An empirical analysis of economic fluctuations in Japan: 1885–1940

An empirical analysis of economic fluctuations in Japan: 1885–1940

Japan and the World Economy 12 (2000) 11±19 An empirical analysis of economic ¯uctuations in Japan: 1885±1940 Shigeyuki Hamoria,*, Naoko Hamorib a F...

88KB Sizes 3 Downloads 47 Views

Japan and the World Economy 12 (2000) 11±19

An empirical analysis of economic ¯uctuations in Japan: 1885±1940 Shigeyuki Hamoria,*, Naoko Hamorib a

Faculty of Economics, Kobe University, 2-1, Rokkodai, Nada-Ku, Kobe 657-8501, Japan Faculty of Information Science, University of Marketing and Distribution Sciences, 3-1, Gakuen-Nishi-Machi, Nishi-Ku, Kobe 657-2188, Japan

b

Received 23 August 1998; received in revised form 17 May 1999; accepted 18 June 1999

Abstract This paper analyzes economic ¯uctuations in Japan's economy before World War II using the vector error correction model. It is found that causality from the real sector to the monetary sector can be observed but not vice versa in those years of Japanese economy. # 2000 Elsevier Science B.V. All rights reserved. JEL classification: O5 Keywords: Japanese economy; Cointegration; Vector error correction model

1. Introduction This paper is intended to empirically analyze business ¯uctuations in Japan using data from a period during the Meiji era until just before the outbreak of World War II. Japan was dramatically transformed following the Meiji Restoration in 1868, and succeeded in modernizing its entire society. Since the Restoration, Japan has introduced many Western in¯uences to rapidly develop its own economy. The early 20th century was a transition period for the Japanese economy. Chenery et al. (1962) pointed out that the year 1914 was a turning point, marking the change from growth balanced between agriculture and industry to an industry-driven economy. One characteristic of Japan's economy at that time was ®erce competition throughout the entire economy. Bankruptcies occurred frequently in every industry, including the banking sector, and price levels often dropped. This is *

Corresponding author. Tel./fax: ‡8166338-1255. E-mail address: [email protected] (S. Hamori). 0922-1425/00/$ ± see front matter # 2000 Elsevier Science B.V. All rights reserved. PII: S 0 9 2 2 - 1 4 2 5 ( 9 9 ) 0 0 0 2 8 - 6

12

S. Hamori, N. Hamori / Japan and the World Economy 12 (2000) 11±19

distinctly in contrast to the current economy, where even price mechanisms are not working satisfactorily, as criticized by other countries. Unfortunately, analysis using recent econometric techniques has not been made of the business ¯uctuations of the period1. This may be attributed to the fact that people have been interested only in the current challenges facing Japan's economy. However, it must be remembered that although Japan's modern economic growth started after World War II, the economy itself had been successful since the Meiji era by taking advantage of accumulations in every industry. Therefore, it would be meaningful to analyze the economy of Japan for the Meiji period. There are several approaches that can be taken in analyzing economic ¯uctuations. One is the calibration approach, which is often used in the real business cycle model (see Kydland and Prescott, 1982; Backus et al., 1992, for examples). This approach has the advantage that one can construct an economic model based on dynamic optimization behavior. However, it also has some weaknesses. For example, it is not easy to solve stochastic nonlinear equations quantitatively, in addition to which the results obtained by this method depend signi®cantly upon the approximation method used in the analysis (Taylor and Uhlig, 1990). The other is the application of time series analysis. Time series analysis itself consists of alternative approaches. One is the extension of the standard vector auto-regression (VAR) developed by Sims (1980). This approach excludes ad hoc restrictions on structural parameters and tries to obtain important information (e.g. exogeneity or causality) from statistical data. Another is the extension of structural VAR developed by Blanchard and Quah (1989) and Giannini (1992). This extension imposes long run or short-run restrictions on structural parameters and analyzes the performance of the model. As Hatanaka (1996, p. 124) pointed out, there is no unequivocal choice between standard VAR and the simultaneous equations model. One way to test economic theory is to identify the model using constraints derived from a portion of the theory and then to test the validity of the remainder of the theory against the observations. In contrast to the standard VAR, this structural VAR model achieves identi®cation of its parameters through economic theories, and thus is subject to Sims' criticism about the validity of economic theories. This paper adopts the standard VAR approach and analyzes what can be found about the economic ¯uctuations from the data, with special focus on the relationship between the real and the monetary sectors. The relationships among major macro variables suggest that there is a cointegrating relation among the real GNP, interest rate, money supply and the price level. It is also found that causality from the real sector to the monetary sector can be observed but not vice versa in those years of Japanese economy. 2. Characteristics of data The present study analyzes annual data from 1885 to 1940 using four variables: real GNP, prices, money supply and interest rates. The GNP de¯ator is used for prices and M1 is used for the money supply. Interest rates are represented by loan-on-deed rates in Tokyo. 1

Hamori and Kitasaka (1997) analyzed the business cycles of Japan for the post-war period.

S. Hamori, N. Hamori / Japan and the World Economy 12 (2000) 11±19

13

Table 1 List of variables Yt Mt Pt It

Real GNP at time t measured in logarithms Nominal money supply at time t measured in logarithms Price level at time t measured in logarithms Nominal interest rate at time t

Table 2 Summary statisticsa

Mean Std. Dev. Skewness Kurtosis Jarque±Bera p-value

DYt

DMt

DPt

DIt

0.032 0.037 ÿ0.182 2.247 1.602 0.449

0.085 0.116 0.631 2.464 4.309 0.116

0.037 0.079 0.530 3.463 3.066 0.216

ÿ0.098 0.906 0.056 3.573 0.781 0.677

a D is the difference operator, e.g., Dxt ˆ xt ÿ xtÿ1 .Skewness is a measure of asymmetry of the distribution of the series around its mean. Kurtosis measures the peakedness or flatness of the distribution of the series. Jarque±Bera is the test statistic for testing whether the series is normally distributed. p-value is the probability value associated with the Jarque±Bera test statistic.

The analysis uses logarithms for variables other than interest rates. Real GNP and GNP de¯ators are cited from Estimates of Long-term Economics Statistics of Japan since 1868, by Ohkawa et al. (1974). Data from Japanese Money Supply, by Fujino (1994) are used for the money supply. Table 1 indicates the symbol for each variable. Table 2 shows the summary statistics of variables. 3. Empirical results 3.1. Unit root test Each variable for the unit root is checked. To analyze data, a method introduced by Phillips and Perron (1988) is employed, which considers serial correlation among error terms2. Here, three types of analysis are performed: one including nothing, one including a constant term, and one including a constant term and a time trend. They are speci®ed as follows:

2

xt ˆ axtÿ1 ‡ ut ;

(1)

xt ˆ a0 ‡ axtÿ1 ‡ ut ;

(2)

We also performed the augmented Dickey±Fuller test and obtained the same results [Dickey and Fuller (1979, 1981)].

14

S. Hamori, N. Hamori / Japan and the World Economy 12 (2000) 11±19

and xt ˆ a0 ‡ a1 t ‡ axtÿ1 ‡ ut ;

(3)

where ut is an error term with mean zero and finite variance. The null hypothesis (H0) and the alternative hypothesis (H1) are shown as follows: H0 : a ˆ 1 and H1 : a < 1: The null hypothesis shows that each variable has a unit root, whereas the alternative hypothesis shows that no variable has a unit root. The unit root test for real GNP, price level and money supply is conducted using the second and the third speci®cation, whereas the unit root test for interest rates is conducted using the ®rst and the second speci®cation. The ®rst and the third order of lags are used for the lag truncation of the Phillips±Perron test. Tables 3 and 4 shows the results of the unit root test. As is clear from these tables, it is found that for each variable, the unit root is included at the level but not at the ®rst difference. 3.2. Cointegration test The Engle±Granger test and the Johansen test are used to analyze the cointegrating relationship among variables (Engle and Granger, 1987; Johansen, 1988; Johansen and Juselius, 1990). If there is no cointegrating relationship between the variables, the standard Table 3 Unit root test: Phillips±Perron test qa

a

Deterministic term None

Y

1 3 1 3 1 3 1 3 1 3 1 3 1 3 1 3

DY P DP M DM I DI a

ÿ4.511** ÿ5.045** ÿ3.385** ÿ3.346** ÿ3.361** ÿ3.613** ÿ1.168 ÿ1.286 ÿ7.549** ÿ7.874**

Constant

Constant and trend

0.587 0.606 ÿ7.269** ÿ7.269** ÿ0.657 ÿ0.711 ÿ11.470** ÿ16.180** 0.039 ÿ0.093 ÿ4.873** ÿ5.077** ÿ2.832 ÿ2.670 ÿ7.503** ÿ7.830**

ÿ1.549 ÿ1.684 ÿ1.586 ÿ1.805 ÿ1.757 ÿ2.128

Note: q is the number of lag truncation. D is the difference operator, e.g., Dxt ˆ xt ÿ xtÿ1 . Shows that the null hypothesis of unit root is rejected at 1% significance level; *shows that the null hypothesis of unit root is rejected at 5% significance level. **

S. Hamori, N. Hamori / Japan and the World Economy 12 (2000) 11±19

15

Table 4 Cointegration test: ADF testa Order of augmented terms

Test statistic

Critical value (5%)

Critical value (1%)

pˆ0

ÿ4.353*

4.22

4.75

a

Note: p is the number of lag truncation based on Schwarz criterion. (*) shows that the null hypothesis of non-cointegration is rejected at 5% significance level. Critical Values for 5% and 1% significance levels are obtained from Engle and Yoo (1987).

VAR is then used for analysis. However, if a cointegrating relationship exists, then the analysis requires the vector error correction model (VECM) that includes error correction terms. More speci®cally, the long-run equilibrium and the short-run dynamic adjustment process of the economic variables should be modeled. The long-run equilibrium is regarded as the cointegrating relation introduced by Engle and Granger (1987) and Johansen (1988). Hence, the long-run equilibrium and the short-run adjustment process are formulated according to the VECM. Table 5 shows the results of the Engle±Granger test. The number of lags is chosen to be zero based on the Schwarz criterion. The table shows that the null hypothesis of no cointegrating relationship is rejected. Table 6 shows the results of the Johansen test (trace test). This result also shows that the system has one cointegrating vector. Thus, we can carry out the analysis by assuming that the four-variable system has one cointegrating vector. The cointegrating regression is obtained as follows: Mt ˆ ÿ 5:621 ‡ 1:099Pt ‡ 1:171yt ÿ 0:0161It …0:281†

…0:050†

…0:079†

(4)

…0:012†

where the numbers in parentheses are standard errors. As is clear from Eq. (4), an increase of 1 percent in the price level corresponds to an increase in the nominal money supply by 1.099 percent. An expansion of 1 percent in real GNP corresponds to an increase in the nominal money supply by 1.171 percent. Finally, an increase of 1 percent in interest rates corresponds to a decrease in the nominal money supply by 0.016 percent.

Table 5 Cointegration test: Johansen test Null hypothesis rˆ0 r1 r2 r3 a

a

Alternative hypothesis r1 r2 r3 r4

Test statistic **

50.830 25.604 6.111 0.748

Critical value (5%)

Critical value (1%)

43.964 26.791 13.338 2.816

47.181 29.509 15.197 3.962

Note: r is the number of cointegrating vector. The lag order of VECM is chosen to be one based on Schwarz criterion. ** Shows that the null hypothesis is rejected at 1% significance level. Critical values for 5% and 1% significance levels are obtained from Osterwald-Lenum (1992).

16

S. Hamori, N. Hamori / Japan and the World Economy 12 (2000) 11±19

Table 6 Hypothesis testing on cointegrating coefficients cointegrating relation: M ˆ a ‡ bP P ‡ bY Y ‡ bI I a Null hypothesis

Alternative hypothesis

Test statistic (p value)

bP ˆ 0 bP ˆ 1 bY ˆ 0 bY ˆ 0 bI ˆ 0

bP 6ˆ 0 bP 6ˆ 1 bY 6ˆ 0 bY 6ˆ 1 bI 6ˆ 0

482.355 3.901 217.221 4.628 1.678

a

(0.000) (0.048) (0.000) (0.031) (0.195)

Note: The numbers in parentheses are p-value.

Based on this result, the following hypothesis tests are performed: (Test 1) H0 : bP ˆ 0 HA : bP 6ˆ 0; (Test 2) H0 : bP ˆ 1 HA : bP 6ˆ 1; (Test 3) H0 : bY ˆ 0 HA : bY 6ˆ 0; (Test 4) H0 : bY ˆ 1 HA : bY 6ˆ 1; (Test 5) H0 : bI ˆ 0 HA : bI 6ˆ 0; where bx is the regression coefficient of variable X. Each test statistic has a w2 distribution with one degree of freedom. Table 6 shows the result of each test. As is clear from the table, the null hypothesis is rejected for Test 1 through Test 4. On the other hand, it is found that the null hypothesis cannot be rejected for Test 5. 3.3. Causality test An analysis is performed based on the VECM that includes an error correction term according to the results of the cointegration test. Estimation results are given in Table 7. In this analysis, the number of lags is chosen to be one based on the Schwarz criterion. Based on the estimated VECM, causality among variables is analyzed. In the time series analysis,

S. Hamori, N. Hamori / Japan and the World Economy 12 (2000) 11±19

17

Table 7 Estimates of vector error correction modela Explanatory variables

Equations

ECtÿ1 DMtÿ1 DPtÿ1 Dytÿ1 DItÿ1 Constant R2

ÿ0.582 0.448 0.006 0.111 0.017 0.046 0.292

DMt

DPt (ÿ2.614) (2.195) (0.022) (0.236) (0.927) (2.280)

0.172 0.165 0.453 0.630 ÿ0.009 ÿ0.012 0.548

Dyt (1.430) (1.500) (3.071) (2.485) (ÿ0.976) (ÿ1.078)

ÿ0.054 0.029 0.055 ÿ0.082 ÿ0.002 0.030 0.040

DIt (ÿ0.657) (0.391) (0.543) (ÿ0.470) (ÿ0.375) (4.007)

0.471 0.577 2.421 2.368 0.002 ÿ0.269 0.089

(0.252) (0.337) (1.051) (0.599) (0.012) (ÿ1.579)

a

Note: ECtÿ1 is the error correction term shown as follows: ECtÿ1 ˆ 5.621 ‡ 1.000Mtÿ1 ÿ 1.099Ptÿ1 ÿ 1.171ytÿ1 ‡ 0.016Itÿ1. Numbers in parentheses are t-statistics

variance decomposition and impulse response function are often performed. However, they are crucially dependent on the ordering of variables. On the other hand, the causality test has the advantage of being independent of the ordering of variables. To explain the procedure of the hypothesis testing, let the equation of a variable DXi(i ˆ 1,2,3,4) be speci®ed as follows: DXit ˆ ai0 ‡ ai1 ECtÿ1 ‡ ai1 DX1tÿ1 ‡ ai2 DX2tÿ1 ‡ ai3 DX3tÿ1 ‡ ai4 DX4tÿ1 ‡ uit ; (5) where uit is the error term with mean zero and finite variance, variables Xit correspond to real GDP, price, money supply and interest rate at time t, and ECt is the error correction term at time t. The null hypothesis is that Xj(i 6ˆ j) does not Granger cause Xi. This null hypothesis means that ai1 ˆ aij ˆ 0, which shows that the coefficients of the corresponding variable and an error correction term are simultaneously zero. Other cases can be considered in the same manner. The test statistic has a w2 distribution with degrees of freedom equal to two. Table 8 shows the results and Fig. 1 illustrates the results. As the ®gure shows, interest rates and the real GNP have relatively high exogeneity in the system. The causality from interest rates to money supply is found. The causality from real GNP to money supply and

Table 8 Causality testa Explanatory variable Y Y M P I

8.943 (0.011) 6.812 (0.033) 0.568 (0.753) a

M

P

I

0.213 (0.899)

1.001 (0.606) 7.558 (0.023)

0.365 (0.833) 6.827 (0.033) 3.436 (0.179)

9.382 (0.009) 0.566 (0.754)

Note: The numbers in parentheses are p-value.

1.355 (0.508)

18

S. Hamori, N. Hamori / Japan and the World Economy 12 (2000) 11±19

Fig. 1. Illustration of causality.

price level can be observed. Finally, money supply and the price level are dependent on each other. 4. Conclusions The current study reports the results of a time-series analysis of several macro-variables of the Japanese economy before World War II. This analysis presents some interesting facts. The relationships among major macro variables suggest that there is a cointegrating relation among real GNP, interest rate, money supply and the price level. The causality test shows that the majority of ¯uctuations in the real economy are independent of the money supply. In contrast, the money supply and the price level are mutually affected. In particular, we note that real GNP, interest rates, and the price level in¯uence the money supply, which shows that money supply is endogenously determined. This fact may re¯ect the policy reaction function of the monetary authorities. Acknowledgements The authors would like to thank the Editor Ryuzo Sato and an anonymous referee for their helpful comments and encouragement.

References Backus, D.K., Kehoe, P.J., Prescott, F.E., 1992. International real business cycles. Journal of Political Economy 100, 745±775. Blanchard, O.J., Quah, D., 1989. The dynamic effects of aggregate demand and supply disturbances. American Economic Review 79, 655±673. Chenery, H.B., Shishido, S., Watanabe, T., 1962. The pattern of Japanese economic growth 1914±1954. Econometrica 30, 98±139. Dickey, D., Fuller, W.A., 1979. Distribution of the estimates for autoregressive time series with a unit root. Journal of the American Statistical Society 74, 427±431. Dickey, D., Fuller, W.A., 1981. Likelihood ratio statistics for autoregressive time series with a unit root. Econometrica 49, 1057±1072. Engle, R.F., Granger, C.W.J., 1987. Co-Integration and error correction: Representation, estimation, and testing. Econometrica 55, 251±276. Engle, R.F., Yoo, B.S., 1987. Forecasting and testing in co-integrated systems. Journal of Econometrics 35, 143± 159.

S. Hamori, N. Hamori / Japan and the World Economy 12 (2000) 11±19

19

Fujino, S., 1994. Japanese Money Supply, Keiso Shobo, (in Japanese). Giannini, C., 1992. Topics in Structural VAR Econometrics. Springer, Berlin. Hamori, S., Kitasaka, S., 1997. The Business Cycle in Post-War Japan An Empirical, Approach. Nova Science Publishers. Hatanaka, M., 1996. Time-Series-Based Econometrics, Oxford University Press, London. Johansen, S., 1988. Statistical analysis of cointegration vectors. Journal of Economic Dynamics and Control 12, 231±254. Johansen, S., Juselius, K., 1990. Maximum likelihood estimation and inference on cointegration: With applications to the demand for money. Oxford Bulletin of Economics and Statistics 52, 169±210. Kydland, F.E., Prescott, E.C., 1982. Time to build and aggregate fluctuations. Econometrica 50, 1345±1370. Ohkawa, K., Takamatsu, N., Yamamoto, Y., 1974. Estimates of Long-Term Economic Statistics of Japan since 1868, Toyo Keizai Shinposha, (in Japanese) Osterwald-Lenum, M., 1992. A note with quantiles of the asymptotic distribution of the maximum likelihood cointegration rank test. Oxford Bulletin of Economics and Statistics 54, 461±472. Phillips, P.C.B., Perron, P., 1988. Testing for a unit root in time series regression. Biometrica 75, 335±346. Sims, C.A., 1980. Macroeconomics and reality. Econometrica 48, 1±48. Taylor, J.B., Uhlig, H., 1990. Solving nonlinear stochastic growth model: A comparison of alternative solution methods. Journal of Business and Economic Statistics 8, 1±19.