One factor causing rise in import spending is the rise in domestic prices, in which case consumers tend to substitute locally produced goods for foreign products. This becomes a big problem especially if excessive importation hampers economic growth through worsening domestic demand for locally produced products. Using monthly data from the years 1999 to 2008, we investigate the dynamic effects of changes in the Philippine consumer price index to import spending. We find that over the long run, import spending is unitary elastic with respect to changes in the consumer price index, but in the short run, imports can be very responsive.
The Philippines has been experiencing a negative balance of trade in 18 out of 20 years from the years 1990 to 2009 and this trend has not shown any signs for potential improvement. As of January 2010, the Philippines experienced an import growth rate of 30.3%, which was the largest in the past 7 years. Furthermore, January import level was 9.4% higher than December 2009. (Aquino, 2010)
[Insert Table 1 here]
Having a ballooning trade deficit has several implications. One of which is a decline in competitiveness of a sector in the economy, for instance increase in labor cost. Based from the theory of absolute advantage, a country would specialize in producing goods that it will incur the lowest possible cost, which is usually those products which are made with the abundant factor in the economy; hence, it will import goods that require relatively higher cost of production, which are usually those made with scarce inputs. This means that excessive importation of a good can drive the prices of its scarce factor inputs down, and the abundant factor to go up. Secondly, a deficit can cause currency depreciation. If a country would be forced to import goods due to lack of domestic production or increase in domestic demand for imports, that country would be requiring more foreign currency; thus, eventually depreciating home currency and this would have cumulative negative effects to the economy. If a country would continuously experience currency depreciation, it will harm its image to multinational corporations that are searching for potential countries to locate their production networks, branches or subsidiaries; thus, losing millions or even billions worth of investments. Lastly, a worsening trade deficit may be a positive indicator of high growth, if high growth was coupled with high inflation. For instance, an extremely fast pace of uncontrolled growth can cause inflationary pressures, and consumers tend to escape from high domestic prices via buying imports. (Is a trade deficit harmful, n.d.)
By using a monthly dataset from 1999 to 2008 from BSP and employing dynamic econometric modelling, this paper aims to answer these questions:
This study will hopefully be able to support BSP's move to achieve its primary object and prove the importance of their inflation targeting strategy. Controlling inflation would then bring a twofold benefit, lowering domestic prices, which in turn lowers cost of living, and hampering too many imports, which when not sustained is detrimental to the economy.
This study aims to study the dynamics behind the effect of domestic prices on import spending. By looking at the dynamics behind it, the paper could try to suggest ways to control excessive importation. However, due to the technicalities of dynamic regression, we could only employ monthly data from 19992008.
There are numerous discussions on the determinants of import demand and it has been the main attraction of policymakers and people in the academic world. According to Aydin, M., Ciplak, U. & Yucel, M. (2004), the two main approaches in examining the effects of movements in exchange rate on balance of trade are “elasticities” approach and “trade balance” approach. The former is said to be an indirect representation and latter is a direct representation in investigating balance of trade relationships.
The elasticities approach is mainly focused on formulating and estimating import demand functions. As quoted by Havrila, I. (2009), “Houthakker and Magee (1969), Leamer and Stern (1970), Goldstein and Khan (1985) and Gafar (1988) emphasized that the demand for imports is directly influenced by the level of domestic income and international competitiveness,” which is represented by foreign prices of a good relative to domestic price levels.
Generally, the traditional aggregate import demand function used in the literature is represented as such:
where, M is the quantity of imports of the country, Pm is the import price index, Pd is the domestic price index and Y is the real GNP of the country. This specification allows us to examine the relative effects of prices on the import levels, not on absolute terms. In effect, this measures everything in “real” terms, which may not be useful for our study.
Other specifications of the import demand function were used in studies that came out during the late 1970s. One of these representation changes is the “splitprice specification”. (Havrila, 2009) Instead of using price ratios of foreign and domestic prices, they expressed their equations as such:
where, M is the quantity of imports of the country, Pm is the import price index, Pd is the domestic price index and Y is the real GNP of the country. Due to the nature of the study, a direct use of the import price and domestic price indices at their index form not that useful, as measuring changes in import spending entails millions of pesos in ranges, while price indices are measured on point scales.
Some of the literature that separately included import prices and domestic prices in their model specification are Murray and Ginman (1976) and Haynes and Stone (1983), and they claim that this model specification would provide a clearer analysis of movements of “foreign prices and exchange rate or domestic trade barriers”. (Havrila, 2009)
In a similar discourse made by Warner and Kreinin (1983), they included an exchange rate variable in their model for the period during the period where a floating exchange rate regime was implemented. The import demand functions that they used are as follows:
For the years 1957:1 to 1970:4 periods:
For the years 1972:1 to 1980:4 periods:
where, M is the quantity of imports of the country, Pm is the import price index, Pd is the domestic price index, Y is the real GNP of the country and ER is an importweighted effective exchange rate.
By including an exchange rate variable, the authors concluded that it seemed to have affected the “import volume in several major countries”. (Warner and Kreinin, 1983) While the inclusion of the exchange rate would be a helpful determinant of import spending, we want to isolate the effect of domestic prices, thus there is no need of inclusion of the said variable.
After providing literatures that utilized an elasticities approach, we will now proceed to the discussion of balance of trade approach. According to Aydin, M., Ciplak, U. & Yucel, M. (2004), in this specification method, “the trade balance variable (magnitude being either a monetary value or an index) is regressed on exchange rate, real income, and other related macroeconomic variables”. Miles (1979) study on the relationship of devaluation and trade balance and balance of payments is a discourse that used this approach. The author utilized the “seemingly unrelated and pooled time series regression techniques” and the equations that he used are as follows:
where, TB is the level of trade balance in country i, f.o.b. exports of goods minus c.i.f. imports, BP is the level of the balance of payments in country i, proxied by the official settlements definition, and Yi is the level of output in country i, i.e. the GNP measured in domestic currency. gi and gR are the growth rates of income in country i and the restofworld R, respectively. Mi and MR are the ratios of the average level of highpowered money to output. The variables Gi and GR are the ratios of government consumption to output of country i and the rest of the world R. The variable ER is the exchange rate of country i and the rest of the world R. All ‘rest of the world' variables are constructed using a nominalGNPweighted average of the variable in various countries.
Miles' (1979) study provided new conclusions to the world of trade balance analysis, he concluded that “a devaluation did not improve the trade balance but improved the balance of payments and that there exist no clear relationship between devaluation and the real variables”, but this result was criticized by Himarios (1985), who identified some weaknesses on the model Miles used. Himarious investigated the same study but used a longer time period in his sample. He was able to prove that there exists a relationship between devaluation and trade balances and that devaluation is potentially a good and useful policy tool.
BahmaniOksooee (1985) also used the balance of trade approach in order to prove the Jcurve hypothesis. And he successful did after estimating this model:
where, TB is the trade balance, YW is the world income, expressed as an export weighted index, M is the domestic highpowered money, expressed as an index, MW is the rest of the world highpowered money, Y is the level of the real output, E is the exchange rate, i.e. an index of export weighted effective exchange rate, and P stands for the domestic price level, index of wholesale prices.
After evaluating both specifications, we have decided to use the elasticities approach and incorporate a dynamic component to the model specification. This would allow us to measure the direct changes in percentages brought about by changes in one of the variables. While the trade balance approach takes into account the magnitudes of the different variables, it fails to take into account relative differences. Also, using the indexlevel values pose a problem to measurement since the regressand and regressors are expressed in different units.
The Bangko Sentral ng Pilipinas (BSP) has always been keen in maintaining target inflation levels to sustain economic growth. It has used different monetary policy tools in the past to control monetaryinduced inflation, brought about by excessive expansion of the monetary base.
BSP's mission is to “promote and maintain price stability in bringing about conducive to a balanced and sustainable growth of the economy.” It adopted the inflation targeting framework during January of the year 2002 as a means to attain their mission. For the year 2010, BSP inflation target is 3.5% to 4.5%.
Some of the tools of monetary policy used by BSP are open market operations, reserve requirements, discount rate, moral suasion, acceptance of fixedterm deposits, and printing new money.
This is a situation wherein BSP publicly buys or sell government securities from banks and financial institutions in order to increase or decrease money supply. This allows BSP to influence the prices of commodities and services and ensuring that it maintains its target inflation level.
Repurchase and reverse repurchase agreement is an example of an OMO. BSP temporarily buys government securities from a financial institution with the intention to sell it back at a predetermined future date at a specified rate. Repurchase transactions temporarily increases money supply and would eventually decrease it by the same level during the maturity date of the transaction. Meanwhile, outright purchases and sales of securities is a permanent purchase/sale made by BSP from/to the market.
Reserve requirements are percentages of deposit liabilities that bank must hold as cashinvault or on deposit at the BSP, which the bank cannot lend out. RR has two components, 11% statutory requirements and 8% legal requirements. These represent opportunity cost to the banking institution, since they are unusable funds stored in the vault or stored at a lower rate at the BSP.
BSP provides discount window loans for lending of reserves to borrowing institutions in exchange for promissory notes as collateral. The interest rate charged is slightly above than that of repurchase and reverse repurchase agreements.
This involves psychological pressure to bear on individuals and institutions to conform to its policies. Some examples would be press releases, letters and phone calls to erring banks.
This is rarely used by BSP due to its inflationary characteristic. Miscalculations of the amount of money to be printed could cause potential long run problems in the economy as what happened in Zimbabwe.
These tools are employed for the BSP to control and curb out unnecessary rise in prices, with the hopes of achieving macroeconomic stability.
Pure theory of international trade says that the direction of trade for a certain good will be from the low cost of production area to the high cost of production area.
Consider a twogood closed economy A with a production possibilities frontier PP'. Since it does not allow trade to happen, then its optimal production should be equal to the optimal consumption, i.e. at point E. If for example the economy operated at point K, in which more good Y was produced than the equilibrium market condition. At this point, the relative cost of production of good Y is higher than what is charged by the market for units of good Y. Assuming perfect competition, firms would quickly adjust by shifting its resources from production of Y to production of X, causing the relative cost of production to increase, with cost of producing X increasing and Y decreasing. The relative prices of the goods (Px/Py)A is determined by the slope of the line tangent to the production possibilities frontier PP' and the social indifference curve CC'. At this price, the relative cost of production is the same as the market prices, which removes any incentive for the firms to shift its volume of production away from point E in the PPF.
Now, assume that this closed economy has a single neighbour B. This closed economy A now has decided to trade with its neighbor, which produces the same type of goods that it produces. This option expands its consumption opportunities. Now, since the country decides to open its doors to international trade, optimal production point is where the relative domestic cost of production is equal to trading prices (Px/Py)W. This is represented at point D. Consumption would then be at point B. Total imports would be MM' and exports to be SS'.
Not all economies, however, can quickly adjust its production point from E to D. If, for example, the economy's production is stuck at the original level E, and the economy is allowed to trade, then it would choose to import the cheaper good. If this is the case, domestic cost of production would not be equal to world prices. Since both economies must agree on a single trading price, then the excess production of economy A, which is good X, will be sold to the international market at a higher price that the cost of production but less than the foreign cost of production. Similarly, it has to import the relatively cheaper good Y. Total imports would then be MM' and total exports would be SS'.
When the relative domestic price is less than the relative world price as in the graph above, then it means that relatively, the good being imported is cheaper when imported than produced locally. This causes the economy to just import it rather than locally producing it. In symbols,
Then either
Or
In either case, the country is importing the good that it produces at a higher cost, i.e. economy A imports good Y and economy w imports good X.
This result is sensitive to price changes. If for example, good Y in economy A has become more expensive, resulting to the relative price becoming lower, then the amount of imported good Y by the economy A is expected to increase.
Notice that since
Then
Or
And, since MM' is the import level associated with the relative price level before good Y has became cheaper in economy A, and MM” is the import level associated with the new relative price level of economy A, then notice that . Thus, the more a good becomes expensive when produced locally, ceteris paribus, an economy decides to import more of it,
Blanchard (2003), together with many economists, defined imports as a function of income and real exchange rates. Specifically, the import function can be defined as
where Y is the domestic absorption or income, and the real exchange rate.
Similarly, Blanchard (2003) also defines real exchange rate as
where is the nominal exchange rate between two countries, as the foreign price level and as the domestic price level. Thus, the ratio between the foreign and domestic price level constitute the relative prices between the two countries contemplating a trade relationship.
Assuming a linear relationship between the variables in (1.1), we can rewrite (1.1) as
where a1 and a2 are positive parameters.
Isolating the effect of domestic income and foreign prices, and taking the natural logarithm of both sides, we get
For the purpose of this study, we will use the consumer price index as a representation of domestic price level. Thus, we can write (1.4) as a regression model like (1.5) below.
for positive parameters and .
Our study would focus in a more general form of (1.5), by subjecting (1.5) to various lag specifications to see the dynamics of price changes on imports.
Data used for the study is monthly Philippine nominal import spending and Consumer Price Index (base year 2000) from 1999 to 2008. Natural logarithm of the values were taken to achieve a better fit, and also to facilitate interpretation of the coefficients in terms of elasticities. The CPI base year used was the year 2000 so that prices will be somewhat realistic, then when using the 1994 CPI base year.
Before we carry out the necessary estimation procedures, it is imperative that we first test for the presence of unit roots in the data, as well as cointegration to guard against spurious regression.
One important assumption of the Classical Linear Regression Model is the stationarity of the time series data to be used as inputs in estimation of (1.5) using OLS. By stationarity, the stochastic process has a time invariant mean and variance, and the value of the covariance between two time periods depends only on the distance or gap between these two time periods and not the actual time the covariance is computed. (Gujarati, 2003). This weak stationarity definition is the basis for the so called White Noise process, wherein the process has zero mean, constant variance and no serial correlation.
To check the stationarity of our data, we employ the Augmented DickeyFuller test in Eviews. Basically, we test for the presence of unit roots by employing a regression such as (2.1) below. A unit root simply means that we still need to “difference” the data to achieve stationarity.
The general form of the Augmented Dickey Fuller regression, which is derived from the definition of the basic random walk equation is
Some parameters can take the value of zero if needed. For example, when all parameters are nonzero, such model is called Random Walk with Drift and Deterministic Trend. If only is zero, then such model is called Random Walk with Drift. If both and are zero, such model is then called Random Walk, which is the most parsimonious assumption as to nonstationarity.
The null hypothesis of the Augmented Dickey Fuller test is , or there is still a unit root, while the alternative hypothesis is that , or the data series is already stationary. Critical values are provided for by Dickey and Fuller and MacKinnon, since the usual tstatistic critical values do not apply to the test. However, the usual testing procedure applies. We get the computed value by dividing the parameter estimate of by its standard error, and if this computed value is less than the critical statistic, then we do not reject the null hypothesis that there is still a unit root. Otherwise, when the critical value is less than the computed statistic, then we reject the null hypothesis that there exists a unit root, and we can say that the data is already stationary.
The value of m is often determined empirically such that there exists no serial correlation on (2.1). In this paper we employ the Vector Autoregression Model specified in (7.1) and (7.2), using the Hendry topdown approach, i.e. using an initial lag of 13 since the data we have is monthly in nature, and determine the number of lags associated with the smallest information criteria (Akaike, Schwarz and the HannanQuinn information criteria).
[Insert Figure 3 and 7 here]
As we can see from Figure 3, lnCPI has a trend, which suggests that there is a possibility of nonstationarity. Figure 7 also show that lnCPI has a possible unit root, since the spikes of the correlogram exceed the confidence interval for stationarity.
From the Vector Autoregression Lag Criteria, we can see that the optimal lag to be used for the ADF test is 2, as suggested by the Schwarz and the HannanQuinn Information Criterion.
We run three forms of the ADF regression, relying on three different assumptions regarding the nonstationarity of lncpi, the random walk, random walk with drift and random walk with drift and with deterministic trend. To check which of the three regressions to interpret, we employ a variation of the Wald's test, testing the three regressions two at a time, and selecting the regression which was selected the most times by the test.
Under this test, we test whether the restrictions are valid versus the restrictions are not valid. Restrictions are simple those parameters set or assumed to have a value of zero. Table 5 shows that the Random Walk is the best suited model to represent lncpi. We compare the Critical value to the following statistic:
This statistic follows the DickeyFuller distribution with m degrees of freedom in the numerator, and degrees of freedom in the denominator.
[Insert Table 4 and 5 here]
Critical Values 

ADF Stat. 
1% 
5% 
10% 

Random Walk 
3.186774 
2.5833 
1.9427 
1.6171 
From Table 4, we can see that the computed ADF statistic of the Random Walk specification is 3.187, which is way larger than the 10% critical value of 1.6171, suggesting that lncpi has a unit root.
We continue testing lncpi for possible unit roots, but this time, we test its first difference.
[Insert Figure 4 and 8 here]
By visual inspection, we can see that D(lncpi) is perhaps already stationary. We employ the ADF test again, relying on the three assumptions, and then performing the Ftest three times to arrive at the best representation of D(lncpi).
[Insert Table 6 and 7 here]
Critical Values 

ADF Stat. 
1% 
5% 
10% 

Random Walk with Drift 
4.326017 
3.4875 
2.8863 
2.5798 
Table 7 shows that the proper model to interpret is the random walk with drift. From table 6, we can see that the computed ADF statistic is 4.326017, much less than the 1% critical value of 3.4875. Thus, we can say that D(lncpi) is already stationary, and LnCPI is integrated at order 1, which means it has a single unit root.
[Insert Figure 5 and 9 here]
We employ the same procedures done to check the stationarity of LnCPI in checking whether lnImports is stationary or not. From Figure 5 and 9, we can say that the variable is nonstationary by visual inspection. We again employ the ADF test on the three assumptions, check which is the better fit and conclude on the stationarity of LnImports.
[Insert Table 8 and 9 here]
Critical Values 

ADF Stat. 
1% 
5% 
10% 

Random Walk with Drift and Trend 
3.426305 
4.0387 
3.4484 
3.1491 
From Table 9, the correct model specification is the random walk with drift and trend. From table 8, the computed ADF statistic is 3.4263, bigger than the 5% critical value of 3.4484. This means that lnImports is not yet stationary, so we do the same test on the first difference.
[Insert Figure 6 and 10 here]
By visual inspection, we can somewhat say the D(lnImports) is stationary. To check, we employ the ADF test.
[Insert Table 10 and 11 here]
Critical Values 

ADF Stat. 
1% 
5% 
10% 

Random Walk 
6.597516 
2.5834 
1.9427 
1.6172 
Based on the result from table 11, the correct model specification is random walk. From table 10, the computed ADF test statistic is 6.597516, which is less than the 1% critical value of 2.5834. We then conclude that lnImports is integrated at order 1, meaning it has a single unit root.
Cointegration means that there exists a long run or equilibrium relationship between variables. (Gujarati, 2003) Two or more variables may be cointegrated even if they are not stationary, which guards us against spurious regression results.
Stochastic processes have specific properties that the definition of cointegration taps into. Specifically, if two stochastic processes and are integrated at the same order d, then a linear combination of them is integrated at some order d* where .
In symbols
with .
Two variables are now cointegrated even if they are not stationary, but a linear combination of them is stationary.
Now that we have shown that both variables lnCPI and lnImports are I(1) stochastic processes, we perform cointegration tests to know whether the two variables are cointegrated or not, i.e., if there exists a linear combination of them that is integrated at order 0. This is done to prevent spurious regression results, i.e. very fit regression model, but in reality there is really no relationship between the two variables.
The GrangerNewbold rule of thumb states that there is a possibility of a spurious regression if the computed statistic is greater than the computed Durbin Watson d statistic. From table 12, it can be seen that this is the case, so we have to proceed using formal tests. Note that the GrangerNewbold rule of thumb does not conclude about the lack or existence of cointegration, only a possibility.
The Cointegrating Regression Durbin Watson (CRDW) test is a quick way to check the existence of cointegration between two variables. (Gujarati, 2003)
Since , where is the coefficient of the lagged value of Y is the ADF test, if , then it can be said that
The null hypothesis of the test is that (there is cointegration), versus the alternative hypothesis that (the regression is spurious). In this test, we compare the computed Durbin Watson dstatistic from the critical values provided for by Sargan and Bhargava.
Significance Level 
Critical region 
1% 
d < 0.511 
5% 
d < 0.386 
10% 
d < 0.322 
Since the computed dstatistic in Table 12 is 0.487253, we do not reject the null hypothesis that there is cointegration between the variables.
The Augmented Engel Granger test taps into how cointegration is defined. The AEG test accepts that the data is stationary if the residual of a linear combination of them is stationary.
To employ the test, first, we perform our OLS regression on Equation (1.5), and test for the existence of unit roots of the residual via employing the Augmented Dickey Fuller tests. In our case, we performed a regression on (1.5), but adding 11 dummy variables representing the months of January to November, to account for the seasonality of the data series. Only 11 variables were used to prevent perfect multicollinearity or the dummy variable trap.
[Insert Figure 11 and 12 here]
Via observation, we can hunch that the residual of the regression using equation (1.5) is already stationary, and we test our claim via the use of the ADF tests performed thrice, one per assumption of the nonstationarity of the data series.
[Insert Table 14 and 15 here]
Critical Values 

ADF Stat. 
1% 
5% 
10% 

Random Walk with Drift 
2.606209 
2.5831 
1.9427 
1.6171 
From Table 15, we can see that the residual follows a random walk. And from table 14, the ADF statistic computed is 2.6062, lower than the 1% critical value of 2.5831, which means that the residual is stationary.
Via the definition of cointegration, since the residual is stationary, equation (1.5), or the regression with dummy variables, is now a cointegrating regression.
The Johansen method, instead of relying on the Least Squares assumption, uses maximum likelihood to estimate the cointegrating equation. (Sorensen, 2005) Two values are derived from this test, the Trace Statistic and the Maximum Eigenvalue statistic. We employ the Johansen Cointegration test using 2 lags of the variables to be consistent with the VAR lag length criteria provided earlier.
[Insert Table 16 here]
From Table 16, we can see that regardless of the assumption made as to the intercept or trend, there exists at least one cointegrating vector of lncpi and lnimports. This reinforces the results of the previous tests that lncpi and lnimports are cointegrated.
While there exists a long run equilibrium relationship between two variables, the cause and effect relationship remains unclear. In our case, is it that changes in consumer price index causes changes in imports, or is it that the changes in imports causes changes in consumer price index? We employ the Granger Causality Test to verify or justify our choice of dependent and independent variable.
The Granger Causality test provides us with the existence of “predictive causality,” i.e. past values of one variable contains information useful in predicting another variable. (Gujarati, 2003) The Granger Causality test allows us to know which variable causes which, using the information provided by the variables themselves, together with some choice of lag lengths.
Specifically, we first perform the following regression models:
Equation (3.1) says that lnImports is influenced by the past values of lnImports and lnCPI, while equation (3.2) says that lnCPI is influenced by the past values of lnImports and lnCPI. and are assumed to be uncorrelated. Note that the results provided by the Granger Causality test varies with the choice of lag length, that is why we must be careful is choosing this. (Gujarati, 2003)
To justify our use of lnImports as dependent variable and lnCPI as our independent variable, we need to show that there is at least a unidirectional causality from lnCPI to lnImports, that is, the coefficients of the lagged vales of lnCPI in equation (3.1) should be statistically different from zero, and the coefficients of the lagged values of lnImports on equation (3.2) should be statistically the same as zero, or a bilateral causality, meaning that the lagged values of both variables have predictive power that can be used for the other variable, that is, all coefficients when taken together are significant.
Using n=2 lags of each variable (to be consistent with the VAR lag length criteria used), we employ the Granger causality test in Eviews and the result is shown in Table 17.
[Insert Table 17 here]
Since the pvalue of the first test that the coefficients of the lagged values of lnCPI to be equal to zero is 0.02981, and the pvalue of the second test that the coefficients of the lagged values of lnImports to be equal to zero is 0.42155, we reject the null hypothesis that lnCPI does not Grangercause lnImports, and do not reject the other null hypothesis that lnImports does not Grangercause lnCPI. Thus, our choice of variables are justified, i.e. there is only a unidirectional causality from lnCPI to lnImports at lag length 2.
Having proven that the data does not violate stationarity and does not produce spurious regression results, we proceed to estimate the effect of domestic CPI on import spending in the Philippines.
The first regression in Table 18 shows the result of OLS regression of equation (1.5) without taking into account differences in import levels due to seasonality. Specifically, the results are
Before interpreting the said model, we first detect for CLRM violations, and correct those which are detected. Table 19 shows the detection and correction steps. The final OLS model, free from autocorrelation, is as follows:
Regression results show that, for every 1% increase in the Philippine consumer price index, import spending will increase by 1.079031%. This value is close to 1. In fact, via a test done and shown on the latter part of Table 19, we cannot reject the null hypothesis that this value is equal to 1. Therefore, in the long run, we can say that import spending is unitary elastic to domestic prices.
We know that imports in the Philippines are somewhat seasonal, since Filipino businesses tend to prepare for the holiday season through massive importation of products. Also, some products are imported on a seasonal basis, i.e. agricultural products, since some of them are only available in certain times of the year. Thus, we modify equation (1.5) to take into account of differences in import levels per month of the country.
We introduce 11 dummy variables to equation (1.5) to represent the months January to December. The choice of the number of dummy variables is to prevent the dummy variable trap. Specifically, equation (1.5) now becomes
In this formulation, we account for the differences in imports levels per month using differential intercepts , with the month of December as the base year. Regression results are shown in Table 18 and are summarized here.
As shown in Table 18, equation (4.4) is better than (4.1) since it takes into account differences in import spending per month, and is supported by a statistical test.
To correct for CLRM violations, we use results in Table 20 and summarize them here.
As we can see, for every 1% increase in the consumer price index, import spending increases by 1.05484% in the long run. This value is very significant, and is statistically equal to one. (Table 20) This means that regardless what month we are in, import spending is unitary elastic to consumer price index, or domestic prices.
However, now we can see the seasonal autonomous import level differences across months. Imports of the month February is significantly lower than that of December, while imports of the months March, April, June, July, August, September, October and November are significantly higher than that of December. Meanwhile, January and May imports are statistically similar to December imports. This is perhaps due to the fact that during March and April, people are importing more products which are either school related items (for grade school and high school students), agricultural products which are seasoned to be harvested during these months, or maybe excess energy sources since these are usually the start of summer months of every year. The months of September to November are usually in preparation for the holidays, so businessmen stock their items in advance, while process are relatively cheaper.
Import spending do not instantaneously react to changes in domestic prices. Sometimes, it takes a few periods before the effect of changes in CPI be felt on the import spending levels. Also, current changes in CPI might also affect future decision of importation. Employing a dynamic model would allow us to study these lagging effects. However, one basic estimation problem is to determine the length of lag, i.e. the length of time that a policy variable affects our dependent variable. Sometimes, this is an empirical question. (Gujarati, 2003)
The general form of a dynamic model is as follows
where p is the number of lagged independent variables, and q is the number of the lagged dependent variables on the right hand side of the equation. Equation (5.1) is also known as the autoregressive distributed lag model, or ARDL(p,q). We refrain from using the 11 dummy variables because their inclusion might result to a model that is expensive in terms of consuming degrees of freedom.
More specific forms when p=0, in which (5.1) is known as the autoregressive regression model or AR(q), and when q=0, in which (5.1) is known as the distributed lag model or DL(p).
In this model, we only consider a distributed lag DL(p) model, i.e.
To solve the problem of determining the optimal lag p, Alt and Tinbergen suggests that we use the bottomup approach to determining the optimal lag. We do this by introducing lags of the independent variable one at a time per regression, and check the individual significance of this variable and the overall significance of the regression model. We stop when either the coefficient of the added variable becomes insignificant or becomes counterintuitive. (Gujarati, 2003)
We employ regression model (5.2) using the AltTinbergen approach and come up with a distributed lag DL(1) model shown in Table 21. Corrected results are as follows:
We stop at the first lag since the sign has already become counterintuitive. These results show that as the current CPI increases by 1%, current imports will increase by 5.60%, ceteris paribus, and future imports would decrease by 4.539%, ceteris paribus. From Table 21, the sum of the two coefficients is statistically the same as 1, suggesting that the unitary elastic property was spread over two periods.
A closer look would allow us to interpret the coefficients in a way that it is not that counterintuitive. Given that incomes are fixed in this period, if current prices increase, people would just purchase the cheaper good(in this case, those which are imported) and use current purchase as substitute for next period purchase. In this way, current imports increase and next period imports decrease, since current and future consumption are substitutes in light of intertemporal economics.
Rather than assuming that the total effect of a certain variable is spread at only a finite time interval, Koyck assumes that policy variable changes in the far past also affect current values of our dependent variables, and this effect goes smaller and smaller geometrically as one further progresses into the past. This is the essence of the Koyck regression. Rather than having a problem for the value of p, it is just assumed that the model (5.2) has a p which goes infinitely large.
Since it is plausible that the more recent phenomenon has a larger impact that events which happened in the distant past, Koyck assumed that the effect of one policy declines geometrically for every next period. With this, estimation of the infinite DL would not be a problem anymore since we can employ mathematical tricks to simplify our regression model.
For Koyck, coefficients decline geometrically. This is shown by the following
Mathematical manipulation using equation (5.4) converts the infinite DL model into an ARDL(0,1) form like that of equation (5.5)
Equation (5.5) can be estimated by using Ordinary Least Squares if there is no endogeneity bias, or employing Two Stage Least Squares if such was the case.
a. Ordinary Least Squares
Estimating (5.5) using our data would result to the results shown in Table 22. A summary is shown below:
As this model is free from CLRM violations as shown in Table 22, including the absence of endogeneity bias, we can interpret it immediately. For every 1% increase in the current CPI, imports spending would increase by 0.3077%, and for every increase in the current (previous) import spending, future (current) import spending would increase by 0.6875%.
Some properties of the Koyck model in (5.6) are as follows:
The long run equation using (5.5) and (5.6) would be
However, we cannot check for the significance of the coefficients of (5.10) due to the nonlinear transformation employed from (5.5) to (5.9). Notice that the long run elasticity using Koyck is close to one, which supports our OLS regression earlier, although the significance is dubious.
b. Two Stage Least Squares
While the previous OLS regression did not suffer from the endogeneity bias (see Table 22), one estimation problem of the Koyck model is the possible correlation of the lagged Y term and the residual. To circumvent this, Liviathan suggested using instrumental variables. (Gujarati, 2003)
Table 23 and Table 24 show the regression results using instrumental variables, with and as instruments, respectively. Both regressions suffer from severe multicollinearity due to the use of the lagged independent variables, which are correlated with the independent variable found in the original Koyck model (5.5). However, the instruments used are justified via the SARG test. (Tables 23 and 24)
Results shown in Table 23, however, is not in accordance to the assumption of the Koyck model that . Thus, we need not interpret it.
The second model, however, can be summarized as follows:
(5.11)
For every 1% increase in the current CPI, imports spending would increase by 0.077%, and for every 1% increase in the current (previous) level of imports, future (current) imports would increase by 0.8784%.
The properties of the Koyck model in (5.11) are as follows:
The long run equation using (5.11) would be
Significance of (5.12) is doubtful since no standard errors are computed, however, by visual inspection, we can see that (5.10) and (5.12) are very different.
In this regression model specification, the elasticity coefficients are not assumed to decline geometrically. Instead, these coefficients are assumed to fit a polynomial of degree m, i.e.
Transforming (5.2) using (5.13) would result to
Equation (5.14) is estimated using OLS if there are no CLRM violations, then the are retrieved using equation (5.13). Standard errors can also be retrieved from the standard errors of (5.14), thereby allowing us to make inferences.
Employing the regression model, we first determine p, which is the optimal lag. We use the Information Criteria in Table 25 to know what p to use given m=2 and m=3, which are most commonly used degrees of the polynomial.
[Insert Table 25 here]
From table 25, we can see than the optimal lag is 9 when the degree of polynomial is 2, and the optimal lag is 12 when the degree of polynomial is 3.
We implement both regressions, and results are shown in tables 26 and 27.
The summary of the regressions are as follows:
Interpretations follow the usual way. One interesting thing to note is that the sum of lags is statistically the same from 1 in both regressions, suggesting that both regression simply state the lag structure of the effect of changes in lnCPI to lnImports, but the total effect remains the same.
While lnCPI and lnImports are cointegrated in the long run, it is plausible to say that in the short run, there exists disequilibrium. It is rather naïve to say that if two variables have long run relationship, then this relationship will hold every single time. There are multitudes of economic shocks that could happen that will drive the short run values to be different from the long run equilibrium values. However, cointegration allows that regardless of short run disequilibrium over the course of life of a policy, all variables will slowly return to their long run equilibrium relationship. This is the theoretical perspective behind the Error Correction Mechanism. (Gujarati, 2003)
Specifically, the error correction model can be specified as
(6.1)
where u is the estimated residual from equation (4.2). This model shows that, when lnCPI changes by 1% in the short run, lnImports changes by also in the short run, plus some factor of the deviation from the equilibrium value of lnImports. can be interpreted as the speed how quickly equilibrium is restored. Apriori, should be negative, since if the disequilibrium is found to be higher than the equilibrium value, the change next period should be lower in order to reduce the disequilibrium.
To estimate this model, we run a regression of (6.1) twice, one with the constant , and another without. We tested which model is better using the standard F test, and concluded that the one without a constant is better (see Table 28). The summary of the corrected model is as follows
The value 3.053278 can be interpreted as the short run CPIelasticity of imports, i.e. in the short run, for every 1% change that occurred in CPI, import spending would change by about 3.05%, if the past import spending value were in equilibrium. Disequilibrium is corrected 0.336 of a month, or about 11 days. This very fast adjustment mechanism is crucial for policy making since it makes inflation targeting very effective in controlling the amount of imports if the cointegrating relation (4.2) were true.
The Vector Autoregression Model is a tool frequently used in forecasting since it is based on atheoretic expectations. It lets the data “speak for itself.” This simultaneous regression model treats all variables as endogenous, and regresses them bases on the past values of all endogenous variables in the system.
The VAR specifies the following systems of equations
Using the VAR Lag Length criteria employed in Table 3, we use 2 as lag length.
OLS Estimates of (7.1) and (7.2) are as follows:
Since there is no autocorrelation, we can interpret (7.3) and (7.4) the usual way we interpret OLS estimates. From (7.3), we can see that the lagged values of lnImports do not affect the current values of lnCPI, a result of the Granger Causality test. For every 1% change in CPI today, CPI next month would increase by 1.5897%, and CPI 2 months from now would decrease by 0.5935%. From (7.4), the only individually significant value that affects lnimports today is the value last month, i.e. for every 1% increase in the value of imports today, next month's imports would increase by 0.6232%.
The Philippines has been experiencing negative trade balance for almost two decades now. With the results of this paper, it has been proven that domestic prices  represented by consumer price index without doubt affect import volumes. Because of this, the Bangko Sentral ng Pilipinas (BSP) has adopted inflation targeting as a tool to ensure price stability that is favourable to a balanced and sustainable growth of the Philippine economy. The purpose of the move of BSP from traditional money supply manipulation to inflation targeting has now been justified by this paper. BSP should continue to control inflation and prepare implementing monetary policy tools if in case actual inflation deviates too much from targeted inflation levels.
An alarming finding of the researchers was seen when the Error Correction Model was estimated. The result showed that it needs only 34% of a period to pass for import levels to adjust to consumer price index changes. For a monthly dataset, this means that only 11 days are needed for import volumes to adjust to changes in inflation level; hence, this tells us that local firms, SME's and transnational entities, are consciously monitoring domestic price levels and are quick to adjust to their purchase or order patterns. This further supports the need for BSP to be accurate in their inflation targeting and to be careful in the use of monetary policy tools.
After the series of dynamic model specifications and estimations, we have conclusively proven that consumer price index positively affect import spending. Furthermore, the degree of elasticity varies with lag specification, but the long run impact remains to be statistically similar regardless of the model specification. In summary, nominal import spending is unitary elastic to the consumer price index.
Given the result of the study, BSP should certainly do its best to control inflation and maintain it at the most desirable or lowest possible level. Furthermore, the government should implement complementary policies that could offset any possible effect of increases in import spending brought about by rise in domestic prices.
For further researches that will be conducted in this field, a bigger monthly dataset that includes the years 1997 and 1998, where the Asian financial crisis started, will make the results of the study even more reliable and could possibly have more interesting results than that of this paper. Also, using a dataset of import spending in real terms could potentially improve the analysis of this relationship, however, a tradeoff between using real import levels to show the volume of output imported but disregarding the price effects of import spending, and using the nominal spending volume, should be considered. Lastly, since inflation and import spending have seasonal characteristics, it will be better if the seasonality components of the variables will also be tested and estimated. This will enhance the results of the regression and make the analysis even more useful.
Source: Essay UK  http://doghouse.net/freeessays/economics/importspending.php
This Economics essay was submitted to us by a student in order to help you with your studies.
This page has approximately words.
If you use part of this page in your own work, you need to provide a citation, as follows:
Essay UK, Import spending.. Available from: <http://doghouse.net/freeessays/economics/importspending.php> [210219].
If you are the original author of this content and no longer wish to have it published on our website then please click on the link below to request removal: