Working Paper No. 81
Money, Exchange Rate, Price Links during
Hyperinflationary Episodes in Developing Economies
Using Hsiao’s approach to Granger non-causality test
Jean-Claude Maswana
Graduate School of Economics, Kyoto University, Japan
Email: maswana@econ.kyoto-u.ac.jp
August 2005
Abstract
The determination of the causal pattern among inflation, money growth, and exchange rate
has important implications for policymakers regarding appropriate stabilization policies in
developing economies. Using Congolese data where the pace of broad money growth and
hyperinflation (23,760% annual change) reached record levels in early 1990s, we use
single-equation multivariate autoregressive models with the optimal lag selected using
Hsiao’s approach to Granger causality. Results indicate feedback causality between
inflation and money growth on one side, and unidirectional Granger causality from money
growth to the exchange rate and from the exchange rate to inflation on the other. These
results suggest that the over-riding goal of disinflation needs to be accomplished initially by
exchange rate stabilization, followed by a direct inflation targeting.
Keywords: Money, exchange rate, hyperinflation, Hsiao’s Granger causality, Congo
1
1. Introduction
Inflation has always been a great concern of policy makers in Less Developed
Countries (LDCs) and analysts within international development agencies. Despite the fact
that the causal relations among inflation, money growth, and exchange rates have been well
studied in the literature, previous empirical studies have produced mixed and conflicting
results on the nature and direction of such links. Although the review of the literature on
inflation and its predictors is beyond the scope of the present analysis, it should be said that
inflationary analyses have been largely dominated by monetarist theories and to a certain
extent by structuralist views.
The first line of research (known as the monetarist approach) focuses on the causal
roles assigned to monetary growth. For instance, the theoretical explanation of inflationary
phenomenon by Cagan (1956) is largely developed within the monetarist view traditionally
represented by Milton Friedman, which asserts that inflation results from money supply in
excess of potential output or demand dictated by trade. As a variant, the fiscal-monetary
approach emphasizes the impact of rising government deficits as cause of expected money
supply growth, which in turn fuels the inflationary process (Sargent, 1973). One of the
crucial characteristics of the monetarist interpretation of inflation is that the increase in the
money supply precedes the rise in the price level.
The main alternative to the monetarist viewpoint is cost-push inflation, mainly
supported by a structuralist approach. This second view explains the role of exchange rate
(depreciation) in an inflationary process and in turning price increases into a vicious cycle.
In the context of LDCs where the share of imported intermediary inputs is large, initial
depreciation is likely to result in higher import prices, which then affect costs and
ultimately prices of products in the economy. Also, the structural dependency of capital
2
imports along with the lack of foreign reserves implies that developing countries have
recurrent balance of payments problems and that currency depreciation is endemic
(Vernengo, 2005). Not only is inflation seen as resulting from balance of payments crises,
but fiscal crises also are thought to be the result of the initial balance of payments crisis.
The effect of exchange rate in the inflationary process can be exacerbated under
conditions where the money supply reacts passively to any inflationary pressures. It is
possible for exchange rate depreciation to play a key role in sustaining the inflationary
spiral regardless of whether the process is initiated by internal rather than external factors
(Burdekin and Burketi, 1996).
Minshki (2004) provides an excellent reconciliatory remark in that sustained costpush inflation is also a monetary phenomenon because it cannot occur without the
acquiescence of the monetary authorities to a higher rate of money growth. Although
theoretically we can distinguish between monetarist and structuralist inflation, it is much
harder to do so in application since both types of inflation are associated with high rates of
money growth.
Studies of LDC inflation causality have focused on Latin-America rather than
inflationary episodes in sub-Saharan Africa, with the exception of Canetti and Greene
(1991). Using vector autoregression analysis to separate the influence of money supply
growth from exchange rate changes on prevailing and predicted rates of inflation in Africa,
Canetti and Greene (1991) find that both exchange rate movements and monetary
expansion affect consumer price changes in a number of Sub-Saharan African countries. In
particular, the two authors find that both the bivariate and trivariate Granger causality tests
suggest that exchange rates had a significant causal impact on prices in Sierra Leone,
Tanzania, and Democratic Congo (then Zaire), while monetary dynamics led inflationary
3
processes in Gambia and Uganda.
The causal nature of these relationships is known to exhibit considerable variation
across countries, and there is a call for empirical investigation for specific cases. The
present study investigates the Granger-causal links among the afore-mentioned variables in
the context of the Democratic Republic of the Congo (hereafter Congo) using Hsiao’s
approach of Granger-causality. The choice of the Congo is justified on the grounds that the
country experienced severe episodes of high inflation in the first half of the 1990s and the
resulting dollarization 1 may have rendered the relationships among price, the money supply,
and the exchange rate more complex, and may well provide new light to understand these
linkages. This question is important because the determination of the causal pattern among
inflation, money growth, and the exchange rate has important implications for policy-maker
choice of appropriate stabilization policies.
The remaining part of this article is organized as follows. Section 2 describes the
model and methodological considerations. Section 3 presents empirical results, mainly
stationarity and cointegration tests along with the results from the Hsiao’s variant of
Granger-causality tests. Section 4 discusses the results, followed by policy implications and
a conclusion.
2. Model, Data, and Methodological Considerations
The present analysis proceeds in four steps. The first involves the Phillips-Perron
(1988) tests of stationarity and Johansen (1988) test of cointegration, followed by a
multivariate cointegration test. The third step focuses on Hsiao’s version (1981) of the
1
Dollarization is the unofficial and partial replacement of the domestic currency by foreign ones
either as store of value, unit of account, or medium of exchange.
4
Granger non-causality method (Granger, 1969) to estimate causality for each equation of
the model. Causality results are then examined with conventional diagnostic tests.
Results from causality tests are highly sensitive to the order of lags in the
autoregressive process. An inadequate choice of the lag length would lead to inconsistent
model estimates, and any inferences would likely be misleading. Appropriate identification
of lag order for each variable requires some care. Hsiao’s approach responds to this concern
by combining the Granger concept of causality and Akaike’s final prediction error criterion,
and is specifically designed to avoid imposing false or spurious restrictions on the model.
For a detailed discussion of Hsiao's version of the Granger causality method, see Hsiao
(1981, 1982), Cheng and Lai (1997), and Bajo-Montavez (2002).
Hsiao’s variant of Granger-causality proceeds as follows. Suppose we want to test
Granger-causality for two stationary variables, Xt and Yt. Consider the models:
m
X t = α + ∑ β i X t -i + u t
(1)
i =1
m
n
i =1
j =1
X t = α + ∑ β i X t -i + ∑ γ j Yt - j + v t
(2)
where α is a constant term, β and γ are coefficient of exogenous variables, and ut and vt are
white noise error terms. The following steps are used to apply Hsiao’s procedure.
(i)
Consider Xt a univariate autoregressive process as in (1), and compute its final
prediction error criterion (FPE) with the order of lags i varying from 1 to m. Choose the lag
m that yields the smallest FPE and denote the corresponding FPE as FPEX (m, 0).
The corresponding FPE is
FPE (m) =
(T + m + 1) SSE
x
T − m −1
T
(3)
where T denotes the number of observations in the regression and SSE is the sum of squared
5
residuals. The determination of causality is performed as follows.
(ii)
Treat Xt as a controlled variable with m lags, add lags of Yt to (1) as in (2), and
compute the FPEs with the order of lags j varying from 1 to n. Choose the lag n that yields
the smallest FPE and denote the corresponding FPE as FPEX(m, n).
The corresponding FPE is given by
FPE (m*, n) =
(iii)
(T + m * + n + 1) SSE ( m*, n)
x
T − m * −n − 1
T
(4)
Compare FPEX (m, 0) with FPEX (m, n). If FPEX (m,0) > FPEX (m, n) then Yt is said
to Granger-cause Xt, whereas if FPEX (m, 0) < FPEX (m, n) then Xt is not Granger-caused by
Yt.
Reverse causality (whether Xt Granger-causes Yt) is determined by repeating steps
(i) to (iii) with Yt as the dependent variable.
In practice, the implicit assumption that Xt and Yt are stationary has to be confirmed
before proceeding with (1) and (2). If the series are non-stationary with unit roots, they
have to be transformed into stationary ones by means of a difference filter. If variables are
all non-stationary but some linear combination of the three series is stationary, it should be
checked whether there is any systematic co-movement among variables over the long run.
Such cointegration would imply that any standard Granger causal inferences will be invalid
and error correction models should be adopted. Unit root and cointegration techniques are
well documented (Engle and Granger, 1987; Johansen, 1988; Johansen and Juselius, 1990).
If variables are cointegrated it is useful to modify (1) and (2) by incorporating an
error-correction term as follows,
6
m
ΔX t = α + ∑ β i ΔX t -i + δz t −1 + u t
(1)
i =1
m
n
i =1
j =1
ΔX t = α + ∑ β i ΔX t -i + ∑ γ j ΔYt - j + δz t -1 + v t
(2)
where zt is the vector error-correction term (Engle and Granger, 1987). Notice that if Xt and
Yt are I(1) but are not cointegrated, no error correction mechanism binds the two variables
and three is no one period lagged error term in (1’) and (2’).
One purpose of the present paper is to analyze the trivariate causal relation of pricemoney supply-exchange rate. Testing for Granger-causality in the trivariate case requires
amending (1’) and (2’) by adding a third variable, W. The trivariate model examines the
causal relationship between Xt and Yt conditional on the presence of Wt,
m
p
i=1
k =1
ΔXt = α + ∑βi ΔXt-i + ∑θk ΔWt −k + δzt −1 + u t
m
n
p
i=1
j=1
k=1
ΔXt =α + ∑βi ΔXt -i + ∑γ j ΔYt - j + ∑θkΔWt−k +δzt−1 + vt
(3)
(4)
with the corresponding FPE given by
FPE (m*, n, p ) =
(T + m * + n * + p + 1) SSE (m*, n*, p )
x
T
T − m * −n * − p − 1
(5)
In the trivariate case, the relevant comparison is between FPEΔX(m, 0, p) and FPEΔX(m,
n, p) where (m, 0, p) and (m, n, p) are the combinations of lags leading to the smallest FPE
in each case. If FPEΔcpi(m, 0, n) > FPEΔcpi(m, n, p) money supply (y) Granger-causes CPI
(x) conditional on the presence of the exchange rate (w). The differenced time series have
no information about the long-run relationship between the trend components of the
original series since these have, by definition, been removed. "Standard" Granger-causality
7
tests may at best describe only short-run relations among price, money growth, and the
exchange rate.
Before moving into estimation results, a word on the data series is in order. The
analysis uses monthly data covering the period January 1990 to September 1996 in Zaire.
All data series are obtained from Beaugrand (1997). Price level (denote lnP) is natural
logarithm of the consumer price index (December 1981 = 1). Money expansion is the
natural logarithm of money outside of banks (denoted lnM). The exchange rate is the
natural logarithm of the parallel Zaire per unit of US dollars, period average (denoted lnZ ).
The use of the parallel exchange rate is justified on the grounds that the official exchange
rate did not affect any transaction in the economy at the margin and that the parallel market
represented the relevant marginal price for most transactions in the Congo over the whole
period under analysis (IMF, 1996). The nominal exchange rate is expressed as the price of
dollars in terms of domestic currency.
A visual inspection of the data in Figures 1, 2 and 3 shows a distinct tendency for the
three series to move together while showing no tendency for the series to revert to a
constant mean, which suggests that the series are nonstationary. The stationarity of the
series is examined by a more formal testing in the next section.
Figure 1. Price
Figure 2. Money
20
24
16
20
12
16
8
12
4
8
Figure 3. Exchange rates
28
24
20
16
12
0
8
4
1990
1991
1992
1993
lnP
1994
1995
4
1990
1991
1992
1993
lnM
1994
1995
1990
1991
1992
1993
1994
1995
lnZ
8
3. Estimation Results
3.1. Results from stationary and cointegration tests
This section explores causality between variables in a multivariable setting to avoid
possible spurious results due to the omission of relevant variables. Before proceeding with
cointegration tests, it must be established that the variables are integrated processes of the
same order. All the three variables are subjected to the Dickey-Fuller and Augmented
Dickey-Fuller tests (Dickey and Fuller; 1979, 1981). An intercept and a time trend is
included in the ADF regression and the null hypothesis of a single unit root cannot be
rejected at the 10% level for the three variables lnP, lnM, and LnZ. The three series each
become I(0) after first differencing as shown in Table 1 at the 5% level for prices and
money growth, and at the 10% level for the exchange rate variable.
Table 1. Unit Root Tests
Level
First Difference
DF
ADF
DF
ADF
LnP
-2.11
-2.54
-4.18*
-4.35*
LnM
-1.32
-1.94
-8.31*
-8.39*
LnZ
-2.13
-2.48
-3.01**
-3.22**
* 1%, **10%
The three variables are integrated of the same order and we proceed to the next step.
The order of the VAR model often plays a crucial role in empirical analysis, and the Akaike
Information Criterion (AIC) selects the number of lags required in the cointegration test. A
VAR model is first fit to the data and the AIC gives lag 5 as the appropriate lag structure.
To test for cointegration, we use the maximum eigenvalue and the trace tests suggested by
9
Johansen (1988) and Johansen and Juselius (1990) shown in Table 2. Our tests fail to reject
the null hypothesis that the variables are not cointegrated. Starting with a maximum of
three cointegrating vectors, both Trace and Eigenvalue tests indicate no cointegration at 5%
among price, money growth, and exchange rate. The lack of cointegration between
variables suggests that there exists no long-run relationship among variable under
consideration.
Table 2. Cointegration results
Hypothesized
No. of CE(s)
None
At most 1
At most 2
Eigenvalue
0.199856
0.126069
0.006525
Trace
Statistic
26.22698
10.17356
0.471317
0.05
Critical Value
29.79707
15.49471
3.841466
Prob.**
0.1220
0.2676
0.4924
*denotes rejection of the hypothesis at the 0.05 level.
**MacKinnon-Haug-Michelis (1999) p-values.
Trace test indicates no cointegration at the 0.05 level.
We used trend (unrestricted), constant, and five lag of each of the variables to test the
cointegrating relationship.
Since cointegration is ruled out, our results include only the differenced term, which
captures short-term adjustments. Therefore, the analysis proceeds with equations (1’), (3),
and (4) with the coefficient δ assumed equal to zero. Note that because of the presence of
unit root in the three variables, they are used in first difference form. However, for
simplicity, we still retain the terms money and exchange rate to represent the rate of change
of these two variables while the term “price” and “inflation” are used interchangeably
hereafter.
3. 2. Results from Hsiao’s version of Granger Causality
The lag lengths m, p and n were set at 12 because in applied econometrics,
maximum lags of 12 are generally used for monthly data. For causality tests, Table 3
reports the minimum FPEs of the three univariate autoregressions with both lnP and lnM at
10
lag 1 and LnZ at lag 4. Causality is established by comparing the minimum FPE from a
bivariate and a trivariate VAR. As indicated in Table 4 in the price equation, the exchange
rate (LnZ) is added first as the first manipulated variable in the first step, and lnM is added
to the previous equation in the next step. Since the FPE obtained in the first step is smaller
than the one obtained in the second step, we conclude that lnM Granger-causes inflation.
Subsequently, the lnM variable is first added to the equation, followed by LnZ, and the
result points to the conclusion that LnZ Granger-causes lnP in the short run.
For the money supply equation in Table 4 the exchange rate is entered into the
equation first followed by inflation, and the result infers inflation Granger-causes money
expansion. Next, inflation is entered into the equation followed by the exchange rate and it
is found that the exchange rate Granger-causes money growth in the short run.
Applied to the exchange rate equation, the above procedure leads to two findings.
One is that inflation does not Granger cause the exchange rate, and the other that monetary
expansion Granger causes the exchange rate.
Table 3. FPE of One-Dimensional AR Processes
Order of Lags
0
1
2
3
4
5
6
7
8
9
10
11
12
FPE of lnP
0.0705
0.0496*
0.0501
0.0516
0.0527
0.0527
0.0537
0.0554
0.0571
0.0586
0.0599
0.0609
0.0622
FPE of lnM
0.0383
0.0321*
0.0328
0.0338
0.0347
0.0358
0.0369
0.0379
0.0391
0.0400
0.0406
0.0416
0.0429
FPE of LnZ
0.0757
0.0555
0.0512
0.0475
0.0474*
0.0489
0.0503
0.0518
0.0533
0.0544
0.0562
0.0572
0.0589
(*) indicates lag order selected by FPE criterion at 5% level
11
Table 4. Results of the Hsiao’s version causality tests
Inflation
(lnP)
Money
growth
Controlled
variable
First
manipulated
variable
lnP (1)
LnZ (8)
lnP (1)
LnZ (8)
lnP (1)
lnM (0)
lnP (1)
lnM (0)
lnM (1)
LnZ (4)
lnM (1)
LnZ (4)
lnM (1)
lnP (1)
lnM (1)
lnP (1)
lnZ (4)
lnM (3)
lnZ (4)
lnM (3)
lnZ (4)
lnM (8)
lnZ (4)
lnM (8)
Second
manipulated
variable
rates (lnZ)
Fvalue
Causality
Inferences
5.63*
lnM causes
0.041
lnM (7)
8
0.032
2
0.034
LnZ (9)
2
0.027
lnP
8.96*
7
0.036
lnP (1)
(lnM)
Exchange
FPE
8
0.034
lnP
2.46**
7
0.036
LnZ (3)
7
0.037
0
0.041
2.13**
5
0.045
0
LnZ does not
cause lnM
5.36*
1
0.053
lnP (3)
lnP causes
lnM
0
0.040
lnP (6)
LnZ causes
lnP does not
cause LnZ
4.26*
lnM causes
LnZ
Note: (*) and (**) denote significance, respectively, at 5% and 10% level.
lnP, lnM and lnZ are in first difference form.
As shown by Hsiao (1982), for X to be a direct Granger-cause of Y, X should
Granger-cause Y both in bivariate and multivariate models. It follows that bivariate results
can be found by comparing minimum FPEs given by univariate estimations in Table 3 and
bivariate FPEs in Table 4. Bivariate Granger-causality is found between all pairs of
variables except from inflation (lnP) to money growth (lnM). This exception somehow
weakens the conclusion derived from trivariate Granger causality from inflation to money
growth. It should be note that results in Table 4 are further confirmed by the F-value
revealing that coefficients in each of the trivariate equations are significant at either the 5%
12
or 10% levels.
The causal relationships in Table 4 are summarized in Figure 1 showing that money
growth is at the crossroads of this complex relationship between the three variables. This
may also be interpreted as the fact that in the DRC money supply changes were mainly
used by authorities as automatic responses to price and exchange rate shocks often
originated from either fiscal or external stimuli.
Graph.1. Trivariate causality
lnP
lnM
LnZ
4. Explanation of the causality relationships
The present results suggest feedback causality between inflation and money growth,
along with unidirectional Granger causality from money growth to the exchange rate and
from the exchange rate to inflation. It is important to review the theories consistent with
these results. Money supply changes predicting inflation is consistent with the monetarist
theory that the root cause of inflation is an increase in money supply beyond its demand,
that "inflation is always and everywhere a monetary phenomenon" as Friedman (1968) puts
it. This finding corroborates that of Beaugrand (1997) in that at the beginning of the
inflationary process in the early 1990s, attempts by the Congo’s government to finance
13
through money creation resulted in an increase in the price level.
After the sharp decline in government revenue from 11% of GDP during 1986-89 to
about 5% during 1990-94 which accompanied the democratic unrests following 1990, the
government of the Congo (then Zaire) was unable to recapture its spending through taxes.
The government essentially resorted to printing money which resulted in hyperinflation
reaching a cumulative increase in prices from October 1990 to December 1995 of 6.3
billion percent (IMF, 1996).
What is also noteworthy is the present result that inflation and money growth
predict each other and are contemporaneously determined. The feedback between inflation
and money supply changes could be interpreted as a validation of the claim that the
Congolese monetary authority allowed the money supply to respond passively to demand
with inflation causing monetary growth. Similar to the German case as noted by Webb
(1985), the central bank did not aim for target growth rates of the money supply, but rather
aimed to guarantee that neither the corporate sector nor the government would need to
restrain its activities because of unavailable foreign exchange for imported intermediate
inputs.
During the hyperinflation episode, the Congolese Central Bank followed an
accommodative monetary policy that let the size of the money stock be determined by
government financing, and marginally by the private sector choice of the dollar and local
currency in its portfolio mix.
The pace of broad money growth reached record levels in early 1990s, from about
84% in 1986-89 to more than 2,000% during 1990-94. According the IMF (1996), in 1994
alone, broad money grew by 5,546%, which mainly reflected the deficit in government
operations (2,284%).
14
A critical element in the finding that money growth Granger-causes the exchange
rate is that with inflation becoming permanent and sustained, currency substitution
increases making the distinction between domestic currency depreciation and movement in
foreign exchange rate closely correlated. An increase in the money supply causes
depreciation, and following monetary theory, the exchange rate depends on relative money
supplies in the long run.
Our results confirm the finding by Canetti and Greene (1991) that in the Congo the
exchange rate Granger causes inflation. The prediction of inflation by the exchange rate
follows the fact that depreciation implies that the domestic price of imported and exported
final goods increases. Because these goods enter the Consumer Price Index (CPI), inflation
increases. The causality from this inference supports the structuralist view considering
inflation as tightly linked to exchange rate changes. Furthermore, the effect of exchange
alteration on inflation may reflect the fact that importers pass on exchange rate changes to
buyers rather than absorbing them in their profit margins, the pass-through of exchange rate
changes.
Under widespread currency substitution, the domestic price level and exchange rate
become equivalent, or the latter may become the most pertinent forward indicator for
expectations and measurement. In such a situation, depreciation would itself contribute to
further inflation if that is what people expect. Anticipating future depreciation, economic
agents increase the demand of foreign currencies in anticipation of expected depreciation
and other deterioration of economic fundamentals.
The present results furnish evidence that relationships among inflation, money
growth, and the exchange rate in the Congo over the 1990s have been complex as attested
15
by the successive failures of disinflation attempts by different transitional governments
during 1994-1996.
5. Policy Implication and Conclusions
The present study utilizes Hsiao’s approach to Granger-Causality to test the
relationships among money growth, the exchange rate, and inflation in the Congo during
the 1990s. Results reveal a feedback causality between inflation and money growth on one
side, and unidirectional Granger causality from money growth to the exchange rate and
from the exchange rate to inflation on the other.
Although the monetarist literature on inflation strongly holds the relative money
supply as the main cause of inflation, the three variables analyzed in the present paper
appear to be involved in a more complex web of interdependences. In particular, money
supply and the exchange rate appear to be mutually connected to inflation, both directly and
indirectly.
These results suggest that a reduction in the money supply will not simply reverse
the consequences of its previous increase unless the central bank takes accompanied
measures susceptible of bringing the exchange rate under control. The evidence supporting
the role of both money growth and exchange rate as leading indicators of inflation leads to
concern of how to turn this evidence into a useful tool for monetary assessment.
A key condition for a successful stabilization policy in an economy with currency
substitution is the commitment of the government to balance its budget, and rigorously
manage domestic and foreign currency cash flows of the public sector to minimize
unnecessary volatility in the foreign exchange market. Stabilization policy should tackle
high inflation by not only relying on the money supply but also focusing on variables that
16
impact the circulation of foreign currencies. Furthermore, variations in expected
depreciation change the share of foreign currency related to domestic currency for current
transactions implying instability of money demand and difficulties for implementing the
money targeting rule. In addition, the extensive currency substitution brings about
exchange rate instability, alters the demand for domestic money, and makes the
implementation of monetary policy futile.
The conclusion is that the over-riding goal of disinflation needs to be accomplished
initially by exchange rate stabilization, followed by a direct inflation targeting. Such a
sequence of targets is appropriate for developing economies: exchange rate targeting to
bring inflation down to single-digit levels, then money growth targeting since control
instruments such as open market operations is not available, and finally inflation targeting
as data and forecasting tools become sophisticated.
17
References
1. Bajo, R., Montavez, G., 2002. Was there monetary autonomy in Europe on the eve of EMU? The
German dominance hypothesis re-examined. Journal of Applied Economics 2 November 2002, 185207.
2. Beaugrand, P. 1997. Zaire's hyperinflation, 1990-96. Working Paper WP/97/50,
Washington, DC: International Monetary Fund.
3. Burdekin, R.C.K., Burketi, P. 1996. Hyperinflation, the exchange rate and endogenous money:
post-World War I Germany revisited. Journal of International Money and Finance 15, 599-621.
4. Cagan, P. 1956. The monetary dynamics of hyperinflation. Studies in the Quantity Theory of
Money. University of Chicago Press; Chicago, 25–117.
5. Canetti, E., Greene J. 1991. Monetary Growth and Exchange Rate Depreciation as Causes of
Inflation in African Countries: An Empirical Analysis. Center For Economic Research on Africa
Research Monograph Series. School of Business-Montclair State University, New Jersey.
6. Cheng, B. S., Lai, T.W. 1997. An investigation of cointegration and causality between energy
consumption and economic activity in Taiwan. Energy Economics 19, 435-444.
7. Dickey, D. A. and Fuller, W. A. 1979. Distribution of the estimator for autoregressive time series
with a unit root. Journal of the American Statistical Association 74, 427-431.
8. Dickey, D. A., Fuller, W. A. 1981. Likelihood Ratio Statistics for Autoregressive Time Series
with a Unit Root, Econometrica, 49, 1057–1072.
9. Engle, R.F., Granger, C.W.J. 1987. Co-integration and error correction: representation,
estimation, and testing. Econometrica 55, 251-276.
10. Friedman, M. 1968. Dollars and Deficits: Inflation, Monetary Policy and the Balance of
Payments. Prentice-Hall, Englewood Cliffs.
11. Granger, C.W.J. 1969. Investigating causal relations by econometric models and cross spectral
methods. Econometrica 37, 428-438.
12. Hsiao, C. 1981. Autoregressive modeling and money income causality detection. Journal of
Monetary Economics 7, pp. 85–106.
13. Hsiao, C. 1982. Autoregressive modelling and causal ordering of economic variables. Journal of
Economic Dynamics and Control 4, 243-259.
14. IMF. 1996. Zaire-Background Information and Statistical data. IMF Staff Country report No.
96/28. Washington, D.C.
15. Johansen, S. 1988. Statistical analysis of cointegration vectors. Journal of Economic Dynamics
and Control 12, 231-254.
16. Johansen, S., Juselius, K. 1990. Maximum likelihood estimation and inferences on cointegration
with application to the demand for money. Oxford Bulletin of Economics and Statistics 52, 169-210.
18
17. Minshki, F. S. 2004. The Economics of Money, Banking, and financial markets, 7th ed.
International Ed. Boston.
18. Phillips P., Perron, P. 1988. Testing for a unit roots time series regression. Biometrika 75, 335346.
19. Sargent J.J., Wallace, N. 1973. Rational Expectations and the dynamics of hyper-inflation.
International Economic Review 14, 328-350.
20. Vernengo M. 2005. Money and Inflation. Handbook of Alternative Monetary Economics. to be
edited by Philip Arestis and Malcolm Sawyer. http://www.econ.utah.edu/~vernengo/, retrieved on
April 15th, 2005.
21. Webb S. B. 1985. Government Debt and Inflationary Expectations as Determinants of the
Money Supply in Germany 1919-23. Journal of Money, Credit and Banking 17, 479-492.
19