Abstract
This paper uses microeconomic panel data to examine differences in the cyclical variability of employment, hours, and real wages for skilled and unskilled workers. Contrary to conventional wisdom, it finds that, at the aggregate level, skilled and unskilled workers are subject to the same degree of cyclical variation in wages. However, the quality of labor input is found to rise in recessions, inducing a countercyclical bias in aggregate measures of the real wage. The paper also finds substantial differences across industries in the cyclical variation of employment, hours, and wage differentials, indicating important interindustry differences in labor contracting.
IN a recent paper, Reinhart and Reinhart (RR) provide an interesting perspective on the macroeconomic fluctuations in Colombia: a test of two opposing theories—the neoclassical synthesis versus a “real” technological or preference shock as the source of output changes. The empirical analysis was done using annual data over 1960–87. Following a battery of unit-root tests, variables (except the interest rate) were differenced to achieve stationarity. After experimentation with a wide combination of variables, a six-variable vector autoregressive (VAR) system was finally chosen involving money, real GDP, prices, wages, interest rates, and exchange rates. By imposing restrictions on the system, a neoclassical and a real business cycle structure were obtained. The results show that in the case of Colombia the neoclassical-Keynesian framework better describes output dynamics.
In view of the important monetary policy implications that could follow from these results, the purpose of this note is to reexamine the main findings in RR by using the Bayesian vector autoregression (BVAR) estimation technique (Litterman (1986)) to explain the dynamic between money, output, and prices using quarterly data for Colombia over 1976:1–1993:1. Mamingi (1992) has shown that causality results based on VAR can be misleading under temporal aggregation. Also, the use of BVAR allows the use of level data. It is well known that in the process of first differencing, one may lose important long- and medium-run information (Lutkepohl (1991), Maddala (1992)). Unfortunately, the causality relations found using the current methodology are quite different from the “three” dynamic properties of the Colombian economy reported by RR. However, responses of the variables to monetary shocks are similar.
I. BVAR Model Estimation
Quarterly data from Colombia were used for real GDP, money, consumer price index (CPI), exchange rate, nominal interest rate on three-month time deposits, and rural wages as measured by cost per hour (1975 = 100).1 GDP, money, CPI, and wages were seasonally adjusted. Except for the interest rate, logarithms of variables were used. Data prior to 1976:1 were used as initial observations that contain lags.
Results of univariate analysis of these seven series are shown in Table 1. All variables are characterized as I(1) processes with positive drift. The interest rate is the only exception, characterized as I(1) process without drift.
Six variables consisting of the logarithms of real GDP, money, CPI, exchange rate, rural wages, and the real interest rate were included in the final analysis. Using RATS and following the suggestion of Doan (1990), four lags were included and the relative values of the Theil U statistic for one to four steps ahead were used to choose among the alternative specifications of the model. The prior variances were adjusted until no more improvement in Theil U could be found (Doan (1990), Litterman (1986)). Figure 1 shows the causality relations between variables based on the F statistics (10 percent significance). Using the impulse command in RATS, the responses of the system of equations to a first-period standard error shock in money were obtained for eight quarters ahead of the final BVAR model. The results are depicted in Figure 2.
Time Series Properties of the Macroeconomic Variables
Time Series Properties of the Macroeconomic Variables
| Test for one-unit root | Test for two-unit root | ||||
|---|---|---|---|---|---|
| ADF with drift | 10 percent crit. val. | ADF with drift | 10 percent crit. val. | ||
| Seasonally adjusted variables | |||||
| Log money (LSMONEY) | 1.8585 | -2.5892 | -4.4277 | -2.5896 | |
| Log real GDP (LSGDP) | -0.9175 | -2.5903 | -2.6812 | -2.5907 | |
| Log price index (LSICPI) | 0.6037 | -2.5892 | -4.1137 | -2.5896 | |
| Log rural wages (LSGANF) | -0.0931 | -2.5892 | -3.2414 | -2.5896 | |
| Exchange rate (LDEV) | -1.5568 | -2.5896 | -4,3125 | -2.5899 | |
| Nominal interest rate (DTF) | -0.1426 | -1.6183 | -4.4969 | -1.6183 | |
| Real interest rate (RINT) | -1.1913 | -1.6183 | -5.6551 | -1.6183 | |
Time Series Properties of the Macroeconomic Variables
| Test for one-unit root | Test for two-unit root | ||||
|---|---|---|---|---|---|
| ADF with drift | 10 percent crit. val. | ADF with drift | 10 percent crit. val. | ||
| Seasonally adjusted variables | |||||
| Log money (LSMONEY) | 1.8585 | -2.5892 | -4.4277 | -2.5896 | |
| Log real GDP (LSGDP) | -0.9175 | -2.5903 | -2.6812 | -2.5907 | |
| Log price index (LSICPI) | 0.6037 | -2.5892 | -4.1137 | -2.5896 | |
| Log rural wages (LSGANF) | -0.0931 | -2.5892 | -3.2414 | -2.5896 | |
| Exchange rate (LDEV) | -1.5568 | -2.5896 | -4,3125 | -2.5899 | |
| Nominal interest rate (DTF) | -0.1426 | -1.6183 | -4.4969 | -1.6183 | |
| Real interest rate (RINT) | -1.1913 | -1.6183 | -5.6551 | -1.6183 | |
Causal Relationships Among the Variables
Citation: IMF Staff Papers 1993, 003; 10.5089/9781451930856.024.A009
Causal Relationships Among the Variables
Citation: IMF Staff Papers 1993, 003; 10.5089/9781451930856.024.A009
Causal Relationships Among the Variables
Citation: IMF Staff Papers 1993, 003; 10.5089/9781451930856.024.A009
Responses to Change in One Standard Deviation in Money
Citation: IMF Staff Papers 1993, 003; 10.5089/9781451930856.024.A009
Responses to Change in One Standard Deviation in Money
Citation: IMF Staff Papers 1993, 003; 10.5089/9781451930856.024.A009
Responses to Change in One Standard Deviation in Money
Citation: IMF Staff Papers 1993, 003; 10.5089/9781451930856.024.A009
II. Main Findings
Present results contradict the dynamic properties of the Colombian economy found in RR:
—Money exogeneity in the Granger sense is not observed. Exchange rates influence changes in money. Variations in the exchange rate reflect the international reserve level and show a strong correlation between it and monetary expansions (see Carrasquilla (1992)).
—As in RR, feedback among nominal magnitudes produces complicated dynamics in the entire price sector. In addition, complex movements are observed in the real sector since prices appear to influence real GDP.
—In contrast with RR findings, real GDP induces changes in exchange rates affecting the real interest rate. Again the external sector appears to be a very important factor not only in affecting monetary variables but also in touching the real sector. In this model co-movements between income and money could arise from past changes in income and money.
The impulse response analysis for money confirms the approximate long-run monetary neutrality and the tendency of the domestic price level to increase. However, as compared with RR, a smaller increase in real GDP is seen during the first three quarters followed by a slight decrease in GDP and a simultaneous increase in CPI. During the first four quarters, the nominal exchange rate depreciates by twice the increase in the CPI; in the second year, the CPI surpasses the nominal exchange rate two times over, suggesting a real exchange rate appreciation.
III. Conclusions
Although the causality relations among the macroeconomic variables are found to differ from the three dynamic properties of the Colombian economy reported by RR, responses of the variables to monetary shocks are similar in both cases.
The difference in the results could indicate structural changes in the Colombian economy owing to the liberalization of the economy and the elimination of exchange rate controls in Colombia during recent years. Loss of valuable information because of temporal aggregation and first differencing may also explain the anomaly.
REFERENCES
Carrasquilla, Alberto, “Acumulacion de reservas y politica macroeconomica: Colombia 1990–1992,” Macroeconomia de tos flujos de Capital en Colombia y America Latina (Bogota: Fedesarrollo, 1993).
Doan, T., “Regression Analysis Time Series RATS User’s Manual” (Evanston, III.: VAR Econometrics, 1990).
Litterman, Robert B., “Forecasting with Bayesian Vector Auto regressions—Five Years of Experience,” Journal of Business and Economic Statistics, Vol. 4 (1986), pp. 25–38.
Lutkepohl, Helmut, “Introduction to Multiple Time Series Analysis” (New York: Springer-Verlag, 1991).
Maddala, G.S., “Introduction to Econometrics” (New York: Macmillan, 1992).
Mamingi, Nlandu, “Aggregation and Test for Unit Roots and Co-integration” (unpublished; State University of New York, Albany, 1992).
Reinhart, Carmen M., and Vincent R. Reinhart, “Output Fluctuations and Monetary Shocks: Evidence from Colombia,” Staff Papers, International Monetary Fund, Vol. 38 (December 1991), pp. 705–35.
Maria E. García is a graduate student and Fulbright Scholar at the State University of New York-Albany and holds degrees from the Universidad Javeriana-Bógota and Universidad de los Andes-Bógota. The author is grateful to Professor Kajal Lahiri for helpful comments and suggestions.
Real GDP was obtained from the “Departamento Nacional de Planeacion.” Money (M1), the consumer price index, the wholesale price index (WPI), the coffee price, and exchange rate were obtained from the International Financial Statistics; oil prices from the Energy Information Administration; rural wages from the “Departamento Administrativo Nacional de Estadistica DANE”; nominal interest rate for time deposits (90 days) from the “Banco de la Republica”; and the real interest rate was computed as the nominal rate minus the inflation rate.
