Professional Documents
Culture Documents
CHAPTER I
INTRODUCTION
The Philippine trade deficit amounted to $55 billion in 2007. Before that, the trade
balance had already been in deficit since 2001. In fact, from 1990 to 2007, it only had
trade surpluses in 1999 and 2000 (NSCB, 2008). While still a subject of debate, the
overvalued exchange rate has been blamed as the prime culprit for the persistent trade
deficit of the country (Bautista, 2003). The overvaluation encouraged imports at the
country the Philippines needs imported capital to grow (as cited in Lim, 1992). However,
huge and persistent trade deficits deplete foreign currency reserves needed to finance
debts (Yap, 1988) and seem to hold back growth (Fabella, 2004). Lim (1992) also
2
pointed out that in a country severely in debt like the Philippines, exchange rate policy is
The Philippines had a fixed exchange rate regime up until the 1980’s. The
Philippine exchange rate followed the Bretton Woods system of exchange rate parity
among IMF member-countries, and devaluations were done only in response to foreign
exchange crises (Bautista, 2003). Major examples include the devaluations of 1962,
1970, 1983, and 1984. Of these, only the devaluation of 1962 led to an improvement of
the trade balance and this was because it came during the liberalization and the shift away
from import-substituting industrialization (Lim, 1992). A more recent event was the
Asian Financial Crisis of 1997. Overvaluation was also cited as a major cause (Sta. Ana,
2008). Because the crisis forced the peso to weaken drastically, the problem of
overvaluation lessened after 1997. However, the exchange rate has again become an issue
as the peso again strengthened due to remittances (Bautista, 2003). Sta. Ana III et al
(2008) estimated that peso overvaluation reached almost at 1.5% and have pushed for
Economic theory maintains a strong and close relationship between the exchange
rate and the trade balance. If the sum of import and export elasticities is greater than
rate will improve the trade balance. This relationship makes the exchange rate an
of the exchange rate and the trade balance. These are to: test for the Marshall-Lerner
estimates of trade elasticities. Detecting S-curves on the other hand can be done by
analyzing the cross-correlation function between the trade balance and the terms of trade.
Backus et al (1994) found that the trade balance was negatively correlated with present
and future movements in the terms of trade but positively correlated with the past
that the Marshall-Lerner condition may not always apply. Others (Bahmani-Oskooee and
Ratha, 2004) see the Marshall-Lerner condition as a long run result of devaluations and
hypothesize that in the short-run, the trade balance actually deteriorates before it starts to
This study shall attempt to clarify some issues regarding exchange rate policy. A
position. However it may also lead to higher prices and cause a contraction in aggregate
demand. Edwards (1985) found that devaluations have a contractionary short-run effect
on output in in Least Developed Countries (LDCs) but also that output expands after one
year. Moreover the Bangko Sentral ng Pilipinas or the BSP traditionally dislikes a
2008). Results from the study are hoped to assist the country’s policymakers in
formulating strategies to address the country’s trade deficits and foreign debt. It is also
hoped that the study will provide a reasonable timeframe for estimating the time expected
for reforms to take effect. Knowledge of the dynamics of trade adjustment to exchange
4
rate changes will also help policymakers deal with and adapt to possible tradeoffs.
Examining the Philippine trade adjustment will also contribute to the debate on the
conflicting goals of government; overvaluing the peso to facilitate debt repayment and
foreign investment versus undervaluing the peso to make local exports more competitive.
The study will also contribute to the local literature by using up to date methods and
models. What few studies done in the country used questionable methods and therefore
can still be improved upon. Most did not check the time-series properties of the variables
and did not take into account the feedback effects that are present in flexible exchange
rate regimes. The study will also avoid aggregation bias and the difficulty of constructing
accurate proxies for world variables by adopting the recent trend of using a bilateral
approach.
trade deficits in particular, complicate and hamper growth. Devaluation has been a
conventional tool to improve the trade balance. However, this has achieved mixed results.
Major devaluations cited earlier did not lead to an improvement in trade except for the
The study will attempt to answer the following questions: a) does devaluation
actually improve the Philippine trade balance? b) does the trade balance adjustment
The general objective of this study is to determine the path of trade balance
adjustment to changes in the exchange rate. Specifically the study seeks to:
1. describe the trade balance of the Philippines with its top two trading partners
CHAPTER II
Traditional trade models maintain a strong relationship between the trade balance
and the exchange rate. Satisfaction of the Marshall-Lerner condition indicates that a
depreciation improves the trade balance, however this does not take time into account
(Blanchard, 2000). In many cases, depreciation does not improve the trade balance
the trade balance improves, making the path of adjustment look like a J. Accordingly,
this phenomenon is termed the J-curve phenomenon (Krugman, 2003). Knowledge of this
adjustment process is important for policymakers since they have to consider that they
may have to deal with lower net exports that contract output, at least in the short-run.
detect and quantify J-curves for several countries using different techniques (Bahmani-
him the J-curve results from contracts in force before devaluation but are completed after
7
devaluation. The trade balance improves when new contracts that take the new exchange
rate into account begin to dominate the transactions. Junz and Rhomberg (1973)
supplemented Magee’s explanation. They identified five lags that explain the delay in the
adjustment of the trade balance. A recognition lag reflects the time for exporters and
importers to perceive the changed situation. Upon recognition, there is a decision lag in
changing orders of products. Even if this decision lag is short, lags in delivery also slows
adjustment. There may also be a need to use up existing stocks before new orders can be
made. Finally, a production lag exists when producers have to decide how and whether to
reallocate their production. Taken together, Junz and Rhomberg estimate that such
2.2 Empirical Studies on the J-curve Phenomenon using Aggregate Trade Data
These studies were followed by others which then tried to empirically investigate
the effect of devaluation on the trade balance. In 1979, Miles contended that these studies
were lacking in that they: did not determine if trade balance responses were temporary or
permanent, did not compare trade balance levels before and after devaluation, did not
incorporate other variables that may affect trade such as government policy. Miles’
model included differentials between a country and the rest of the world of the following
consumption to output, and the exchange rate. Employing annual data from 14 countries
during the period 1956-1972, Miles concluded that devaluations improved the balance of
payments and the capital account, not the current account. The improvement in the
8
capital account meant that the effect of devaluation was on portfolio holdings. However,
these results were later disputed. Himarios (1985) pointed out that the real exchange rate
should have been used, not the nominal and that lagged real exchange rate values should
be included. Including these variables, Himarios found that devaluations of the real
expanded Krueger’s model (1973) and regressed the trade balance on domestic and
foreign income, domestic and foreign levels of high-powered money, and imposed an
Almon lag structure on real exchange rate values divided by the domestic price level
(E/P). With respect to the real exchange rate lags, an initially negative coefficient
Four LDCs with different exchange rate regimes were examined; Greece, India, Thailand,
and Korea. J-curves were detected in Greece, India and Korea to different extents.
However, Bahmani-Oskooee redefined the real exchange rate variable in 1989. The
reasoning was that since P represents the domestic price level, E should have been
defined as the units of domestic currency per unit of foreign currency, not the other way
around. This revision produced an inverse J-curve for Greece, India and Korea.
Noland (1989) carried out a similar study for Japan. The author estimated
Japanese trade response to relative price changes through a gamma distributed lag model.
A different approach was put forth by Demirden and Pastine (1995). The authors
exchange rate regimes since exchange rate changes will tend to affect and be affected by
9
other variables such as income and these “feedback effects” are not reflected in OLS
estimation. Feedback effects are not as significant in fixed exchange rate regimes since
devaluations are done by government, not by the market. As such, Demirden and Pastine
used a VAR model to “endogenize” the variables. Their results suggested that feedback
A more recent study was done by Moura and da Silva (2005) for Brazil. Impulse
response functions were used in an array of models that included Markov-switching and
vector error-correction models. They found that the Marshall-Lerner condition holds for
2.3 Empirical Studies on the J-curve Phenomenon using Bilateral trade data
So far most of the cited papers used aggregate trade data. Bahmani-Oskooee and
Brooks (1999) suggest that this approach may result in a sort of aggregation bias. That is,
a country’s trade balance and effective exchange rate could be improving with one
country while simultaneously worsening with another. As such, more recent studies use
bilateral data. Employment of bilateral data is also useful because there is no need to
construct a proxy for income of the rest-of-the-world (Rose and Yellen, 1989).
Rose and Yellen pioneered this approach in 1989 (as cited in Bahamni-Oskooee
and Ratha, 2004). The authors employed US quarterly data from the period 1963-1988.
The log-linear model included as explanatory variables US real GNP, real GNP of trade
partner, and the real exchange rate of the US$ with the currency of the trade partner.
Existence of a single unit root for all variables required first differencing to attain
10
stationarity. Rose and Yellen also tested for co-integration based on the method of Engle
and Granger and used instrumental variables in an attempt to deal with simultaneity bias
among the variables. An application of their model to Germany and Italy yielded negative
findings for the J-curve effect. The authors put this down to possible simultaneity of the
trade balance, exchange rate and output, and the complications arising from the first
differencing procedure.
Marwah and Klein (1996) also conducted a study for US and Canada and their
five largest trading partners using quarterly bilateral data. Instead of defining the trade
balance as the difference between exports and imports, the authors expressed the trade
balance as the ratio of a country’s exports over its imports. Bathmani-Oskooee and Ratha
(2004) suggest that this is advantageous because it makes the trade balance variable unit-
free in addition to allowing a log-linear specification of the model. Marwah and Klein’s
model included as explanatory variables the volume of world trade at constant 1985 US$
prices over GNP of the home country, the real exchange rate of the home country with its
trading partner. Employing the instrumental variable method and imposing a polynomial
distributed lag scheme on the exchange rate, they found evidence for similar J-curves in
Another study by Bahmani-Oskooee and Brooks (1999) pointed out flaws in the
two previous studies. With respect to the study of Rose and Yellen (1989), Bahmani-
Oskooee and Brooks assert that defining the trade balance as net exports makes their
findings sensitive to the unit of measurement, and that they should have used error-
correction modeling rather than the Engle-Granger co-integration analysis and the
Augmented Dickey-Fuller Test. Bahmani-Oskooee and Brooks also point out that the
11
study of Marwah and Klein (1996) suffered from the use of nonstationary data and may
have led to a spurious regression problem. The authors corrected for these and employed
a model similar to that of Rose and Yellen. Instead of the Engle-Granger co-integration
analysis the authors used the error-correction modeling and the Autoregressive
Distributed Lag Approach pioneered by Pesaran and Shin in 1995. The model was used
to examine US bilateral trade relations during the period 1973-1996 with France,
Germany, Italy and Japan, UK, and Canada. In general Bahmani-Oskooee and Brooks
found that dollar depreciation produced mixed short-run trade responses with the six
countries but that in the long run, bilateral US trade balances improved.
disaggregated approach, this time focusing on 108 industries in the US and Australia that
engage in trade. The authors used annual data and co-integration analysis on each
industry and found that only in 35 industries was there evidence of a J-curve.
industries in the US. Monthly data was used covering the period January, 1991 to
August, 2002. Of the 66 industries, only six (6) exhibit the J-curve effect.
These two studies were the first two studies to employ data at the industrial level
(Bahmani-Oskooee and Ardalani 2007). Thus, these reflect the most recent trend in J-
curve analysis.
12
were done on the Philippine trade situation. Dowling (1973) calculated price elasticities
of exports and imports to prove the Marshall-Lerner condition (as cited by Duldulao,
2004). The previously cited study of Miles (1978) also included the Philippines among
the 18 countries examined. With respect to the Philippines, Miles found that devaluation
A number of theses also examined the Philippine trade balance. Cruz (1986)
found exchange rate lags of up to two years are significant in explaining the trade
balance. In particular, the first lagged value is positive while the second is negative. This
means that devaluation only causes a temporary improvement in the Philippine trade
balance, a finding contrary to the J-curve. Cruz interpreted this to mean that the
Lansigan and Lopez (1991) on the other hand employed a variant of Bahmani-
Oskooee’s 1985 model to detect a J-curve for the Philippines. The authors regressed the
trade balance on real foreign and domestic income, real foreign and domestic high-
powered money, real domestic government expenditures, and the real effective exchange
rate. Lagged values of real domestic and world income, and real domestic high-powered
money were found to negatively affect the trade balance while the real effective exchange
Duldulao (2004) pointed out that these studies did not take the time-series
properties, in particular stationarity, of the variables into account. Using data from 1989-
13
variables real domestic and foreign income, contemporaneous real effective exchange
rate, real and foreign high-powered money, and domestic government expenditure. The
author also followed Bahmani-Oskooee in defining trade balance as the ratio of exports
to imports. Duldulao determined that only real domestic and foreign income and real
domestic high-powered money significantly affect the trade balance. The author also
tested for the Mashall-Lerner condition which was then found not to apply to the
Philippine setting.
14
CHAPTER III
THEORETICAL FRAMEWORK
Since exchange rate policy is a major option for countries facing persistent trade
deficits, it is important to understand the relationship between the trade balance and
exchange rate. In this regard, there are three approaches which are used to explain the
The elasticities approach assumes that the increases in the nominal exchange rate,
improves the trade balance through a relative price effect. An increase in the exchange
rate makes imports more expensive and exports cheaper, discouraging consumption of
imports and encouraging production of exports. This holds true when the sum of export
and import demand elasticities exceed unity, or when the Marshall-Lerner condition is
satisfied for the country (Krugman, 2003). This approach is closely related with a
Keynesian framework and the improvement in the trade balance is also accompanied by
rate increases improve the trade balance through expenditure-switching and expenditure
reducing policies. In general, the trade balance improves if domestic production exceeds
domestic absorption. Devaluations facilitate this in two ways. The first is when
devaluations, through a process similar to the elasticities approach, push domestic and
foreign expenditures toward domestic goods. The second is when devaluations are
exports relatively cheaper and the resulting excess demand produces inflation. Inflation
15
effectively reduces real balances which in turn may decrease demand for goods (Cruz,
1986).
In contrast to the two approaches, the monetary approach argues that increases in
the exchange rate only cause a temporary improvement in the balance of payments
competitiveness of the country’s exports. However, this leads to a rise in money stock
which will in time increase aggregate demand and therefore prices. Once money supply
and prices rise enough to counter the short-run relative price effect of the exchange rate
Even if exchange rate policy does impact the trade balance, empirical studies
cited earlier (Bahmani-Oskooee and Ratha, 2004) have shown that the effects may be
delayed over a time period. Depreciation may initially worsen the trade balance before
1999). The exchange rate increase temporarily makes exports cheaper and imports more
expensive, so if adjustment is slow then export receipts will be lower while import
payments will be higher. Eventually as adjustment goes on the trade balance will
improve. This pattern of trade balance adjustment is termed the J-curve phenomenon.
16
increases the deficit to OB because export earnings have become smaller while import
payments became larger. Over time exports and imports adjust and net exports equal its
CHAPTER IV
METHODOLOGY
This study used a bilateral approach to reflect recent trends in the literature. In
particular, the study focused on the top two trading partners of the Philippines, the United
States and Japan. To integrate the three approaches enumerated earlier, the trade balance
was specified as a function of real domestic income, real foreign income, domestic
government expenditure, real domestic high-powered money, real foreign high powered
money and the real effective exchange rate. The general form of the trade balance
where TB is the trade balance, YD is real domestic income, YW is real foreign income,
foreign money supply, and RER is the real bilateral exchange rate.
TB is defined as the ratio of exports to imports. As mentioned earlier, this has the
advantage of making the trade balance variable unit-free and can be interpreted as either
YD is defined as real gross national product of the Philippines. Its expected effect
on the trade balance is negative because a rise in national income increases demand for
imports leading to a deterioration of the trade balance. It can also be positive however. As
Magee (1973) argued, domestic production of importables can also rise faster than the
18
rise in domestic consumption of these such that the net effect will be a decrease in
imports.
related with the trade balance. An increase in world income increases demand for
exports, improving the trade balance. However, as argued again by Magee (1973), the
may deteriorate the trade balance therefore the expected effect is negative (Blanchard,
2001).
money supply raises the public’s demand for goods and services. As individuals perceive
their wealth to their expenditures will increase compared to income, the trade balance
would eventually worsen. However, Miles (1979) also gave reasons that may cause the
effect on the trade balance to be positive. For one, nominal money balances may only be
a small portion of wealth. Money might not also be seen as net wealth by the private
sector. Even if it is, the resulting increase in expenditures may be too small so as not to
MF is defined as the level of trading partner money. The expected effect of this is
opposite of that of MD. Again, for reasons outlined above, the effect may also be
negative.
RER is the real bilateral exchange rate of the Philippines with the trading partner.
The impact of changes in the ER is indefinite. All things held constant, a depreciation
19
makes exports cheaper such that demand for them expands and improves the trade
balance. However, the depreciation also makes imports relatively more expensive so if
adjustment in the quantity of imports is sluggish, the trade balance may deteriorate in the
Previous studies (Cruz, 1986 and Lansigan, 1991) on detecting a J-curve for the
Philippines did not attempt to verify if their time series were stationary or that there
might exist a long-run relationship between the variables. Duldulao (2004) tested the
relationship of the trade balance with the contemporaneous value of the real exchange
The procedure followed four stages. Unit root tests were done on each of the
appropriate lag length on the basis of several information criteria. Johansen’s test was
then used to check for co-integration. If the test suggests a co-integrating relationship, a
vector error correction model was estimated. If the variables do not co-integrate, then the
nonstationary variables were differenced and then applied together with the stationary
variables (if any) in a vector autoregression model. In order to aid in the interpretation of
the model, impulse response functions will be used to examine the effect of shocks on the
trade balance.
20
Regression analysis assumes that the data are stationary or that its mean, variance,
and autocovariances do not vary with time. Ignoring these assumptions in a time series
analysis exposes one to the problem of spurious regression. However, most time series
data are in fact nonstationary. High R2 and t-ratios that may lead to the conclusion that
there is a relationship between variables when there should not be any. Also, each set of
time series data from a specific period cannot be generalized to other time periods
(Gujarati, 2003).
Most macroeconomic time series data exhibit nonstationarity, hence detecting and
dealing with it should be done before estimation. Nonstationary time series can only be
studied for the time period under consideration and its behavior cannot be generalized
into other time periods. Such time series also have a variance that tends toward infinity
since the entire value, including the error term, is carried over from the previous period to
the next.
A formal test for determining stationarity is to look for unit roots using the
Augmented Dickey Fuller or the ADF test. The test involves estimating a regression such
as:
m
Yt = 1 + 2 t + Yt −1 + i Yt −i + t ,
i =1
21
where Yt is the variable of interest in first difference, and lagged difference
m
terms t Yt −i are to be included to ensure that the error term 𝜀𝑡 is serially
i =1
uncorrelated. The null hypothesis (H0) is δ=0 or that the time series is nonstationary. This
is tested against the alternative hypothesis (H1) is δ≠0 or that the time series is stationary.
If the calculated tau statistic is smaller (i.e., more negative) than the critical value, then
H0 is rejected and the series is deemed to be stationary or I(0). If on the other hand the
calculated statistic is larger than the critical value, then H0 can be accepted and the time
series is nonstationary or that it contains a unit root. First differences taken from variables
found to be nonstationary in levels were again subjected to the ADF test to determine if
Choosing the optimum lag length for the tests is important since too few lags may
lead to omission of relevant variables while too many lags may reduce efficiency and
degrees of freedom (Gujarati, 2003). Model selection criteria such as the Akaike (AIC),
Schwarz (BIC) and Hannan-Quinn (HQC) criteria will be the basis for selecting the
appropriate lag length. The model with the lowest value of the information criteria will be
used. If each criterion suggest differing lag lengths, Johansen et al (2000) state that the
HQ criterion is to be preferred.
22
co-integration entails testing the error term for stationarity. However, this study made use
of Johansen’s co-integration test since there are multiple variables involved and there
may be more than one co-integrating relationship (Harris and Sollis, 2003). The Johansen
approach is based on the vector autoregression (VAR) approach (Greene, 2003). Defining
zt as the vector of the endogenous variables [TB RER YD MD GOV] in the model and k
as the lag length, an error-correction representation of the VAR may be written as:
also equal to αβ’ where α denotes speed of adjustment while β is a matrix of long-run
independent columns there are in П. If П has full rank then all variables in z are
stationary and the appropriate model is a VAR in levels while if П has zero rank then
there are no co-integration relationships and the appropriate model is a VAR in first
differences (Harris and Sollis, 2003). If П has reduced rank then there are r < (n-1) co-
integration vectors. The Johansen approach makes use of two tests to determine co-
integration rank, the trace test and the Lmax test, although the trace test is to be preferred.
23
The null hypothesis is that there are at most r co-integration vectors versus the alternative
hypothesis that there are r +1 co-integration vectors. If the trace statistic is greater than
its critical value then the null hypothesis will be rejected. The co-integration rank will be
analysis. If they are not co-integrated then nonstationary variables will be differenced and
a VAR model will be estimated containing the transformed and the stationary variables.
The general form of the vector autoregression with constant if all variables are stationary
is:
zt = A0 + A1zt-1+ …+ Akzt-k + ut
correlated, but are uncorrelated with their own lagged values and uncorrelated with all of
variables in the right-hand side. The expanded form of the model is presented in
Appendix 2.
Impulse response functions will be used to observe the effects of shocks of the
real exchange rate variable on the bilateral trade balance. Impulse response functions
were used to observe the response of the bilateral trade balance to shocks in the real
24
bilateral exchange rate (Gujarati, 2003). Use of the IRF will show whether the plot of
forecast values of the bilateral trade balance in response to a shock in the bilateral
exchange rate follow a J-curve pattern. The expanded form of the model is presented in
Appendix 2.
them. If one variable is above the level suggested by another variable with which it is co-
integrated, then the former can be expected to decrease in the next period. Thus
variable as an error correction term (Salvatore and Reagle, 2002). Bahmani-Oskooee and
Brooks (1999) also contend that error-correction modeling can be a way of detecting the
J-curve effect since it incorporates short and long-run dynamics of the trade balance into
one model. A vector error correction model can help in the detection of J-curves through
the signs of the short-run coefficients as well as the value and significance of the error-
exchange rate variable followed by positive ones. The impulse response of the trade
balance to the real exchange rate will also be plotted to determine whether the response
of the bilateral trade balance to a shock in the bilateral exchange rate follows a J-curve
pattern.
25
The study used quarterly data from the first quarter of 1985 up to the fourth
quarter of 2007 and obtained a sample size of 92 observations. The data used were
quarterly, from 1985-2007. This period was chosen to include only the data from the
flexible exchange rate regime. Most data were gathered from the International Financial
Statistics (IFS) and Direction of Trade Statistics published by the International Monetary
Fund. Transformations were also done in order to ensure uniform units of measurement.
transformed to be Php over national currency of trading partner. The real bilateral
Other nominal data were converted into real terms using real terms using the
consumer price index (2000=100) and were converted into dollars such that the dataset
was composed of real data in billion US dollars. The resulting datasets for the
respectively.
26
CHAPTER V
This section is divided into four parts. It begins by discussing the trends in the
bilateral trade balance and bilateral exchange rate data were. It then presents the results of
the tests for nonstationarity using the Augmented Dickey-Fuller test. The lag lengths used
in the ADF test were determined by maximizing adjusted R2. Nonstationary variables
were transformed and then tested again to ensure that the transformed data were now
stationary. This also helped in identifying the order of integration. Third, the optimum lag
lengths for the Philippines-US and the Philippines-Japan models were selected based on
the Hannan-Quinn Criterion (HQC). Fourth, the models were tested for co-integration. A
vector autoregression was estimated in the model with no co-integration while a vector
error-correction model was estimated in the model with co-integration. In both cases
impulse response functions were plotted to trace over time the expected responses of the
Ratio of Philippine exports to imports with the United States averaged 1.26 for the
period 1985-2007. This means that for the period, exports to the US have been on
average, 1.26 times larger than imports from the US. The highest export to import ratio of
2.09 occurred in 2001 with exports twice the amount of imports. The next three highest
export to import ratio occurred in 1999, 2000, and 2002, making that period a relatively
27
good one for Philippine bilateral trade with the US. However, recent performance has
been worse. The lowest trade ratio in the period was recorded in 2004 with exports just
70.45% of imports. For the last five years the trade ratio has been very close to unity.
On the other hand, Philippine trade with Japan averaged 0.76. This means that for
the period 1985 to 2007, exports to Japan averaged only 76% of imports from Japan.
While exports were 1.31 times larger than imports at the start of the period, the ratio of
exports to imports steadily declined until it reached its lowest point of 0.33 in 1994. Since
then it has risen and reached balance in 2004. From 2004 onwards the ratio of Philippine
exports to imports with Japan has swung between .80 and 1.2 while averaging 1.03.
Philippines-US Exports/Imports
2.5
2
1.5
1
0.5
0
Q1 1985
Q1 1986
Q1 1987
Q1 1988
Q1 1989
Q1 1990
Q1 1991
Q1 1992
Q1 1993
Q1 1994
Q1 1995
Q1 1996
Q1 1997
Q1 1998
Q1 1999
Q1 2000
Q1 2001
Q1 2002
Q1 2003
Q1 2004
Q1 2005
Q1 2006
Q1 2007
Philippines-Japan Exports/Imports
1.4
1.2
1
0.8
0.6
0.4
0.2
0
Q2 1987
Q3 2001
Q2 2005
Q1 1985
Q4 1985
Q3 1986
Q1 1988
Q4 1988
Q3 1989
Q2 1990
Q1 1991
Q4 1991
Q3 1992
Q2 1993
Q1 1994
Q4 1994
Q3 1995
Q2 1996
Q1 1997
Q4 1997
Q3 1998
Q2 1999
Q1 2000
Q4 2000
Q2 2002
Q1 2003
Q4 2003
Q3 2004
Q1 2006
Q4 2006
Q3 2007
Figure 5.1.2 Philippines-Japan Exports/Imports
Source: Author’s computations based from the Direction of Trade Statistics data
For the period under consideration, the real bilateral exchange rate with the US
averaged 40.37 pesos to a dollar. The maximum of 52.21 pesos to a dollar occurred in
2004 while the minimum of 28.99 occurred in 1997. The real bilateral exchange rate was
around 40 pesos per dollar until the early 1990’s. This was followed by an appreciation of
the real exchange rate. Following the 1997 crisis the bilateral exchange exhibited an
In contrast with the relatively high exchange rate with the US, the real bilateral
exchange rate with Japan averaged only 0.35 peso to one yen. The lowest value of 0.19
occurred in 1985, which is also the start of the period. On the other hand, the maximum
of 0.45 occurred in 2004, which is towards the end of the period. The maximum of 0.45
50
40
30
20
10
Q1 2003
Q1 2006
Q1 1985
Q4 1985
Q3 1986
Q2 1987
Q1 1988
Q4 1988
Q3 1989
Q2 1990
Q1 1991
Q4 1991
Q3 1992
Q2 1993
Q1 1994
Q4 1994
Q3 1995
Q2 1996
Q1 1997
Q4 1997
Q3 1998
Q2 1999
Q1 2000
Q4 2000
Q3 2001
Q2 2002
Q4 2003
Q3 2004
Q2 2005
Q4 2006
Q3 2007
Figure 5.1.3 Real bilateral exchange rate defined as peso over dollar
Source: International Financial Statistics
Figure 5.1.4 Real bilateral exchange rate defined as peso over yen
Source: International Financial Statistics
This study therefore presents two contrasting cases. As shown in Figures 5.1.1
and 5.1.3, the Philippine trade situation with the US has been characterized by
comparatively high peso-dollar exchange rates and trade surpluses. On the other hand,
30
Figures 5.1.2 and 5.1.4 show in trade with Japan, the Philippines has had generally low
peso to yen exchange rates and chronic deficits. The summary statistics and the graphs of
the other variables are presented in the Appendix 5. Most of these variables seem to
exhibit seasonality.
The ADF test was carried out on all variables in levels. For each variable, the lag
length that led to the highest R2 was chosen. If the variables were found to be
nonstationary, their first differences were also tested. Stationary first differences mean
that the nonstationary variables are integrated of order 1 or I(1). The results of the tests
1
The critical value at the 5 percent level of significance is -3.41.
31
the 5% level of significance. Their first differences were also tested and these were found
to be stationary, indicating that they are integrated of order 1. . However, the result that
Philippines-US trade ratio is already stationary or I(0) means that it cannot be co-
integrated with the other variables of the model.. The Philippines-US model will then be
estimated by a VAR with the Philippines-US trade ratio in levels while the other
variables are in first differences. On the other hand, all the Philippines-Japan variables
are integrated of the same order so it is still possible for co-integration in that model to
exist.
Various criteria were used to help in choosing the lag lengths. The two criteria
presented in Table 5.3 are the Schwarz/Bayesian Information Criterion (BIC) and the
Hannan-Quinn criterion (HQC). The optimal lag length is the one that minimizes the
criteria. For the Philippines-US model, the two criteria suggest an optimum lag length of
three (3) periods. However, for the Philippines-Japan model, the two criteria suggest
different lag lengths. The BIC suggests a one period lag and while the HQC recommends
a lag of three periods. In such a situation, Harris (2003) cited that the common practice is
to follow the HQC. In any case, it is less dangerous to include excessive lags than
insufficient lags. Given the findings above, both models will be estimated with three lags.
32
Table 5.3 Optimum lag length selection based on SIC and HQC
Philippines- Philippines-
US Japan
The VAR for the Philippines-US model contained a mix of variables in levels and
first differences. Following the results of the ADF tests, the trade balance was expressed
in levels while the other variables were expressed in first differences. US money and
GDP entered the model as exogenous variables because there is no reason to expect that
Philippine policy can affect them. The relevant equation is presented in Table 5.4 and the
entire VAR output is presented in Appendix 6. The results indicate that many of the
estimated coefficients are not individually statistically significant on the basis of the t-
test. This is possibly due to multicollinearity arising from the inclusion of so many lags in
the equation system. However, the equation for the trade balance as whole is statistically
significant because the F-statistic 7.319442 is greater than the critical value at the 5%
level of significance.
33
YW 0.041 0.91
MW -0.043 -0.51
34
Since the coefficients in Table 5.4 are difficult to interpret, the impulse response
of the trade balance to a shock in the bilateral exchange rate was plotted. Figure 5.5 is
shown here because it was difficult to determine the magnitude of the impulse responses.
The impulse responses shown here are responses to Cholesky one standard deviation
Despite the evident seasonality, Figure 5.5 shows a clear J-curve pattern.
Following the shock, the trade balance initially worsens for two periods before beginning
to improve. From the third until the thirtieth quarters, the depreciation shock improves
the trade balance. The shock eventually results in a bigger improvement than the initial
deterioration with the peak arriving seven periods after the shock. The path to
35
improvement is three periods longer than what Miles (1978) found, but both findings
seem to support evidence for a J-curve. The following periods then show a gradual
dissipation of the effect of the exchange rate shock. The effect of the shock was
The analysis for the Philippines and Japan will attempt to look into possible long-
run relationships because all the variables are integrated of order one. Since there are
greater than two variables involved, co-integration will be tested for using the Johansen
approach. As in the previous model, Japan money and GDP were treated as exogenous. A
model with “unrestricted constant” and “restricted trend” was adopted since the data in
levels seemed to exhibit trends while their first differences did not2. The trace and max-
eigenvalue tests were carried out to determine the number of co-integrating equations.
The two tests to determine the co-integration rank of the Philippines-Japan system
both suggest the existence of one co-integrating equation at the 0.05 level of significance,
as shown in Table 5.6.1. The null hypothesis of no co-integration was rejected while the
null hypothesis that there is at most one co-integrating equation was accepted by both
tests. Since the Johansen tests indicate the existence of a long-run relationship, a vector
2
There are five possible cases as shown by Johansen (1995) and cited by Cottrell (2008). These are: Case
1 – no constant, Case 2 - restricted constant, Case 3 – unrestricted constant, Case 4 - constant + restricted
trend, Case 5 - constant + unrestricted trend. Cottrell summarized the criteria for choosing among the cases.
If the variables exhibit a linear trend then the constant should not be restricted and Case 3 is used. Also if
taking first differences eliminates the linear trend then Case 4, unrestricted constant and restricted trend, is
chosen.
36
error-correction model with three lags and co-integration rank 1 was estimated. Again,
Japan money and GDP were treated as exogenous. The relevant equation from the vector
error correction model is in Table 5.6.2 and the entire output is presented in Appendix 7.
Hypothesized Trace
No. of CE(s) Eigenvalue Statistic Prob.**
EC -1.98 0.30
The model has a noticeably lower R2 than the Philippines-US model. Many of the
lagged variables are also statistically insignificant, possibly due to multicollinearity. The
insignificance of the error term however, suggests that the trade balance adjusts to
An impulse response function is again used here to capture short run dynamics of
the trade balance response to the peso-yen exchange rate. The response of Philippines-
Japan trade was forecasted over twenty periods, or five years. Figure 5.7 showed an
unusual result, the shock in the real bilateral exchange rate worsened the trade balance for
three quarters before improving to above one in the fifth quarter. It then declines again to
another low in the seventh quarter and the settles at around -.0025. Figure 5.7 seems to
follow a J-curve pattern but it does not in the end result in a net improvement of the
Philippines-Japan trade balance. In fact, the graph seems to suggest that a shock in the
real exchange rate permanently worsens the Philippines-Japan trade balance by -.0025
every quarter. This conforms with the finding by Duldulao (2004) that the Marshall-
CHAPTER VI
This study sought to identify the key economic variables affecting the Philippine
trade balance from 1985 to 2007. The period chosen represented the flexible exchange
rate regime. Specifically this study aimed to determine the effect of each variable on the
Philippine trade balance and to determine if the J-curve phenomenon applies to the
Philippines. A bilateral approach focusing on the country’s top two trading partners, the
United States and Japan, was adopted in order to avoid aggregation bias and the difficulty
of constructing an accurate proxy for world variables. This move also reflected the recent
trend of bilateral studies on a problem that had earlier been addressed by aggregate
models. Graphical analysis showed that the Philippines has usually had trade surpluses
with the US until recent years while having chronic trade deficits with Japan. After
gathering and transforming the data to make them uniform in measurement, four steps of
analysis followed; unit root testing, lag length selection, vector autoregression or co-
integration analysis, and estimation of the impulse response functions. Unit root tests
were done to determine the order of integration of the variables. Contingent on their
outcome, Johansen’s co-integration test was done to test for a long-run relationship
among the variables. Non-existence of a co-integration relationship means that the model
impulse response functions were plotted to better illustrate the dynamics of trade balance
After summarizing the trends of the bilateral trade balances and exchange rates,
the Augmented Dickey-Fuller test was carried out on all the variables. All but one of the
variables were found to be nonstationary in levels so a second test was done on their first
differences. Their first differences were found to be stationary so it was concluded that all
these variables were integrated of order one while the Philippines-US trade balance was
the only stationary variable in levels. The finding that Philippines-US trade balance is
already stationary in levels meant that analysis of the Philippines-US model would
immediately proceed to VAR. The other variables were differenced once to make them
stationary and the US GDP and money variables were taken as exogenous. A plot of the
impulse response of the Philippines-US trade balance to shock in the peso was shown.
The response of the trade balance to a shock in the exchange rate followed the J-curve
variables were tested for co-integration. Both the trace test and the max-eigenvalue test
found the existence of one co-integrating relationship so a VEC was estimated with
cointegration rank 1. The impulse response of the Philippines-Japan trade balance was
also plotted. A seemingly J-curve pattern was also observed in the response of the trade
balance to the exchange rate but the end effect was still negative, similar to what was
found by Duldulao (2004). Thus this study provides mixed results for the J-curve
phenomenon.
41
CHAPTER VII
This study was limited to only the top two trading partners of the Philippines, the
United States and Japan. This is due to unavailability of data for some of the other large
trading partners. Most bilateral studies on the J-curve effect study five or more countries
but this option was not available in this study. Also, the methods used in the study are
typically used for annual data and might not be less than ideal for data that exhibit
seasonality. However, this was not tested for due to the complexity of the methods
involved. Future studies may expand the analysis to other trading partners, include more
REFERENCES
Ardalani, Z. and Bahmani-Oskooee, M. 2007. "Is there a J-Curve at the Industry Level?."
Economics Bulletin. Vol. 6, No. 26
Backus, D. K., Kehoe, P. J. and Kydland, F. E. 1998. Dynamics of the trade balance and
the term of trade: the J-curve? American Economic Review. 84: 84–103.
Bahmani-Oskooee, M. 1985. Devaluation and the J-curve: some evidence from LDCs.
The Review of Economics and Statistics. 67: 500–504.
Bahmani-Oskooee, M. 1989a. Devaluation and the J-curve: some evidence from LDCs:
errata. The Review of Economics and Statistics. 71: 553–4.
Bahmani-Oskooee, M. and Wang, Y. 2007. The J-curve at the industry level: evidence
from trade between U.S. and Australia. Australian Economic Papers. Vol. 46,
Issue 4
Cruz, O. 1986. The significance of exchange rate in the Philippine trade balance position.
Masteral thesis. University of the Philippines, Diliman.
Fabella, R. 2008. Peso appreciation and the sustainability of Philippine growth: need we
worry?
43
Harris, R. and Sollis, R. 2003. Applies time series modeling and forecasting. John Wiley
and Sons, Ltd.
Krugman, P. 1989. The J-curve, the fire sale, and the hard landing. American Economic
Review. 79: 31–5.
Krugman, P. and Obtsfeld, M. 2003. International Economics Theory and Policy. 6th ed.
Addison-Wesley.
Lansigan and Lopez. 1991. Is the J-curve phenomenon relevant for the
Philippines.Undergraduate thesis. University of the Philippines, Diliman.
Miles, M. A. 1979. The effects of devaluation on the trade balance and the balance of
payments: some new results. Journal of Political Economy. 87(3): 600–20
Marwah, K. and Klein, L. R. 1996. Estimation of J-curve: United States and Canada.
Canadian Journal of Economics. 29: 523–39.
Moura, G. and Da Silva, S. 2005. Is there a Brazilian j-curve? Economics Bulletin. 6(10):
1-17.
Noland, M. 1989. Japanese trade elasticities and the J-curve. Review of Economics and
Statistics. 71: 175–9.
Rodrik, D. 2008. The real exchange rate and economic growth: theory and evidence.
Salvatore, D. and Reagle, D. 2002. Statistics and Econometrics. 2nd ed. McGraw-Hill.
Wickham, P. 2002. Do “flexible” exchange rates of developing countries behave like the
floating exchange rates of industrialized countries? IMF Working Paper.
Zhang, Z. 1996. The exchange value of the Renminbi and China’s balance of trade: an
empirical study. NBER Working Papers Series, Working Paper #5771.
44
APPENDIX
45
Appendix 1. Expanded version of the error-correction representation of the VAR, as applied to the current model
TBt = 1,1TBt −1 + ... + 1, k TBt − k + 1,k +1RERt −1 + ... + 1,2 k RERt − k + 1,2 k +1YDt −1 + ... + 1,3 k YDt −k + 1,3 k +1MDt −1 + ... + 1,4 k YDt −k
+ 1,4 k +1GOVt −1 + ... + 1,5 k GOVt − k + 1 ( 11TBt −1 − 12 RER − 13YD − 14 MD − 15GOV ) + u1t
RERt = 2,1TBt −1 + ... + 2, k TBt − k + 2, k +1RERt −1 + ... + 2,2 k RERt − k + 2,2 k +1YDt −1 + ... + 2,3 k YDt − k + 2,3 k +1MDt −1 + ... + 2,4 k YDt −k
+ 1,4 k +1GOVt −1 + ... + 2,5 k GOVt − k + 2 ( 21TBt −1 − 22 RER − 23YD − 24 MD − 25GOV ) + u2t
YDt = 3,1TBt −1 + ... + 3,k TBt − k + 3,k +1RERt −1 + ... + 3,2 k RERt −k + 3,2 k +1YDt −1 + ... + 3,3 k YDt −k + 3,3 k +1MDt −1 + ... + 3,4 k YDt −k
+ 3,4 k +1GOVt −1 + ... + 3,5 k GOVt − k + 3 ( 31TBt −1 − 32 RER − 33YD − 34 MD − 35GOV ) + u3t
MDt = 4,1TBt −1 + ... + 4,k TBt − k + 4, k +1RERt −1 + ... + 4,2 k RERt − k + 4,2 k +1YDt −1 + ... + 4,3 k YDt − k + 4,3 k +1MDt −1 + ... + 4,4 k YDt −k
+ 4,4 k +1GOVt −1 + ... + 4,5 k GOVt − k + 4 ( 41TBt −1 − 42 RER − 43YD − 44 MD − 45GOV ) + u4t
GOVt = 5,1TBt −1 + ... + 5,k TBt − k + 5,k +1RERt −1 + ... + 5,2 k RERt − k + 5,2 k +1YDt −1 + ... + 5,3 k YDt − k + 5,3 k +1MDt −1 + ... + 5,4 k YDt −k
+ 5,4 k +1GOVt −1 + ... + 5,5 k GOVt − k + 5 ( 51TBt −1 − 52 RER − 53YD − 54 MD − 55GOV ) + u5t
TBt = a1,0 + a1,1TBt −1 + ... + a1,k TBt − k + a1, k +1 RERt −1 + ... + a1,2 k RERt −k + a1,2 k +1YDt −1 + ... + a1,3k YDt −k + a1,3k +1 MDt −1 + ... + a1,4 k MDt −k
+ a1,4 k +1GOVt −1 + ... + a1,5 k GOVt − k + u1t
RERt = a2,0 + a2,1TBt −1 + ... + a2,k TBt − k + a2,k +1 RERt −1 + ... + a2,2 k RERt − k + a2,2 k +1YDt −1 + ... + a2,3k YDt −k + a2,3 k +1 MDt −1 + ... + a2,4 k MDt −k
+ a2,4 k +1GOVt −1 + ... + a2,5 k GOVt − k + u2t
YDt = a3,0 + a3,1TBt −1 + ... + a3,k TBt − k + a3, k +1 RERt −1 + ... + a3,2 k RERt −k + a3,2 k +1YDt −1 + ... + a3,3k YDt −k + a3,3k +1 MDt −1 + ... + a3,4 k MDt − k
+ a3,4 k +1GOVt −1 + ... + a3,5 k GOVt − k + u3t
MDt = a4,0 + a4,1TBt −1 + ... + a4,k TBt − k + a4,k +1 RERt −1 + ... + a4,2 k RERt −k + a4,2 k +1YDt −1 + ... + a4,3k YDt −k + a4,3 k +1 MDt −1 + ... + a4,4 k MDt −k
+ a4,4 k +1GOVt −1 + ... + a4,5 k GOVt − k + u4t
GOVt = a5,0 + a5,1TBt −1 + ... + a5, k TBt − k + a5, k +1 RERt −1 + ... + a5,2 k RERt −k + a5,2 k +1YDt −1 + ... + a5,3 k YDt −k + a5,3 k +1 MDt −1 + ... + a5,4 k MDt −k
+ a5,4 k +1GOVt −1 + ... + a5,5 k GOVt − k + u5t
47
YD
0.35
0.3
0.25
0.2
0.15
YD
0.1
0.05
0
Q1 1997
Q1 2003
Q1 1985
Q1 1986
Q1 1987
Q1 1988
Q1 1989
Q1 1990
Q1 1991
Q1 1992
Q1 1993
Q1 1994
Q1 1995
Q1 1996
Q1 1998
Q1 1999
Q1 2000
Q1 2001
Q1 2002
Q1 2004
Q1 2005
Q1 2006
Q1 2007
Philippine real domestic product in billion dollars
Source: International Financial Statistics
RP GOV
0.0008
0.0007
0.0006
0.0005
0.0004
0.0003
0.0002 RP GOV
0.0001
0
Q1 1991
Q1 1995
Q1 1985
Q1 1986
Q1 1987
Q1 1988
Q1 1989
Q1 1990
Q1 1992
Q1 1993
Q1 1994
Q1 1996
Q1 1997
Q1 1998
Q1 1999
Q1 2000
Q1 2001
Q1 2002
Q1 2003
Q1 2004
Q1 2005
Q1 2006
Q1 2007
0
2
4
6
8
0.1
0
0.02
0.04
0.06
0.08
0.12
0.14
0.16
Q1 1985
Q1 1985
Q1 1986
Q1 1986
Q1 1987
Q1 1988 Q1 1987
Q1 1989 Q1 1988
Q1 1990 Q1 1989
Q1 1991 Q1 1990
Q1 1992 Q1 1991
Q1 1993 Q1 1992
Q1 1998
Q1 2000 Q1 1999
USmoney
Q1 2001 Q1 2000
Q1 2002
Q1 2001
Q1 2003
Q1 2002
Q1 2004
Q1 2003
Q1 2005
Q1 2004
Q1 2006
Q1 2005
Q1 2007
Q1 2006
Q1 2007
54
MD
USmoney
10
15
20
25
30
35
40
45
50
0
5
20
40
60
80
0
100
120
140
Q1 1985
Q1 1985
Q1 1986
Q1 1986
Q1 1987
Q1 1987
Q1 1988
Q1 1989 Q1 1988
Q1 1990 Q1 1989
Q1 1991 Q1 1990
Q1 1992 Q1 1991
Q1 1993 Q1 1992
Q1 1994 Q1 1993
Q1 2001
Q1 2000
Q1 2002
JAPANmoney
Q1 2001
Q1 2003
Q1 2004 Q1 2002
Q1 2005 Q1 2003
Q1 2006 Q1 2004
Q1 2007 Q1 2005
Q1 2006
Q1 2007
55
JAPANmoney
USgdp
56
JAPANgdp
60
50
40
30
20 JAPANgdp
10
0
Q1 1985
Q1 1999
Q1 1986
Q1 1987
Q1 1988
Q1 1989
Q1 1990
Q1 1991
Q1 1992
Q1 1993
Q1 1994
Q1 1995
Q1 1996
Q1 1997
Q1 1998
Q1 2000
Q1 2001
Q1 2002
Q1 2003
Q1 2004
Q1 2005
Q1 2006
Q1 2007
Japan real gross domestic product in billion dollars
Source: International Financial Statistics
TB 1.0000
(0.00000)
YD 341.97
(36.45)
GOV -35000.
(9380.1)
RER 86.71
(11.41)
MD -387.32
(62.20)
trend 0.61
(0.083)
TB -0.0058
YD -0.0046
GOV -6.52e-006
RER -0.0054
MD -0.00017
Log-likelihood = 1757.80
Determinant of covariance matrix = 4.81e-024
AIC = -37.48
BIC = -34.96
HQC = -36.46
60
Equation 1: TBRP-JAPAN
Equation 2: YD
Equation 3: GOV
Equation 4: RERP/Y
Equation 5: MD