2002年-世界发展银行全球_Money_Demand_in_Venezuela___Multiple_Cycle_Extraction_in_a_Cointegration_Framework_68页_2mb

Summary of "Money Demand in Venezuela: Multiple Cycle Extraction in a Cointegration Framework" by Mario A. Cuevas

Core Content

This working paper analyzes the behavior of money demand in Venezuela using both structural time series and vector error correction models (VECMs), focusing on the period from 1993.1 to 2001.4. The study aims to better understand the long-run and short-run dynamics of money demand, particularly in the context of external and internal economic shocks.

Main Views and Key Findings

• Money Demand Fluctuations: Real money balances (measured by M1) in Venezuela have experienced significant fluctuations, increasing by 41% from 1993 to 2001, yet still below the peak reached in 1997.
• Seasonal Cointegration: The preferred model features seasonal cointegration and was estimated using a structural time series approach. This model is robust to changes in opportunity cost variables.
• Common Cycles: A three-year cycle is found to be common across money demand, real GDP, and opportunity cost variables. This cycle is robust to model specification changes.
• Higher Frequency Cycles: Shorter cycles (higher frequency) are also identified, but they are more sensitive to changes in model specification.
• Model Comparison: The structural time series approach provides richer insights into short-run dynamics compared to the VECM approach, which is more focused on long-run behavior.
• Combined Approach: A merged approach is proposed, combining the strengths of both models. It uses VECMs to estimate long-run components and structural time series models to capture short-run dynamics.
• VECM Advantages: VECMs allow for the joint estimation of long- and short-run components and are relatively robust to measurement errors and residual heteroskedasticity. However, short-run dynamics in VECMs are heavily influenced by the specification of the autoregressive lag structure.
• Structural Time Series Model: The model represents the log of real money balances, real GDP, and inflation as a combination of stochastic trends, cycles, seasonal components, and innovations. The stochastic trend is modeled as a random walk with time-varying drift, while the cycles are modeled using trigonometric functions.
• Cointegration: The cointegration hypothesis is tested and supported, indicating that the variables share a common stochastic trend. The cointegrating vector is normalized to unity for the money demand variable.
• Empirical Results: The results from the structural time series model show that the estimated cycles have periods of approximately 2.78 and 2.85 years for the first model, and 1.71 and 2.97 years for the second model. All models show good fit, with R² values above 0.7 for real M1 and GDP, and above 0.3 for inflation.
• Statistical Tests: The model passes normality, heteroskedasticity, and autocorrelation tests at conventional critical levels. The goodness of fit is improved compared to random walk with drift models.

Key Information

• Modeling Approach: The study uses a structural time series approach to model money demand, which allows for the explicit estimation of short-run dynamics.
• Data: Quarterly data from 1993.1 to 2001.4 is used, with real M1 balances, real GDP, and inflation as the main variables.
• Estimation Tools: The models are estimated using the Kalman filter and the STAMP 6.0 software. The likelihood function is optimized to obtain parameter estimates.
• Cointegration: The presence of a common stochastic trend among variables is a key feature of the model, which is consistent with economic theory.
• Adjustment Dynamics: The structural time series approach provides a more transparent and flexible way to estimate short-run dynamics compared to the autoregressive framework used in VECMs.

Model Specifications and Results

• First Model: Uses log real M1, log real GDP, and inflation as endogenous variables. Estimated cycles have periods of 2.78 and 2.85 years. The log-likelihood is 271.43.
• Second Model: Uses the logs of real M1, real GDP, and inflation. Estimated cycles have periods of 1.71 and 2.97 years. The log-likelihood is 275.82.
• Goodness of Fit: The models show a significant improvement in fit compared to alternative models, with R² values of 0.82, 0.74, and 0.30 for the first model, and 0.85, 0.76, and 0.75 for the second model.
• Information Criteria: The Akaike and Bayesian Information Criteria support the model's specification, with values of -3.57, -5.29, -4.76 and -1.83, -3.55, -3.03 respectively.

Conclusion

The structural time series approach provides a more detailed and flexible analysis of short-run dynamics in money demand, while the VECM approach is effective in capturing long-run behavior. The combined approach offers a way to integrate both methods, leading to a more comprehensive understanding of money demand in Venezuela. The three-year cycle is a robust feature of the model, consistent across different specifications, suggesting a significant underlying economic pattern.