A Time Series Analysis of Turkish Trade Patterns at the Sector Level

In our previous research on the pattern of Turkish trade (Erlat and Erlat, 2012) we tried to establish if this pattern had a persistent nature or whether it was dynamic. In doing so we used tools originally developed by Gagnon and Rose (1995) and later used by Carolan, Singh and Talati (1998) and Carter and Li (2002, 2004). These involved (i) classifying the sectors as surplus, balance and deficit sectors and constructing 3x3 contingency tables indicating whether sectors, say, that showed a deficit at the beginning of a period, remained deficit sectors at the end of the period or became balance and surplus sectors; (ii) testing whether the pattern at the end of the period was independent of the pattern at the beginning period, and (iii) constructing histograms regarding the distribution of how long the sectors have been showing surpluses over the period. We consider three aspects of this approach that may require improvement: (i) The results are highly aggregated even though the data used, at least in Erlat and Erlat (2012), are at the SITC 5 digit level. (ii) The results refer to the comparison between the beginning and ending of two periods that are years apart. Thus, how the patterns at the end of the period are reached is not investigated. (iii) The only tools that take the individual sectors and how they behave during the period into account are the histograms. To remedy these shortcomings, we followed Carolan, Mora and Singh (2012)’s lead and applied time series methods to individual sectors to obtain information about the path their trade balances took over the period under consideration. This also allowed us to pinpoint those sectors that have been successful in trade. We first constructed two series using export and import data for the sectors to be considered. First, we have the normalized trade balance for sector i at time t, NBit, to be used as the subject of the time series analysis. Second, we have the normalized trade volume for sector i at time t, NVit, to be used in presenting the results of the time series analysis. The sum of the NVit across i for any t is always 100. It, thus, shows the significance of the ith good (or sector) in overall trade. Since the focus of our time series analysis was the NBit, we established if the trade balance of a given sector increased, decreased or remained the same. This means that we needed to be interested in the long run movement of the NBit; in other words, the trend component in the series. This component may be stochastic, implying the presence of a unit root, or deterministic, implying a trend stationary series. In the second case, the sign of a statistically significant coefficient for the linear trend term will indicate to us the direction of the change while a statistically insignificant coefficient would imply that there has been no significant change in the trade balance of that sector. We used two tests for this purpose. The first one had the existence of a unit root as its null hypothesis and our test for this was the Augmented Dickey-Fuller (ADF) test. The second had stationarity as its null and the test we used was Kwiatowski, Phillips, Schmidt and Shin (KPSS) test. The joint use of the ADF and KPSS tests leads to the classification of the sectors into eight groups. Groups IV-VII contain results where there are no conflicts. Of these IV indicates that the series are nonstationary while VI-VII indicate that they are stationary. Groups I-III and VIII indicate conflicts. However, we used the results in I-III by regarding the stationarity obtained by the KPSS test as an indication that ADF lacks power in the sense that the null of a unit root would have been rejected. The conflict in VIII implies that the NBit has neither a unit root, nor is it stationary. Thus, these sectors were ignored. The data are the same ones used in Erlat and Erlat (2012) and will enable us to compare our results with those obtained in that paper. They are from 5-sectors but we eliminated those sectors that either had no exports or imports or both at any year during the period in question. This reduced the number of sectors to be analyzed to 1118. We also used the technological classification of the data and the presentation of the results as in Erlat and Erlat (2012). When we look at the aggregate results of this paper, we find that there is not much that is new compared to those in Erlat and Erlat (2012). But, when we consider the disaggregated results, we find information about the nature of the dynamism in the sectors classified as such. We find that the number and share in 2001 trade of positive change sectors is larger in all categories except Raw-Material Intensive Goods, a category including more traditional export sectors. sectors but we eliminated those sectors that either had no exports or imports or both at any year during the period in question. This reduced the number of sectors to be analyzed to 1118. We also used the technological classification of the data and the presentation of the results as in Erlat and Erlat (2012). When we look at the aggregate results of this paper, we find that there is not much that is new compared to those in Erlat and Erlat (2012). But, when we consider the disaggregated results, we find information about the nature of the dynamism in the sectors classified as such. We find that the number and share in 2001 trade of positive change sectors is larger in all categories except Raw-Material Intensive Goods, a category including more traditional export sectors. By the same token, Difficult-to-Imitate Research Intensive Goods appears to be the most dynamic sector with 21 top dynamic 5-digit sectors. Hence, we are able to say that Turkey not only has a dynamic tra