econd quarter of 2006, probably due to an unexpected

plunge in energy prices. The last

dummy controls for the economic impacts of the terrorists
attack of September 11, 2001.
Moreover, it is worth noting that the graph of
the real money shows two breaks in
trend. A constant increase in the variable is observed
from the beginning of the sample until
1987. A period of approximately zero growth follows and
lasts until 1995, when it starts
increasing as fast as in the first period. Thus, the
slope of the two growing trends seems to be
roughly the same. Even though the introduction of two
breaks has been attempted, it turned
out more satisfactory to allow for trends in the levels.
This should have the effect of
averaging out the aforementioned breaks. Moreover, since
it is not possible to know a priori
whether these trends cancel out in the cointegrating
relations, the chosen specification allows
these to be trend-stationary and have non-zero
intercepts.

4.3. Lag Length Selection and Residual Analysis

Table 3 shows that the Schwartz criterion (SC)
suggests 1 and the Hannan-Quinn
(HQ) criterion suggests 2. However, when imposing
1 the other misspecifications
tests become much worse, implying that the SC might have
penalized too much. The LM
tests in the last two columns show the left-over residual
autocorrelation in each VAR(
model and seem to accept the absence of autocorrelation
for the VAR with 2 lags.

[Table 3 about
here]

Table 4 presents the results of the multivariate

residual analysis. In particular the null
hypothesis of no autocorrelation is rejected at both the
first and the second lag. Given the
relatively small sample, it is not advisable to rely on
the asymptotic properties of the
estimator, thus the normality assumption turns out to be
relevant and it is safely accepted.
Lastly the null hypothesis of no ARCH effects is accepted
at both lags.

[Table 4 about
here]

Table 5 shows the results of the univariate

residual analysis. The output of the ARCH
and Normality tests reflect the good results of the
multivariate analysis. The skewness and
kurtosis statistics are close to the normal distribution
values, suggesting that by inserting the
three blip dummies the biggest outliers should have been
controlled for.

[Table 5 about
here]

Overall, both the multivariate and univariate tests

suggest that the residuals are well
behaved and therefore that the model is well specified.

4.4. Rank Determination

The cointegration rank is determined according to
Johansen (1996) LR Trace test.
When the sample is small, the asymptotic distributions
are generally poor approximations to
the true distributions. To secure a correct test size one
can apply the small sample Bartlett

8
corrections developed in Johansen (2002). The asymptotic
distribution of the rank test
statistic differs depending on the deterministic
components in the model and on almost any
type of dummy variable3. Therefore, the safest procedure
is to simulate the new critical
values. Table 6 presents the Trace test results with the
Bartlett corrections and the simulated
critical values.

[Table 6 about here]

The choice of the cointegration rank is not

clearly defined, in fact the Trace test
Bartlett corrected suggests that 3 is accepted only
at 10% significance level. Looking at
the significance of the coefficients of the third
cointegration vector in Table 7 it seems that
information regarding the real GDP and the real commodity
prices would be neglected by
choosing 2.

[Table 7 about here]

A graphical inspection of the cointegrating

relations in Figure 4 reveals some
symptom of I(2)ness. The first two cointegration
relationships look fairly stationary, but the
third one presents some indication of cyclical swings.
However, it should be observed that the
lower panel corrected for short-run effects of
each graph is similar to the upper panel

, confirming that the I(2) problem could have been

limited (see appendix for a formal
I(2) rank test).

[Figure 4 about here]

 Moreover, an examination of the characteristic
roots shows that the largest
unrestricted root for 2 is 0.90 and for 3 is 0.78
(pretty far from the unit circle). This
seems to confirm the presence of three common stochastic
trends. Figure 5 shows the roots of
the companion matrix for 3.

[Figure 5 about here]

4.5. Recursive Tests

The graphs of the recursively calculated
fluctuation tests in Figure 6 show that the X-
form of and are in the rejection region at the
beginning of 1995, when the recursion
starts. The test statistics remain at a fairly high level
until approximately 1998. The recursive
graphs of the suggest that the parameter of the second
cointegration relation are
considerably constant over the sample period. The overall
test in the lower right-hand side

3An exception to this are the centered seasonal dummies,

which, by construction, sum to zero over time, and hence
do not change the asymptotic distribution of the rank
tests.

9
panel picks up the non-constancy at the beginning of the
recursions. However, it should be
noted that the R-form looks stable in all , meaning that
the instability is only in the short-run
coefficients. In general, the eigenvalue fluctuation
tests provide a fair picture.

[Figure 6 about
here]

The Max test of Constancy is always lower than

one and shows a slightly higher
volatility after 2003. Figure 7 confirms what is
suggested by the eigenvalue fluctuation tests,
namely that the changes in the eigenvalues are due to
changes in .

[Table 7 about
here]

4.6. Long-Run Exclusion

The matrix gives tentative evidence of long-run
exclusion for the variables in the
system. If a variable is excludable, the coefficients in
the columns must be insignificant.
From the PI matrix there are no clear signs that any of
the variables can be excluded from the
cointegration relations.
A formal LR-test for variable exclusion has been
performed. Based on the results in
Table 8, it is not possible to exclude any variable at
10% significance level for 3.
However, the real commodity prices seem to be a border-
line case.

[Table 8 about
here]
 4.7. Weak Exogeneity and Pure Adjustment
The matrix gives preliminary evidence of weak
exogeneity. If a variable is weakly
exogenous, the coefficients in the rows must be
insignificant; in other words, it represents a
pushing force. The only variable that seems not to react
to any other variables is the real
money.
The formal LR-test for weak exogeneity in Table 9
shows that both real GDP and real
money are weakly exogenous at 10% significance level,
even though the joint exogeneity is
rejected. It has been decided to carry out the analysis
including them in the system in order to
analyze their impact in the matrix.

[Table 9 about
here]

Another LR-test is carried out to test for unit

vector in the matrix. In other words, it
tests whether the cumulated disturbances from the ith
variable do not enter the common trends
(pure adjustment hypothesis). At 10% significance level
Table 10 shows that the null
hypothesis cannot be rejected for the long term interest
rate and the real commodity prices.
Thus, these can be considered pulling forces.

10
[Table 10 about here]

4.8. Identification and Interpretation of the Results

The identifications process starts from the
long-run relationships. It is carried out by
imposing 1 non-testable4 just-identifying
restrictions on each beta vector of the
unrestricted reduced VAR model with rank 3,
with no changes to the likelihood function
with respect to the under identified model. Table 11
presents the estimation results that seem
to suggest that the first beta vector is describing a
money demand relationship with an
unconventional sign on the inflation variable. The second
cointegrating relationship
illustrates the expected correlation between real output,
real money, real commodity prices
and a trend. Finally, the third stationary vector is
expressing a relationship between inflation
and the interest rate spread.

[Table 11 about here]

The over-identified structure is modeled

imposing restrictions accordingly to the t-
value of the vectors' coefficients of the just-
identified structure. This is accepted with a
fairly large p-value of 0.642 (meaning that the
stationarity of the long-run relations cannot be
jointly rejected).
Table 12 permits the detection of the pulling
forces for each cointegration relation.
The money demand relationship is corrected by changes in
inflation, whereas the real
commodity prices are the only equilibrium correcting
force for the second cointegrating
relation. Lastly, deviations from the inflation
expectations relationship are corrected by
inflation itself.
[Table 12 about here]

The short-run identification has been carried out

by removing all the non-significant
variables from each equation of the cointegrated VAR5.
The parsimonious structure cannot be
rejected with a p-value of 0.257 and is broadly
consistent with the classification into pushing
and pulling forces6.
The equation for the interesting variable is
reported in Table 13.

[Table 13 about here]

From the inspection of the matrix in Table 14, it

is evident that some residuals are
highly correlated (positively between real GDP and real
money, negatively between inflation

4 Since we have imposed 1 restrictions, there are

zero degrees of freedom, thus restrictions cannot be
tested.
5 Centered seasonal dummies have been left in the
equations for each variable, even if insignificant.
6 The equilibrium correcting role of the real commodity
prices for the second cointegrating relationship
confirmed.

11
and real GDP, negatively between inflation and real money
and positively between the two
interest rates) and this would suggest a simultaneous
specification of the model. Nonetheless,
such analysis has not been pursued because of the
problems in determining the direction of
the causation between the mentioned variables.

[Table 14 about here]

Overall, the results confirm the existence of the

hypothesized long-run equilibrium
relationship between real money and real commodity
prices, with an effect from real output.
Moreover, there is evidence that the real commodity
prices are the only equilibrium
correcting variable.
It can be concluded that in order to restore the
long-run equilibrium when there is a
real excess (lack) of liquidity, the real commodity
prices need to increase (decrease). Such
increase (decrease) can be achieved through an increase
(decrease) in the commodity prices
stronger (smaller) than the increase (decrease) in the
consumer goods' prices, generating a
larger spread between the two as in Figure 8.

[Figure 8 about here]

The Moving Average (MA) representation of the

data for 3 corresponds to
3 common trends. Since 1 2 just-
identifying restrictions are imposed on
each vector, estimates are not unique and the likelihood
function is unchanged. However, the
space spanned by and is uniquely determined, so that
the estimated long-run impact
matrix C is unique7. The normalization has been placed on
the variables with the highest
residual standard errors, namely real GDP, real money and
real commodity prices.
 The cumulated empirical shocks to the real GDP
have had significant and high
negative effects on the real commodity prices. The
inverse is true, but the impact is much
more moderate. The cumulated shocks to real money have
had positive effects on inflation, as
well as the cumulated shocks to short-term interest rate
on long-term term one and vice versa.
Moreover, the cumulated shocks to long term interest rate
have had positive effects on
inflation. Likewise, the cumulated shocks to inflation
have had similar positive effects on
both the interest rates. Finally, the cumulated shocks to
all the variables have had a positive
effect on themselves. It seems that the long-term
interest rate is purely adjusting (consistently
to the previous findings).
In order to impose the over-identifying
restrictions on , joint weak exogeneity for
real money and real GDP is tested keeping fixed the
restrictions on the vectors. Since the
hypothesis turns out to be rejected at 10% significance
level, a zero row in is imposed only
for the real money, for which the null hypothesis cannot
be rejected with a p-value of 0.505.
The C matrix of the over-identified MA representation is
broadly consistent with the results
obtained for the just-identified structure.
As a robustness check, the CRB has been replaced
with the Conference Board's
Sensitive Materials Index (SENSI). This comprises raw
materials and metals but excludes
food and energy. The results remain broadly the same as
before.

7As when imposing just-identifying restrictions on the

cointegrating relationships, the PI matrix was uniquely
determined, albeit and vectors were not.

12
5. Conclusions
Commodity prices are currently seen as one or the
main source of current inflationary
pressures and there seems to exist, as discussed by
Frenkel (2006), a linkage between
increases in commodity prices in commodity exporting
countries and monetary policy
changes in advanced industrial economies. This, as
suggested, would defy the common
knowledge that changes in commodity prices are solely
impacted by developments occurring
in the commodity markets.
The aim of the paper is to identify the nexus
between the excess of liquidity in the US
and commodity prices over the period 1983-2006 within a
cointegrated VAR framework and
in particular at testing whether the commodity prices
react more powerfully than the
consumer goods' prices to changes in real money balances.
The results provide empirical evidence on the
magnitude of the reaction of
commodity and consumer goods' prices to an increase in
real money. In particular, a long-run
equilibrium relationship between real money and real
commodity prices has been found, with
an effect from real output. The two variables of interest
show positive and significant
correlation.
Moreover, real money seems to be a weakly
exogenous variable and therefore is a
pushing force (away from the equilibrium), whereas real
commodity prices are the only
equilibrium correcting variable to such cointegration
relationship, or pulling force.
Therefore, in order to restore the long-run
equilibrium when there is a real excess
(lack) of liquidity, the real commodity prices need to
increase (decrease). Such increase
(decrease) can be achieved through an increase (decrease)
in the commodity prices stronger
(smaller) than the increase (decrease) in the consumer
goods' prices, generating a larger
spread between the two.
The results have important policy implications.
More specifically, if the magnitude of
the reaction is due to the fact that consumer goods'
prices are slower to react, then their long-
run value can be predicted with the help of the commodity
prices. In other words, the extent
of the rise in the commodity prices acts to predict
subsequent changes in the price of the other
goods, namely the consumer goods, whose price is
initially unchanged.
Moreover, the results also support the idea that
monetary policy cannot only focus on
the core CPI and ignore developments in the commodity
market. In fact, if commodity prices
are very high it might be the case that monetary policy
is loose; therefore they should be
taken into account as a useful monetary indicator. This
conclusion is particularly relevant to
those countries that are adopting an inflation targeting
regime which target is the CPI.

Acknowledgments
I wish to thank Katarina Juselius, Morten Tabor
and Paolo Paruolo for their kind
support in carrying out the econometric analysis. A
special thank to Paolo Paruolo for
providing me with important comments. Note that all
mistakes and opinions expressed in the
paper belong to the author only.

References
Awokuse T. O., Yang J. (2003), "The information Role of
Commodity Prices in Formulating
Monetary Policy: A Re-examination", Economics Letters,
79, 219-224
Bessler, D.A. (1984), "Relative Prices and Money: a
Vector Autoregression on Brazilian Data",
American Journal of Agricultural Economics, 66, 25-30

13
Barsky R. B., Kilian L. (2002), "Do We Really Know that
Oil Caused the Great Stagflation?
A monetary alternative", in Bernanke B., Rogoff K.
(eds.), NBER Macroeconomics Annual
2001, May 2002, 137-183
Bernanke B. S., Gertler M., Watson M., Sims C.A.,
Friedman B.M. (1997), "Symmetric
Monetary Policy and the Effects of Oil Price Shocks",
Brookings Papers on Economic
Activity 1, 91-174
Bahr R., Hamori S. (2008), "Information Content of
Commodity Futures Prices for Monetary
Policy", Economic Modelling, 25, 274-283
Boughton J.M., Branson W.H. (1990), "The use of Commodity
Prices to Forecast Inflation",
Staff Studies for the World Economic Outlook, IMF, 1-18
Boughton J.M., Bronson W.H. (1991), "Commodity Prices as
a Leading Indicator of
Inflation", in Lahiri K., Moore G.H. (eds.), Leading
Economic Indicators � New
Approaches and Forecasting Records, Cambridge University
Press, 305-338
Browne F., Cronin D. (2007), "Commodity Prices, Money and
Inflation", ECB Working
Paper 738, Frankfurt
Cody, B.J., Mills, L.O. (1991), "The role of commodity
prices in formulating monetary policy",
Review Economics and Statistics, 73, 358-65
Dornbusch R. (1976), "Expectations and Exchange Rate
Dynamics", Journal of Political
Economy, 84, 1161-1176
Frankel J. A. (1986), "Expectations and Commodity Price
Dynamics: The Overshooting
Model", American Journal of Agriculture, 68(2), 344-348
Frankel J. A. (2006), "The Effect of Monetary Policy on
Real Commodity Prices", NBER
Working Paper 12713
Fuhrer J, Moore G. (1992), "Monetary policy rules and the
indicator properties of asset
prices", Journal of Monetary Economics, 29(2), 303-336
Garner, A.C. (1989), "Commodity prices: policy target or
information variable?" Journal of
Money, Credit, and Banking, 21, 508-514
Hamori, S., (2007), "The information role of commodity
prices in formulating monetary
policy: some evidence from Japan" Economics Bulletin,
5(13), 1-7
Hua, P. (1998), "On primary commodity prices: the impact
of macroeconomic/monetary
shocks", Journal of Policy Modeling, 20, 767-790
Johansen S. (1996), "Likelihood Based Inference in
Cointegrated Vector Autoregressive
Models, 2.ed. Oxford, Oxford University Press
Johansen S. (2002), "A small sample correction of the
test for cointegrating rank of the vector
autoregressive model", Econometrica, 70, 1929-1961
Juselius K. (2006), "The Cointegrated VAR Model:
Econometric Methodology and
Macroeconomic Applications, Oxford University Press
Olivera, J.H.G. (1970), "On passive money", Journal of
Political Economy, 78, 805-814
Marquis, M.H., Cunningham, S. R., (1990), "Is there a
role of commodity prices in the design
of monetary policy? Some empirical evidence", Southern
Economic Journal, 57, 394-412