2008年-世界发展银行全球_Guatemala___Investment_Climate_Assessment_Volume_2_Background_Notes_on_Productivity_49页_2mb

Summary of Guatemala Investment Climate Assessment (Volume II: Background Notes on Productivity)

Core Content

This report presents a detailed analysis of the investment climate in Guatemala, focusing on its impact on productivity. The study is based on the Guatemala Enterprise Survey and includes methodological notes and productivity results. It highlights the challenges of handling missing data and endogeneity issues in econometric models.

Main Points

1. Methodological Notes

• Missing Data Handling:
  Missing data in production function variables is a significant challenge. To address this, the report introduces an ICA (Investment Climate Assessment) method, which replaces missing values with the expected values based on industry, region, and size dummies. This method increases the sample size by approximately 20% and reduces the percentage of observations lost.

  • For example, in the food sector, the percentage of observations lost was reduced from 26.3% to 5% after applying the ICA method.
  • The replacement strategy is based on the equation:
    $$
    E(J_i \mid D_{T,i}, D_{I,i}, D_{S,i}) = \rho_0 + \rho_{T,J} D_{T,i} + \rho_{T,J} D_{I,i} + \rho_{T,J} D_{S,i}
    $$
    where $J$ represents output, labor, materials, and capital.

• Bootstrap Standard Errors:
  The report also discusses the use of bootstrap standard errors (with 1500 replications) to improve the reliability of the standard errors in the regression models. This is particularly important as the replacement method may reduce variability and lead to downward-biased standard errors.

• Alternative Methods:

  • The complete case method (deleting observations with missing data) is less efficient and may reduce the significance of IC variables.
  • The Heckman selection model is considered but its Lambda is not significant, indicating that it may not be necessary for this dataset.
  • The ICA method is preferred for its parsimony and simplicity, even though it may not fully resolve endogeneity issues.

• Endogeneity Considerations:
  The report acknowledges the endogeneity problem, where explanatory variables might be correlated with the error term, leading to biased and inconsistent estimates.

  • Industry-region-size (I-R-S) averages are used as a potential correction, but their effectiveness depends on whether they are correlated with the error term.
  • Some variables, such as crude plant-level IC variables, are not well instrumented, so they are used as-is.
  • The models are interpreted in terms of conditional expectations rather than causal effects.

2. Productivity Results

• Variables Used:
  The study uses a wide range of variables to estimate productivity, including:

  • Output: Measured by sales (deflated by the Producer Price Index, base 2002).
  • Labor: Total number of permanent and temporary workers, and total hours worked per year.
  • Materials: Total costs of intermediate and raw materials (excluding fuel), deflated by PPI.
  • Capital: Net book value of machinery and equipment, deflated by PPI.
  • User cost of capital: Defined as 15% of the net book value of machinery and equipment.
  • Labor cost: Total expenditures on personnel, deflated by PPI.

• Dependent Variables:

  • Exports: A dummy variable indicating whether exports exceed 10% of total sales.
  • Foreign Direct Investment (FDI): A dummy variable indicating whether the firm has foreign capital.
  • Real wages: Calculated as total labor cost divided by the number of workers.
  • Employment: Total number of workers (permanent and temporary).

• Elasticities and Contributions:

  • The ICA elasticities and semi-elasticities are presented, showing the impact of investment climate variables on productivity.
  • The percentage contribution of IC and C variables to the Olley and Pakes decomposition is also analyzed.
  • Security costs are found to have a significant positive effect on productivity, while crime has a negative effect.
  • The percentage of female workers has a small negative impact on productivity, but the effect is not statistically significant in all cases.
  • IC block variables such as days to clear customs, electricity supply, and water outages are shown to have varying impacts on productivity.

3. Key Findings

• The ICA method significantly improves the representativeness and efficiency of the dataset by reducing the number of missing observations.
• Bootstrap standard errors provide more accurate inferences, especially when the replacement method reduces variability.
• The Heckman model is not necessary for this dataset, as the Lambda is not significant.
• The endogeneity of certain variables is acknowledged, but the ICA method is used due to its simplicity.
• The impact of IC variables is interpreted as marginal effects on the conditional expectation of productivity, not as causal relationships.
• Security is a major factor influencing productivity, with a positive contribution.
• Gender impact is minimal and not always statistically significant.
• Sampling differences between 2003 and 2007 data are noted, which could affect the interpretation of results.
• The Olley and Pakes decomposition helps in understanding the efficiency term and the contribution of IC variables to productivity.
• IC variables have varying impacts on employment, real wages, and export probability, with some showing significant contributions.
• The percentage of female workers has a small negative effect on log-productivity, but the magnitude is not always significant.

Conclusion

The report provides a comprehensive analysis of the investment climate in Guatemala and its effect on productivity. It emphasizes the importance of addressing missing data and endogeneity issues in the regression models. The ICA method is recommended for its ability to maintain sample representativeness and improve efficiency. The findings highlight the significant role of security, access to electricity, and water supply in influencing productivity, while also noting the limited impact of gender-related variables. The analysis underscores the need for further research to better understand the causal relationships and simultaneous effects of IC variables on firm performance.