Money neutrality: causality effect
Download
Report
Transcript Money neutrality: causality effect
Empirical study of causality
between Real GDP and
Monetary variables.
Presented by :
Hanane Ayad
overview
• Since the early 60’s , the literature concerning
the interaction between monetary variables and
real income, especially in U.S, have taken an
new direction.
• Many of those models have been constructed to
take into consideration some sophisticated
channels through which money can influence
Real Income.
• However, the empirical results gathered so far
still do not lead to a clear path as to whether
monetary variables do have a direct or whatever
kind of effect on the real GDP.
Chronology of literature
• The earliest study dealing with the
causal relationship between money and GDP
was performed by Freidman and Schwartz
(1963).
• Nine Years later, Sims(1972) came to argue that
it is quite inappropriate to distinguish between
cause and effect based only on correlation
pattern.
• Eight years later (1980),Sims suggested a vector
auto regression process that takes into
consideration the eventual effect of additional
controlling macroeconomic variables.
Chronology of literature
• The causality between money and GDP that Sims found
in (1972) disappered as soon as he included a short
term interest rate as a controlling variable.
• Others later found a strong evidence of causality when
they use log level data and no significant effect of money
on income when they adopt first difference data
• Stock and Watson (1989) adopted a first differenced log
data, they found that M1 growth affects GDP growth
when they included a Trend function, this monetary
effect vanished once they increased the sample size,
and added interest rate as an explanatory variable
• Dufour and Tessier (1997)came up with a more
ingenious and logically coherent specification VARMAEchelon.
Objectives of the empirical study
• Establish a causal link between monetary
variables(m1 interest rate) and real GDP .
• Investigate the “nature” of the dynamic of
this relationship.
• Detect a long run relationship and the
speed of convergence into equilibrium (if
there any)
methodology
• The most commonly used test for causality is the
standard granger causality test.
• The best method that can be used to test the
causality of cointegrated variables is the ECM
procedure.
• The main advantage of this methodology: it does
not only detect the causality effect, but also
gives an idea about the long run relationship
between the variables.
• If the variables are not cointegrated, the ECM is
inappropriate, the best alternative in this case is
the use of a VAR in difference process.
• In this paper, I intended to test the causality between
between some Monetary variables: m1, interest rate,
CPI, and the real GDP using the ECM procedure.
• Before using the ECM, we need to make sure that all the
four time series have the same unit roots.
• If it is the case, we then test for cointegration among the
four variables, if they are cointegrated, we then can
proceed to the second step of ECM, and run the long
run regression: D(RGDP)=a+b1D(RGDPt-1)+b2D(M1t1+b3D(INTRTt-1)+b4D(CPIt-1)+b5Ut-1+et
• In the case of no cointegration, we, use the VAR
differenced, and test for the significance of the lagged
independent variables.
• If the variables do no have the same unit roots, we just
use the standard Granger causality test
Econometrics analysis
Simulations reveals that the four variables
are all I(1)for example of the real M1:
• ADF (M1 LEVEL)
• ADF Test Statistic-0.837416
•
•
1% Critical Value*-3.5889
5% Critical Value-2.9303
10% Critical Value-2.6030
• ADF M1 (FIRST DIFFERENCE)
• ADF Test Statistic-3.950100
•
•
1% Critical Value*-3.5930
5% Critical Value-2.9320
10% Critical Value-2.6039
Econometrics analysis
• ADF RES2 (LEVEL)
• ADF Test Statistic-2.287506
•
•
1% Critical Value*-3.6019
5% Critical Value-2.9358
10% Critical Value-2.6059
• The residuals have a unit root of more than o,
hence, the variables are not cointegrated, in this
case, the use of ECM procedure is
inappropriate.
• The best method than can be adopted in the
absence of cointegration is the VAR in difference
cointegration
• We say that Xt and Yt are cointegrated if there is a long run
relationship between Xt and Yt.
• If we have two non-stationary time series that have the same unit
roots, let say
Xt = T(t) + e(x)t, and
•
Yt= z(t) +e(y)t , where T(t) , Z(t) are trend terms,
•
e(x)t and e(y)t are white noise.
• If these two series can be written as a linear combination so that the
trend terms cancels out, then we can say that Xt and Yt are
cointegrated.
• This can be done through testing the unit root for the residuals
resulting from regressing Yt on Xt (Xt and Yt ) has the same unit
root. The residual should have unit root less than that of the
variable.
• Less say Xt →I(b) and Yt → I(b)
• If Ut → I(d) where d is less than b. Them we can say that Xt and Yt
are cointegrated.
Unit Root
• Yt = a + bt + Ut where Ut is the white noise and t
is the trend factor.
• Yt-1 = a + b(t-1) + Ut-1
• ∆Y = a + Ut - Ut-1.
• If we manage to cancel out the trend factor by differentiating only
one time this means that Yt has one unit root.
• The DF test is done as follows:
• H0: THE VARIABLES HAS 1 UNIT ROOT
Ha: b < 0.
• ADF test done the same way, but we just add other lag difference
terms of the dependent variable. So that we can control for higher
order correlation.
• If ADF is less than ADF critical value we don’t reject the null.
Var in Difference
• ∆Yt = C1 + a11∆Yt-1 + a12∆Xt-1 + U1t
• ∆Xt = C2 + a21∆Yt-1 + a22∆Xt-1 + U2t
We notice that the dependent variables are
different but the set of the independent
variables are the same.
Results of the VAR
• After analyzing the VAR in difference results, it
seems plausible( Based on t-statistics) that:
• Interest rate has a direct effect on M1
• M1 has a direct effect on CPI
• CPI has a direct effect on GDP
• Overall
• Even though M1 does not have a direct effect on
RGDP, it affects RGDP indirectly through CPI
conclusion
• This paper dealt first with detecting if there is a
long relationship between variables: unit rot
testing was used to achieve this objective.
Econometrics results reveals that monetary
variables have an impact and a causal
relationship on real GDP, but there were no
evidence that they play any statistical significant
role in the determination of real GDP in the long
run.