Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Economies
César R. Sobrino
School of Business & Entrepreneurship
University of Turabo
PO BOX 3030
Gurabo, PR 00778
Email: sobrinoc1@suagm.edu
ABSTRACT
In this paper, I examine the comovements between the US real gross domestic product (RGDP)
and the Puerto Rican RGDP. To accomplish this, I employ a cointegration test and the Vahid and
Engle (1993)’s common feature test. Using data from 1947 to 2008, the findings indicate there
are common movements between Puerto Rico and the US in the long and short run. Likewise, the
evidence shows that Puerto Rico has been more sensible to US permanent shocks than US
transitory shocks.
En este documento, analizo los movimientos comunes entre el PBI real de Estados Unidos y el
PBI real de Puerto Rico. En este cometido, utilizo un test de cointegración, así como, el test de
Vahid&Engle(1993) sobre movimientos cíclicos comunes. De 1947 al 2008, los resultados
indican que Estados Unidos y Puerto Rico presentan una tendencia y fluctuaciones cíclicas
comunes. Además, se encuentra que Puerto Rico ha sido más sensible a los choques permanentes
provenientes de Estados Unidos que a los choques transitorios que provinieron del mismo país.
Puerto Rico is a self-governing non-incorporated territory of the United States and its
economy works under the US legal frame: the American dollar is the legal tender on the
island; there is a common customs union; and there are no factor mobility restrictions.
In this paper, I analyze the comovements between the US and Puerto Rico following the
techniques employed by Herrera (2003), who tested the comovements between US and
Mexico after the adoption of the North American Free Trade Agreement (NAFTA). The
stages of this scheme are: (1) identify a long-term relationship between the US real gross
domestic product (RGDP; QUS) and the Puerto Rican RGDP (QPR) using a cointegration
test; and (2), conditional to cointegration, find a short-run relationship between both
RGDPs using Vahid and Engle (1993)’s test. This approach is different from that
employed by Alameda (2000), who examines the Puerto Rican business cycles.
Using annual data (1947–2008), the evidence indicates that the two economies share a
in US permanent shocks. This finding is consistent with the fact that, in 1950–2007, the
RGDP per capita of Puerto Rico grew by almost eight times, while the US RGDP per
capita grew by almost three times. Likewise, both economies share common cycles, and
However, this last outcome needs more revision because Puerto Rico’s RGDP is just
accounted annually for and with some delay, which would mislead short-run outcomes.
2
This paper is organized as follows: Section 2 presents a brief review of the econometric
techniques used and some empirical findings. Section 3 presents the cointegration and
common cycle tests and data; section 4 reports the outcomes of the tests; and section 5
concludes.
Cointegration (Engle and Granger, 1987; Stock and Watson, 1988; Johansen, 1991)
examines the existence of common stochastic trends among non-stationary time series. In
a bivariate case, two time series of order p are cointegrated when there is at least one
linear combination of them of the order p-1. Likewise, conditional to cointegration, the
common feature test (Vahid and Engle, 1993) examines common cycles among times
series. In this case, the matrix of cointegrating parameters is used to determine the
Engle and Kozicki (1993) analyze common features among stationary time series.
In this approach, the common feature test is not conditional to cointegration and the
(1997) extend Engle and Kozicki’s approach in analyzing common features when they
are not synchronized. These movements are called “codependent cycles.” These two
For Brazil and Argentina, which are major trading partners, Engle and Issler
(1993), Arnaudo and Jacobo (1997), and Mejía (1999, 2000) have found that the
countries share strong comovements. For Mexico and the US, which are also major
trading partners, Mejía, (1999, 2000) finds common cycles between both countries from
1950 through 1995. However, using a different sample, Torres (2000) only shows strong
3
evidence of common cycles from 1948 to 1979. Finally, Herrera (2004) found that, after
NAFTA adoption, both countries have shared a common trend and common cycles.
Consider n variables of integration order I(1), which can be represented in the next VAR
,…..(1)
coefficients with elements Ajk (i); and ε is the (n*1) matrix of disturbance terms
After some operations of adding and subtraction in (1), the following VECM of order p it
is obtained:
,…(2)
Johansen (1991)’s cointegration test focuses on the rank (r) of π. If 0 < r < n, there are r
linear combinations of the series, which are I(0). In other words, r is the number of
4
observations and the lambdas are the eigenvalues of π. In the first test, the null hypothesis
is the number of cointegrating vectors ≤ r, and in the second test, the null hypothesis is
Once a long-term relationship is found among the series, the cointegrating vector is used
to find Vahid and Engle (1993)’s statistics for the common cycles test. In this approach,
the canonical correlations between {Δxt} and { xt-1, Δxt-1,…,Δxt-p} are estimated. The null
hypothesis that the dimension of the cofeature space is at least s (the quantity of the
smallest squared canonical correlations) is tested with this statistic. The statistic
squared canonical correlations. This statistic has a Chi-square distribution with s(np+r)-
5
3.1. Data and Modeling
For the period 1947–2008 (yearly data), the US RGDP, hereafter QUS, is provided by the
Fed of St. Louis web page, while the Puerto Rico Planning Board provides the RGDP for
Puerto Rico,1 hereafter QPR. Thus, n=2. In Figure 1, both series are shown in logs. At first
glance, over the 60 year period, it looks like they share a common trend and short-run
comovements (mostly in the 1970s); however, after the 2001 US recession, it seems that
In order to determine the integration order of the series, Augmented Dickey-Fuller test’s
outcomes are used; the results are shown in Table 1. They indicate that neither process is
stationary because the null hypothesis cannot be rejected. In the first differences, the null
hypothesis, which is that the process has a unit root, is rejected at 5% of significance for
To test the optimal lag order of the VAR, the Akaike Information Criterion (AIC) is
employed; the results are reported in Table 2. They indicate that the optimal lag order at
levels is 3, which means that the optimal lag order is 2 in the first differences. Thus, when
, ……(3)
1
I thank Dr. Ervin Martinez, Director of the Puerto Rico Planning Board, for making the Puerto Rican data
available.
6
Where L is a matrix of lag operators and . The number of
In this form, the matrix of deterministic variables allows changes regarding the inclusion
of unrestricted and restricted constants, as well as unrestricted and restricted linear trends.
other words, the coefficients of the restriction are included in the cointegrating vector.
all s>0.
4. Outcomes
Including a drift and a restricted linear trend, the outcomes of λtrace and λmax tests are
shown in Table 3 and Table 4, respectively. In the first test, the null hypothesis that r≤1
cannot be rejected at 5% significance, while the null hypothesis that r=0 is rejected at 5%
significance. In the second test, the null hypothesis that r=0 is rejected at 10% and the
null hypothesis that r=1 cannot be rejected at 5% significance. All of these results
indicate that there is at most one cointegrating relationship between QUS and QPR over the
When it is not included as a restricted trend, in both lambda tests, the null hypothesis that
r=0 cannot be rejected. The restricted linear trend is consistent with data showing that the
two RGDPs veer apart after the 2001 US recession. Likewise, QPR does not show any
breaks; for that reason, no dummy has been included. However, to be accurate, (1) is run
7
including a time dummy (1991–2009=1; otherwise, zero). This dummy is not significant
at 5%.2
β is (1, -0.22, -0.02), normalized with respect to the coefficient of QUS. The second one is
the cointegrating coefficient of QPR, and the last one is the coefficient of the linear trend.
The restricted α is (-0.56,-0.36), which means that when both series (or one of them)
deviate from their long run equilibrium, the adjustment of QUS is faster than the
adjustment of QPR.
On the other hand, in order to get the canonical correlations, the cointegrating restriction
(β) is used to test common cycles. Reported in Table 5, this test indicates that there is a
common feature between QUS and QPR. The null hypothesis that s>0 cannot be rejected at
Both the cointegrating and cofeature vectors are shown in Table 6. Puerto Rico is more
outcome is consistent with the fact that, from 1950 to 2007, the RGDP per capita of
Puerto Rico has grown by almost eight times while the US RGDP per capita has grown
by three times.3
2
For Puerto Rico, Alameda (2000) divides his analysis into two periods: 1967:01–1991:03 and 1991:03–
1998:06. Using a Chow test and the Economic Activity Index as a proxy variable of real GDP, He finds
that there is a break in 1991:03.
3
The Puerto Rico RGDP per capita was $3,327.01 in 1950, and $26,212.98 in 2007; the US RGDP per
capita was 12,989.94 in 1950, and 42,886.92 in 2007 (Penn Tables 6.3).
8
In the short run, QPR positively responds at almost 1.6% to a 1% change in US transitory
shocks.4 Alameda (2000) finds that, since the late 1980s, US transitory shocks have been
diminishing in terms of their importance for Puerto Rican RGDP fluctuations. However,
this last result needs more revision because the Puerto Rico RGDP is just determined
annually and with some delay,5 which would mislead short-run outcomes.
5. Conclusions
The outcomes presented in this paper show that Puerto Rico is more sensible to US
almost 4% to a 1% change in US permanent shocks in the long run. In the short run, QPR
The long-run results are consistent with the increase of RGDP per capita in Puerto Rico
over the 60-year period. However, in Puerto Rico, regrettably, only the annual GDP is
determined and this frequency does not accurately describe short-run fluctuations as is
the case with quarterly data. In addition, another shortcoming is that the annual GDP in
Puerto Rico is accounted for from July to June, while the US GDP is accounted from
January to December. Evidently, the Puerto Rican GDP is reported with some delay.
Conference Board in the US and the Government Development Bank in Puerto Rico,
4
In contrast, after 1993 (when NAFTA was adopted), as reported by Herrera (2004), the Mexican RGDP
became less sensible to US permanent shocks than US transitory shocks. In the first case, the real output
increased by less than 1% in response to a 1% rise in US permanent shocks. In the second case, the real
output increased by almost 4% in response to a 1% change in transitory shocks.
5
Bram, J., F. Martinez, and C. Steindel (2008).
9
these indexes are recorded monthly on the basis of the construction, manufacturing, and
service sectors, as well as job growth and consumer outlays. Thus, these indexes might be
useful to short run co-features between both countries, because they accurately account
for short-run fluctuations. In this sense, Engle and Kozicki (1993) and Vahid and Engle
10
References
Bram, J., F. Martinez, and C. Steindel (2008) “Trends and Developments in the Economy
of Puerto Rico,” New York Federal Reserve, Current Issues 14(2).
Engle, R., and J. Issler (1993) “Common Trends and Common Cycles in Latin America,”
Revista Brasileira de Economia 47, pp 149-176.
Engle, R., and S. Kozicki (1993) “Testing for Common Features,” Journal of Business
and Economic Statistics 11, pp. 369-396.
Engle, R., and C. Granger (1987) “Cointegration and Error Correction: Representation,
Estimation and Testing”, Econometrica 55, pp 251-276.
Herrera, J. (2004). “Business Cycles in Mexico and the United States: Do they share
common movements?” Journal of Applied Economics, 7(2), pp. 303-323.
Mejia, P. (2000) “Asymmetries and Common Cycles in Latin America: Evidence from
Markov Switching Models,” Economia Mexicana 9, pp. 189-225.
Vahid, F., and R. Engle (1993) “Common Trends and Common Cycles,” Journal of
Applied Econometrics, 8, pp. 341-360.
Vahid, F., and R. Engle (1997) “Codependent Cycles,” Journal of Econometrics 80, 199-
221.
11
Figure 1: US RGDP vs. Puerto Rico RGDP
(1947-2008)
30.5
24
30.0
23
29.5
22
29.0
21 28.5
28.0
20
50 55 60 65 70 75 80 85 90 95 00 05
12
Table 1: Unit Root Test (ADF)
Null Hypothesis P-value
1 -9.50551
2 -9.6399
3 -9.64575*
4 -9.55505
5 -9.46928
13
Table 3: Johansen's Trace Test (restricted trend)
Null Hypothesis: # of Trace Critical Value at
Eigenvalue P-value
cointegrating vectors Statistic 5%
Table 6: US Real GDP and Puerto Rico Real GDP Common Dynamics
Long Run
Normalized cointegrating vector (β)
log(QUS) log(QPR) TREND
1 -0.22 -0.02
(-0.03027) (-0.00149)
Short Run
Normalized common feature vector( )
Δlog(QUS) Δlog(QPR)
1 -0.53
(-0.1985)
14