.0163 |

Menoufia | -0.0649 | 0.0540 | -0.0523 | 0.0302 | -0.0331 |

Behairah | 0.0164 | 0.0085 | -0.0249 | 0.0086 | 0.0085 |

Beni Suef | -0.0436 | 0.0310 | -0.0443 | 0.0337 | -0.0231 |

Fayoum | -0.0536 | 0.0191 | -0.0360 | 0.0366 | -0.0340 |

Menia | -0.0437 | 0.0527 | -0.0384 | 0.0311 | 0.0017 |

Total simple | -0.0267 | 0.0263 | -0.0357 | 0.0248 | -0.0113 |

^{(a)}Mean changes for the period 1990–2008

Considering a situation of scarce statistical information, the focus of the study is novel in the sense that it proposes four alternative empirical models, with the purpose of looking for convergence in the results to validate to some extent the conclusions and to minimize the effect of working with a deficient database.

Indeed, the results of the estimation for the four models are consistent. The four models provide similar results for the technical efficiency (considering the total sample), and they all indicate the negative contribution of the technical progress component and the positive contributions of technical efficiency and scale components to total productivity growth. In addition, models B1 and B2 coincide in the negative contribution of the allocative efficiency component.

These results provide evidence that the cotton production system in Egypt offers some deficiencies. Although the levels of technical efficiency improve during the considered period, in this production system the total productivity does not increase. The results for model B2 show that the mean of CTFP of the total sample for the period 1990–2008 is -1.13% per year. The unfavorable evolution of productivity is mainly due to the negative average contributions of CAE (-3.57%) and TP (-2.67%). Therefore, this lack of productivity growth can be one of the causes of cotton production decline.

The allocative efficiency component is the one with the greatest relative importance in determining the negative behavior of TFP. The presence of allocative inefficiency provides evidence that, during the period of analysis, the inputs were not allocated in the correct proportions; i.e., the input combination that minimizes the cost of production was not chosen. This result might be related with the characteristics of the exploitations and the workers. If the farms are small (and in fact the average farm size in Egypt is about 0.6 hectare, which hinders the use of machinery) and the workers are inadequately trained, this might prevent the inputs from being used adequately and in the correct proportions [

The negative contribution of technical progress does not mean that in average there was no technical progress in cotton production; what this result probably indicates is that its contribution was small, and that there are possibly other factors whose negative effect on production outweighs the small positive effect of the possible technical progress. Another possible reason is that the existent technology might not be used appropriately.

This production system leads to varying results in the different cotton-producing provinces of Egypt, since the results for the changes in productivity and in its components are not homogeneous across them. This might suggest that some provinces have worse conditions for the production of cotton.

In light of these research results, and from the point of view of establishing an agricultural policy for the cotton production system in Egypt, some recommendations can be made:

Policy makers should improve the database of the cotton production sector. Only with a good database it is possible to get to know the productive reality of the cotton production sector appropriately, and a sufficient knowledge of this productive reality is required in order to establish an effective agricultural policy. This improved database should contain sufficient statistical information at the disaggregated levels (at provincial and farm levels). With this kind of statistical information it would be possible to identify the specific factors that influence allocative efficiency and technical progress.

Policy makers should aim to improve the total productivity of cotton with the purpose of reducing the production costs and increasing the degree of competitiveness of the Egyptian cotton production. It would therefore be useful to identify and to tackle the factors that cause the negative contributions of the allocative efficiency and the technical progress components to total productivity growth. Although these factors are not precisely identified due to the lack of statistical information, following previous literature in the agricultural field (e. g., [

Finally, policy makers should take into account that the behavior of productivity and its components is not homogeneous across provinces. Therefore, it seems reasonable to propose specific measures for each of them and to analyze the possibility of orientating production towards the provinces where productivity shows a relatively better behavior.