Размещено на http://allbest.ru

Asia-Pacific region: empirical analysis of the effects

Introduction

trade relation econometric

Last decades can be characterized as the period of active international economic integration that covers almost all countries in the world. A complex of political, economic, geographical characteristics of countries determines intensity of this process in particular directions. Awareness of the extent of significance of these factors is crucial when defining international economic policy priorities of countries.

Economic interests of many large economies are now concentrated in Asia-Pacific region due to its fast economic growth and intensive technological development in strategically important spheres. This attractive features of countries belonging to this region makes other countries to interact with them more intensively in order to enhance their own economic growth and intensify technology exchange occurring in the process of international trade for achieving the higher level of technological development.

Russian economy of XX century can be characterized as autarky based on almost complete self-sufficiency principle under control of government. It means that the dependence on the exchange with other countries was minimized. This closed economic regime negatively influenced trade and investment international flows between Russia (in this specified time - the Soviet Union of Socialist Republics) and capitalistic unit of countries. In 1980s the idea of economy based on the government plan discredited itself and the Soviet economy took a step forward transformation into market economy: privatization of property and market modernization began. After this and even more after the collapse of the Soviet Union the process of intensive economic integration of Russia in the world economy took place.

Previously economic relations between Russian Federation and Asia-Pacific region were passive. However, increasing power of countries of this region, their active economic growth and technological development, geographic position of Russia regarding Asian countries and some other factors contributed to change of this situation. For realization of national economic interests of Russia it is crucial to raise the level of its international cooperation with countries of Asia-Pacific region and to form clear foreign economic policy in this direction.

Some steps were taken forward increase of economic integration of Russia in Asia-Pacific region. One of the most significant among them is the fact that since 1988 Russia is the participant of becoming increasingly significant APEC (Asia-Pacific Economic Cooperation), the entry into which brought economic benefits through extending international trade geography of Russia and trade familiarization with such trade destinations as Singapore, Hong-Kong, Thailand and other previously strange for RF trade destinations. The participation in APEC led to liberalization of international trade and investment activities and strengthening of trade ties due to reducing barriers between countries-participants of the forum. Nowadays, Russia is a powerful member of APEC: it intensively takes part in the process of modification of APEC and has a significant weight in discussing strategically important issues. Furthermore, the indicators of active integration of Russian Federation in APEC were such events as Asian economic forum in 2004, where perspectives of investment and energy cooperation in APEC region were discussed, and the summit APEC in Vladivostok in 2012.

Besides membership in APEC Russian Federation economically integrates in Asia-Pacific region through participation in other international organizations: Shanghai Cooperation Organization (since 1996), BRICS (since 2006), Association of South East Asian Nations (ASEAN, since 1996) and - and cooperation with other economic blocks.

The shift of international activities of Russian Federation to Asia-Pacific region is a weighty argument to investigate economic integration there. It is crucial in understanding of their trade patterns, which in turn is helpful for determining of main directions of foreign economic policies and predicting of perspectives of further economic cooperation with this region.

Besides analysis of the influence of participation in economic trade agreements of countries on export-import trade flows, geographic factors are taken into consideration with use of gravity model approach.

The process of economic integration and priorities in international trade are determined by a complex of characteristics of countries. Geographic group of features, which is a focus of current study, indicate transportation costs of international trade, which are simultaneously the reason and the consequence of international economic integration.

Geographic group of characteristics includes area, distances, common border, access to seas, common language, colonial links and others. The reason of choice of these characteristics of countries as the focus of the research is that they do not only indicate transportation costs, but they are crucial when investigating trade patterns of Russian Federation. International trade of this country with other countries is highly dependent on geographic peculiarities due to a great length of Russian borders, its vast area, a large number of neighboring countries and other factors.

One of the most frequently used methods of investigation of influence of geographic factors on international trade flows is gravity approach. Initially it was used by J. Tinbergen (1962) as the way of determining the character of interconnection between trade values and economic size of trading countries and distances between them on the base of Newton's gravity law. Economic sizes are determined by values of gross domestic product. They can be described as economic weight of country in trade pattern.

For the purposes of the research data on international trade flows and geographic characteristics of 29 countries of Asia-Pacific region was downloaded and consolidated into one database. The choice of countries is grounded on export and import values: countries with no or low trade flows were excluded from observation, because without them there is nothing to estimate.

Empirical part of the work contains estimation of data described above with use of one of the most common approaches to gravity model analysis - Poisson Pseudo Maximum Likelihood, that is considered to be suitable by majority of empirical works investigating international trade flows (Silva and Tenreyro, 2006; Melitz and Toubal, 2014) The appropriate method should contribute to avoidance of two main challenges that usually occur when analyzing international trade flows with appliance of gravity. The first problem that leads to biased estimations is heteroscedasticity, which occurs because data downloaded from different sources has different quality. Another challenge is sample selection, which takes place because of excluding of zero flows, that often occur in trade data, from observation. Research conducted by Silva and Tenreyro (2006) proves the fact that estimation of log-linear gravity regressions in multiplicative form in presence of heteroscedasticity and sample selection problems leads to distorted results.

All things considered, the major purpose of current research is to estimate the effects of international economic integration of Russia in Asia Pacific region through explanation of differences and changes in export-import trade flows with use of gravity model approach. One more side purpose is to choose the most appropriate method of applying gravity model to explanation of international trade flows.

Main hypothesis of the work can be formulated as following:

1 Geographic characteristics are significant factors influencing export-import flows between countries in Asia-Pacific region.

2 Export-import trade flows are the inverse function of distance between trading partners.

3 Export-import trade flows are proportional to economic sizes of trading countries that are expressed in gross domestic product values.

4 Remoteness from other countries in the world positively influences export-import trade flows.

5 Poisson Pseudo-Maximum Likelihood is an appropriate method of gravity model estimation.

The rest of the paper is organized in the following way: Section 2 provides the existing, mainly empirical literature concerning observed issue, Section 3 contains description of data and variables, its empirical analysis with help of OLS and PPML and estimation results, Section 4 concludes.

1.Literature review

where is dependent variable, - coefficients of explanatory variables, - explanatory variables, - the error term.

The gist of the method is minimizing the sum of differences between predicted and observable values. Thus, the goal of the method is to minimize the sum of squared and to define the equation, characterizing linear dependence:

Log-linear OLS regression equation looks as follows:

The method suffers from considerable problems that makes it inappropriate in majority of cases, especially in analysis of macroeconomic data, such as in our paper. Trade data usually satisfies only the one data requirement - macroeconomic studies usually employ enormous datasets. However, the amount of missing, extremely high or extremely low values is also large.

Application of OLS method to gravity model gives biased results. Firstly, it cannot cope with heteroscedasticity problem, which occurs due to significant up or down deviation of variable value from the average value. Secondly, this method obviously cannot deal with problem of zero flows either as it automatically excludes zeros because the logarithm of zero is undefined. Even if all trade values are positive, the error value is correlated with covariates, that leads to inconsistence of estimations.

2.1.2 PPML - Poisson Pseudo Maximum Likelihood Estimator

The failure of OLS estimator to solve main gravity challenges makes us to choose another estimator. The most frequently used in empirical literature estimator that is used for analysis of our models is Poisson Pseudo Maximum Likelihood estimator (PPML). The adequacy of it for gravity-based models was explained by Silva and Tenreyro (2006). They stated that method gives consistent estimations of gravity model regressions unlike OLS estimator under only one assumption - correctly chosen variables for model specification.

The Poisson regression estimates expected values according with following:

According to Silva and Tenreyro (2006) taking into account fulfillment of Jensen's inequality of logarithm of expected value of variable to expected value of its logarithm (E[ln ln(E[)) log-linear estimator gives biased results when heteroscedasticity of errors takes place.

In current paper we apply the PPML with population averaged effects as in the paper by Cameron and Trivedi (2013). The model averages individual effects, and the main condition of results consistence is the same as in the pooled OLS:

There are three main properties of PPML estimator that makes it suitable for estimation of gravity models. Firstly, it should be noted that Poisson distribution of data is not necessary for PPML method application, so it can be used in analysis of not only count data, but also of more extended nonlinear models with use of dummies such as gravity model. Secondly, as it employs the logarithm of the error it includes variance, thus it is able to cope with its heteroscedasticity, which makes it impossible to obtain consistent unbiased results due to correlation of error term with explanatory variables. Thirdly, it naturally include zero values, which are automatically dropped from the process of estimation by OLS method because logarithm of zero cannot be counted. However, trade data abounds in zero values because not all countries are willing to export some goods to all countries. Third property was stated by Khosla (2014): PPML coefficients of explanatory variables can be interpreted as semi-elasticities, that makes interpretation as simple as in the case of OLS.

2.2 Data description

2.2.1 Variables description and expected signs

In order to achieve the aim of the paper empirical analysis of panel dataset, compiled from several gravity and trade data sources, was conducted. Data was collected for 29 countries of Asia-Pacific region listed in Appendix A between 1996 and 2012.

The list of variables, their short descriptions, measure values and sources of data on them are presented in Appendix B. There are six sources, employed for compiling our dataset: Trade Statistics Database (UN Comtrade), World Integrated Trade Solution (WITS) website, the World Bank website, Research and Expertise on the World Economy website CEPII, Kellogg Institute for International Studies exploring democracy and human development website and UN Department of Economic and Social Affairs website.

Countries are organized in pairs (every country of the region with every country of the region) and their own characteristics and geographical bonds between them are included for every year in the observed period.

In models analyzed in further sections logarithms of export and import values are used as dependent variables. They are main indicators of international trade intensity between countries in pairs.

In the first interpretation of gravity model by Tinbergen (1962) only variables, expressing economic sizes of trading countries and circle distance between them, were employed for explanation of trade patterns between countries. We also use this variables and make an accent on their significance and interpretation in the process of result estimation. Economic sizes of trading countries are expressed in Gross Domestic Product values, the indicator of cost of final goods and services that are produced on the territory of the country in observed year. This measure was brought for both countries in pairs. Tinbergen (1962) stated that trade flows positively depend on economic sizes of both countries, because higher GDP value of country-reporter means greater ability to export goods and services abroad whereas GDP value of country-partner is the indicator of value of demand for these exported goods. Thus, empirical analysis conducted in further sections is expected to reveal positive signs of coefficients for both of these variables.

Another explanatory variable that is of great importance is distance between countries in pairs. Hummels (2001) stated that distance between destinations is a proxy of costs on transportation of goods and services from country-exporter to country-importer. In the CEPII dataset four approaches to estimation distance were employed, however we took only two of them as different specifications of models: geodesic distance between capitals and weighted by population distance. According to Tinbergen (1962) again, trade flows should negatively depend on distance. Thus, we expect that regardless the way of distance calculation, the estimations of this variable will give negative sign.

For calculation of geodesic distance between capitals the great circle formula is used:

)

where R=6371 - radius of the Earth, - latitudes of capitals of countries i and j respectively, - longitudes of cities in countries i and j respectively. Latitudes and longitudes of cities should be in radians, so that if they are in degrees, they should be multiplied by 57.3.

However, there are some problems with use of distance calculated in such way. Firstly, the North Pole is not taken into consideration, because there are no any trade roots. Furthermore, if area of countries is large enough calculation of distances only between capitals can lead to wrong results. Moreover, according to Head (2003) costs of transportation include also cost of loading goods in the transport and other activities that do not depend on distance between destinations.

The approach of weighted by population distances uses share of population in capitals in the overall population of each country in the pair. The formula for weighted distance calculation introduced by Head and Mayer (2002) is as follows:

where - weighted distance between countries i and j, - population of city k in the country I, - overall population of the country i, - population of city l in the country j, - overall population of the country j, - sensitivity of bilateral trade flows to distance between cities k and l in countries i and j respectively (). is considered to be equal 1.

It is crucial for those who investigate trade patterns to estimate distance between destinations for several reasons. Firstly, as it was said above, it indicates the amount of possible transportation costs. Secondly, more distant countries might have greater cultural differences, that makes an obstacle for international trade. Thirdly, it is more problematically to get information about markets of distant countries. One more explanatory variable that is often used in contemporary empirical literature is remoteness, that was counted with use of formula suggested by Head (2003):

where - distance between countries i and j, - share of GDP value of country-partner of country i in the world GDP.

According to formula above remoteness is distance weighted by GDP. So it may be considered as one more approach to distance calculation. Initially, another formula of remoteness calculation was suggest by Helliwell (1998):

However, this formula contains a significant mistake: if distance is high and the relative economic size of country is small, the remoteness can overestimated. So, the formula by Head (2003) is a more appropriate way of distance estimation. Remoteness was counted for both types of distances described above. The sign of this variable is expected to be positive according to statement of Silva and Tenreyro (2006) that trade flows between countries are larger for countries that have larger distances to all other countries observed, reflected by the remoteness variable.

Besides distance, the analysis employed other variables showing geographic bonds of countries in the pair are:

1. Contiguity or existence of common border - dummy, that is equal 1 if countries in the pair share common border and 0 if they do not. The expected sign for this variable is plus as transaction costs of transportation goods to neighboring countries are lower than to distant countries. Furthermore, it is more possible, that these countries are culturally similar, that increases the willingness to export there.

2. Common official language - dummy, that is equal 1 if population in both countries speaks common official language and 0 if they do not. It shows cultural identity of countries and facilitates better interaction due to ability to communicate using common language. Consequently, it is expected that sign will be positive.

3. Dummies for being the part of the same continent: dummy for being a part of America, dummy for being a part of Asia and dummy for being a part of Pacific. Countries are more often trade with countries belonging to the same continent due to its lower remoteness. The sign of this variables is expected to be positive, however there can be exceptions in some specifications.

4. Dummy for being part of one country in the past or now. It equals one if countries in the pair have ever been united by one country and 0 if not. The expected sign is positive as the variable also indicates cultural bonds of countries. Moreover, it is more possible that population speaks common language in such countries.

5. Colonial dummy variable equals 1 if countries have ever had a common country-colonizer and equal 0 if not. The same justification as for the previous variable is acceptable. The sign is expected to be positive.

Another block of geographic variables includes:

1. Area of country-partner - the inclusion of this variable can be justified also by increase of transportation costs with greater area of country-partner as the city-destination may be distant from the border. Furthermore, Silva and Tenreyro (2006) stated that smaller countries are usually more open to international trade flows. Sign of the variable is expected to be negative for this reason.

2. Landlockedness of both countries - dummies for country-partner and for country-reporter that equal 1 if there is no access to seas and 0 if the country is not landlocked. The expected sign is negative, as absence of access to seas increases transportation costs. The only one country -Mongolia - is landlocked among those that we included into our analysis. So, we can investigate influence of its landlockedness on trade of Asia-Pacific countries with it.

Geographic variables are crucial when analyzing international trade flows. Distance, common language, common border, access to seas and others are indicators of transaction costs for movement of goods and services. These variables explain why some countries trade with these particular countries. It is expected that weighted distances better fit our regression analysis due to the fact that in the list of countries some have a considerable area and estimating distance between only capitals or the most populated cities can give biased results.

The last block of variables reflects existence of agreements between countries in pairs. The dataset for agreements was downloaded from Kellogg Institute for International Studies exploring democracy and human development website and initially included types of agreements: 1 - preferential trade agreement, 2 - free trade agreement, 3 - customs union, 4 - common market, 5 - economic union. However, among observed in our research countries there is no any pair that had 4^th and 5^th and the 3^rd one have only 3% of country pairs, so it is not significantly influence integration in the region. For 1 and 2 types dummy variables, that equal 1 if countries have this type of agreement and 0 if they do not, were created. Participation in agreements liberalizes international trade by lowering tariffs for countries-members, consequently signs of coefficients are positive.

2.2.2 Data heteroscedasticity and descriptive statistics

Firstly, it should be decided whether the problem of heteroscedasticity of trade variables exists in order to prove that OLS method is not appropriate estimator for gravity model. For this purpose, Breusch-Pagan and White's tests were conducted. The null hypothesis for both of them is that variances are constant, that means that there is no heteroscedasticity problem. Results of tests are presented in tables in Appendix C.

The null hypothesis about homoscedasticity of trade variables is rejected since chi-square statistics has p-value less than even 1%. It means that problem of heteroscedasticity of trade variables exists. Thus, OLS estimations are inefficient and standard errors are biased. This can lead to erroneous interpretation of regression results. That is the reason for choice of another method of estimation gravity model, which is able to cope with problems associated with use of OLS.

Table with descriptive statistics of continuous variables is presented below.

Some crucial points can be revealed from this table. Firstly, the number of observations for import and export is less for export and import values: we have some missing variables. Secondly, standard deviation form average value is significantly high and there can be outliers both to up and down direction. Thirdly, minimum values of both variables are low. These 3 points show that our data suffers from the problem of zero values, which was described above. Moreover, according to areas minimum and maximum values, both extremely large countries (like Russia) and extremely small (like Singapore).

Descriptive statistics for binary variables is presented in appendix D. We can observe that the majority of countries included into our research is situated in Asia, and the situation when both countries in the pair are situated is most frequent for Asia too. Only 5.7% of countries have common borders and 17% of pairs of countries have common official language. There are 14.53% pairs of countries that have free trade agreements and 16.02% that have preferential trade agreements.

2.3 Models and estimation results

2.3.1 Ordinary least squares estimations

The first step of our empirical analysis is to estimate dataset employing method of OLS as it was suggested in the first work applying gravity law to explanation of trade flows between countries by Tinbergen (1962). The set of variables is the same in his work, though one more specification was added - distance weighted by population is included into the set of variables estimated.

Equation estimated is the following:

where - the trade flow from country i to country j (export or import), , - gross domestic products of countries i and j respectively, - distance between countries i and j (geodesic between capitals or weighted by population).

Table below presents results of estimation. (1) and (3) columns present results for specification with geodesic distances between capitals and (2) and (4) columns are results for specification with weighted by population distances.

Standard errors in parentheses

^* p < 0.05, ^** p < 0.01, ^*** p < 0.001

All coefficients are significant. Our results coincide with conclusions, made by Tinbergen: signs of coefficients for GDP values of trading countries are positive and signs for distances are negative. That means that international trade flows are proportional to economic sizes of trading countries and are an inverse function of distance between country-reporter and country-partner.

Now more variables are included into regression, particularly we added remoteness variable dummies for common official language, contiguity, colonial ties, free trade agreements, preferential trade agreements, location in common continent America and location in common continent Asia. Furthermore, two different types of remoteness of country-reporter from other countries in the world are employed: calculated with use of geodesic distances and calculated with use of weighted by population distances.

Table for results is represented below. Columns (1) and (2) show specification with use weighted distances and (3) and (4) column represent specification with use of geodesic distance between capitals of countries in pairs.

Standard errors in parentheses

· p < 0.05, ^** p < 0.01, ^*** p < 0.001

According to r2 and significance of coefficients the 1^st specification with weighted distances suits better for the purposes of the research.

Signs of all variables that are significant except common continent dummies coincide with expected signs. We have only one insignificant coefficient for variable reflecting previous and current colonial ties between countries (having the same colonizer) that may be connected with the fact that the low quantity of pairs of countries in this region that have ever had colonial bonds. Furthermore, cities like Hong-Kong (a part of China), which have large trade turnover, were included in list of countries. That may bias the estimated parameter.

Application of OLS to our data revealed the same result for main variables as in Tinbergen's work: trade flows are proportional to economic sizes of countries and negatively depend on distance between them. As it was justified by Silva and Tenreyro (2006) the larger remoteness ratio from other countries in the world the greater export-import flow in observed pair of countries. So, the sign coincides with their conclusion.

As it was stated before our dataset doesn't satisfy all limitations of OLS method. And application of this estimator gives wrong interpretations of results. For example, GDP of both partner and reporter in all described researches are close to 1. Coefficients in the table prove this peculiarity of OLS estimator, that shows that heteroscedasticity impact is greater in such models. It seams reasonable to move to another method of estimation. We have chose one of the most frequently used - Poisson Pseudo Maximun Likelihood estimator (PPML).

2.3.2 Poisson Pseudo-Maximum Likelihood estimations

Population-average effects PPML estimator was chosen for estimation of the model, because the variance of variable for areas of countries is large and population has uneven distribution. In such way it can be taken into consideration.

Another set of variables was chosen, comparatively to OLS estimation.

Firstly, the dependent variable is expressed in logarithm of trade turnover that is counted as sum of export and import flows. That variable expresses the total international trade activity of country-partner. Dummies for common continents, being a part of the same country and having in the past or in the present common colonizer were excluded from observation.

The (1) column reflects specification that uses weighted by population distances and the (2) - that uses geodesic distances and remoteness ratios respectively.

Standard errors in parentheses

^* p < 0.05, ^** p < 0.01, ^*** p < 0.001

All coefficients except contiguity in 2^nd specification are significant and have predicted signs. The 1^st specification has a more appropriate set of variables, particularly due to use of weighted distances. As it was stated in previous sections coefficients can be interpreted as semi-elasticities as in the case of OLS estimation.

The model proves results of all researchers investigating effects of geographic characteristics on values of export-import trade flows: rise in GDP values of countries-partner leads to rise in their trade turnover (if GDP of country-reporter rises on 1% trade turnover rises on 7,7%) and distances have negative signs in both specifications (if weighted distance is 1% greater between countries than trade turnover is lower on 8,89%).

The increase of country-reporter's remoteness calculated with use of weighted by population distance on 1% means greater trade turnover on 4,37% as it has a positive sign. Consequently, the statement by Silva and Tenreyro (2006), that greater remoteness of the country-reporter from the rest of the world enlarges its trade turnover with the country-partner, was proved.

If the value of variable for area of country-partner, that has negative sign as predicted, is 1% higher the trade turnover is lower on 0,78%. This coincides with statement by Silva and Tenreyro (2006) that smaller countries are usually more open to international trade flows.

There is only one country that is landlocked - Mongolia. We can estimate an effect of its landlockedness on its trade turnover with other countries. If Mongolia is a reporting country, than its landlockedness lower trade turnover on 1,4% and if it is a partner, than on 2,88%. This appears due to higher transportation costs associated with no alternative means of transport to export or import goods.

Having a common border rises trade turnover on 0,7%. It is associated with lower transaction costs, because neighboring countries are less distant and they are more culturally similar to a country-reporter sue to its proximity.

Common official language spoken by population in both reporter and partner countries enlarges their trade turnover on 6,05%. The cause of this effect on bilateral trade flows is that trading units can easier understand each other as they speak the same language as their partner. Furthermore, this variable is the indicator of cultural similarity between countries too.

Having trade agreements between countries enlarges trade turnover, as it was stated by Anderson and Yotov (2016) leads to larger export-import trade flows: if countries in the pair have preferential trade agreement, their trade turnover is on 3,43% larger, if free trade agreement - on 1,08%. This is associated with a greater integration of countries in Asia-Pacific region.

All things considered, comparing two methods obviously leads to choice of PPML. Firstly, it was stated that it copes with two main challenges of gravity data: heteroscedasticity and zero trade flows. Secondly, it is generally assumed that logarithms of GDP of country-reporter and country-partner should be close to 1. In OLS method this suggestion takes place, whereas PPML reveals semi-elasticities that are significantly less than 1. This also reflected in coefficients of all other explanatory variables, that are common in these methods: contiguity . This reflects that homoscedasticity of data significantly influences OLS estimations and leads to biased estimations. Estimations of data with use of PPML method with population-averaged effects gives more realistic coefficients for variables.

2.3.3 Case of Russia

Russian Federation has unique geographic peculiarities that cannot be avoided when analyzing its trade patterns with other countries: enormous area, a great amount of neighboring countries, the location both in Asia and Europe and etc. After difficult for this country XXth century its extent of international trade liberalization have increased and now it is more opened for cooperation with other countries in the world. However, the trade accent moves from Europe countries to Asia-Pacific countries that have grown economically a lot in last decades. That requires an empirical analysis. In the previous section, we have revealed that the 1^st specification gives better results. The results of its test for Russia are represented in the table below. However due to absence of free trade agreements between Russian Federation and countries of Asia-Pacific region there is no sense to include this variable, furthermore the variable for landlockedness of country-reporter can also be deleted as it contains only of zeros, as we know that only Mongolia is landlocked in our list of countries. Furthermore, there is no sense to include remoteness as it is similar for all partners.

Standard errors in parentheses

· p < 0.05, ^** p < 0.01, ^*** p < 0.001

All coefficients are significant at least on 5% level. The increase of the value of GDP of Russia on 1% increases trade turnover on 1,31%, increase in GDP value of country-partner increases it on 5,21%. The 1% greater distance weighted by population between Russia and country-partner means the lower on 9,6% value of trade turnover. Greater on 1% partner's area is associated with lower Russia's trade turnover on 1,05%. Landlockedness of Mongolia lowers Russian trade turnover on 0,03%. Trade turnover with countries-partners that have common border with Russia is 1,62% higher. The participance in preferential trade agreements with USA and Canada increases trade turnover with them on 7,49% comparatively with other countries.