5. Dynamics of Entry and Exit
Section 4 showed while New Zealand's aggregate exports have diversified substantially over the past 18 years, both in terms of the composition of products exported and the countries traded with, this diversification has slowed since 1995. At the same time, over the past 10 years we have seen a mild degree of increased concentration in the upper quartile, with large exporters making up a greater proportion of total value over time.
Changes in the degree of concentration can occur through:
(1) Doing more of the same – Export value can increase with little change in the composition of either products exported, firms involved or trading partners through increasing the volume of exports or through changes in the unit price of existing export products.
(2) New export activities – new firms beginning to export, new products being introduced, new markets being targeted.16
(3) Exit from some previous export activities – the exit of some firms from export markets, shifts away from some products and markets.
In this section we examine the dynamics of export market entry and exit. Section 5.1 deals with summary statistics which indicate the overall level of entry and exit at product, firm and relationship level. Section 5.2 decomposes aggregate export growth into shares from new products, new destinations and new firms. Section 5.3 looks at the duration of observed trade relationships.
5.1 Dynamics of Export Market Entry and Exit
A better understanding of the factors that determine success in export markets have significant implications for government policy on international connections and export development assistance. New market entry is associated with a variety of fixed (sunk) costs, in the form of investments in market information and marketing, developing networks and distribution chains etc. If they are specific to both the product and the country concerned, these sunk costs may be incurred each time a new relationship is begun. Past trade relationships may reduce the cost of new entries, by giving firms better information about the markets they are going into, or the ability to build off existing networks. Alternatively, if past export experiences have been unsuccessful, there may be reputational barriers that raise the cost of re-entry or the initial entry costs to new firms. In either case, the impact of past experience is likely to decline as the gap between exporting spells increases, as networks and market knowledge deteriorate. Empirical studies of firm level export behaviour find significant persistence in export performance, implying that sunk costs are an important determinant of export market entry (Bernard and Jensen 2004; Roberts and Tybout 1997).
It has been suggested that the costs of market entry are much higher for New Zealand firms, due to the size of our market and the distance to other major markets.17 This also translates, particularly in non-traditional industries, to a lack of reputation in offshore markets. These difficulties may prevent firms from entering export markets, but may also impact upon the success of firms which do enter.
The level of sunk costs is likely to differ across industries, products and destinations. For example, it may be lower in products where New Zealand already has an established reputation for quality, or in products which are relatively undifferentiated and may require less marketing. Equally, some differentiated products may have low sunk costs if their main means of exporting is through internet-based orders. This suggests that intermittent or one-off exporting is not problematic for some firms, as they have low costs of entry and exit. However, for other firms and industries, the costs of export market entry may impact on their ability to begin exporting, their ability to maintain exports over the medium term, and the benefits of consistent market presence over intermittent spells of exporting.
In this section we present some simple statistics which illustrate the degree of volatility in export patterns at the 10-digit product level and the firm level over the years 1996 to 2005. Overall, we find that, while there are a significant number of products and firms that exported continuously over the period, there is also a large group that export for relatively short periods.
5.1.1 Product Level Analysis
Of the 10,090 products which have been exported from New Zealand between 1996 and 2005, between 64 and 72% have a positive export value in any given year. Just over 40% are consistently exported in every year throughout the ten year period (though not necessarily by the same firm). The rest experience at least one year in which they are not exported. Over 10% of products exported are in only one year. Figure 7 shows the distribution of products according to the number of years in which they are exported.
Figure 8 shows the number of products exported per year, divided according to whether it is the first year in which they are observed, they have been observed in previous years but not in the year immediately before, or if they have been exported in both the current year and the year immediately previous. Clearly, this decomposition is affected by the length of past history available. For example, all products show up as "new" in 1996. In any given year after the first three there are considerably more "previous" products than there are "new" products.
This points to a significant degree of volatility in the exports of some products. Among the products that were not consistently exported throughout the ten years, most experienced multiple spells of exporting (Table 8). That is, they were exported for one or more years, followed by a year or more gap, then another spell of exporting. Many of these spells were very short – almost half lasted only a single year. This is discussed further in Section 5.3 below.
5.1.2 Firm Level Analysis
Patterns at the firm level are quite similar, with around 16% of firms which ever export exporting in every year, 38% exporting in only one year, and the rest being fairly evenly spread between two and nine years. Just under 40% of intermittent exporters experienced more than one export spell.
The degree of intermittency in firm-level export patterns is outlined in Table 9. Overall, around three-quarters of exporting firms experience only a single spell of exporting over the ten year period. However, for 38% of these firms, this is because they export in only a single year. Only 22% of firms export in eight or more of the ten years.
As noted above, the existence of sunk costs for export market entry would tend to imply that that firms have a strong incentive to maintain their export relationships once they have incurred these costs. The level of intermittency among New Zealand exporters raises questions about whether these intermittent exporters are successfully taking advantage of specific market opportunities (such as a low exchange rate or a fortuitous offshore order) or whether they have made concerted efforts to enter export markets but have encountered barriers and been unsuccessful. Future work incorporating the International Engagement module of the Business Operations Survey 2007 may shed light on this issue.
5.2 Decomposition of Aggregate Export Growth
As noted above, changes in the value and concentration of exports over time can come from a number of sources. It may be that aggregate export growth occurs through increasing volumes or through increases in the unit value of goods sold offshore. Alternatively, growth in exports may be driven by the introduction of new products, links to new markets, or the entry of new firms. Finally, some export value will be "lost" due to existing export relationships coming to an end.
In this section, we decompose changes in export value over time according to four sources: new entries, exits, ongoing relationships and intermittent exports. This analysis takes it's motivation from the work of Evenett and Venables (2002), who perform a similar decomposition of aggregate exports from 23 developing countries. They find that most of the growth in exports from these countries comes from the introduction of new products and from the "geographical spread of trade" (the export of existing products to new markets). They also find very little exit of products. In this section we replicate the results of Evenett and Venables for New Zealand, and extend these by looking at a more disaggregated product classification and by considering entry and exit at the firm level.
The reader should note that the results at the firm level should be treated with a certain amount of caution. Enterprises which have changed their legal status over time lead to some overstatement of the numbers of and value attributed to entering and exiting firms.
5.2.1 SITC 5-Digit Product Analysis
Following Evenett and Venables (2002) and using the SITC5 data from 1988-2005, products are defined according to their export performance in the first and last four years of observation, and classified into 4 status groups:
- Continuing - the product was exported at some point between 1988 and 1991 and between 2001-2005.
- Exit - the good was exported between 1988 and 1991 but not in 2001-2005
- Entry - the good was exported between 2001 and 2005 but not in 1988-1991.
- Intermittent - the good was not exported in either period (but was exported at some stage in between)
Values refer to the average value over the four year periods. As such, the measured growth in exports refers to the difference between annual average exports in the last four and first four years of the sample. This smoothes some of the variability in export values from year to year.
The use of a four year period at either end also mitigates somewhat the problems of left- and right- censoring. That is, we reduce the probability of mis-identifying goods as entries if they were not exported in 1988 but had been exported previously. However, the reader should keep in mind that many products which are labelled as "entries" may in fact have been exported in previous years, and those defined as "exits" may yet be exported again in 2006 and future years. The data shows a considerable amount of intermittent exporting, such that a four year gap need not imply permanent or long-term exit.18
Our preferred analysis differs from that of Evenett and Venables in two main ways. Firstly, we do not include a threshold for significance.19 Evenett and Venables impose a threshold for "economic significance" in export values, with exports which fall below this threshold being counted as not exported. As such, a product which is exported in low values at the start of the period and increases over time would be counted as an entry in their analysis but as a continuing product in ours. Secondly, our main results are presented using SITC 5-digit product classifications, while Evenett and Venables use SITC 3-digit groupings.20
Around 87.5% of all products were exported both in the early period and the later period, and are therefore classified as continuing products under the definitions above. Around 5.6% are classified as entries, while 4.7% count as exits. 67 products were exported intermittently – that is, they were not exported in either the early or later periods, although they were exported at some point in the full sample period (Table 10).
We can also consider the total contribution that each status group made to export value growth over this period, by calculating the sum of export growth associated with each status group. This shows that almost all the increase in export value in New Zealand came from increases in the value of existing products, rather than from new entries. Of the total increase in export value of $13.4 billion, $13.6 billion came from increased exports of existing products, while only $13.3 million came from new products. The difference of -$140 million comes from the exit of products which were exported in the early period but no longer exported in 2002-2005.
There are some differences in the pattern of dynamics according to SITC groups (Table 11). Product groups which seemed to be "punching above their weight" in terms of new product entries included Food and Live Animals (14% of new entries, 11% of total products), Crude Materials, inedible (18% of entries and 8% of all products), and Chemicals and related products not elsewhere specified (20% of new products and 16% of the total). In contrast , Manufactured Goods Classified Chiefly by Material and Machinery and Transport Equipment saw a lower proportion of new goods (16 and 17% respectively) relative to their share of the total (27 and 21 percent).
In Crude Materials, the level of entry seems to be associated with a high degree of churn, as this group also made up a high proportion of exits (20.1%). Conversely, churn in Machinery and Transport Equipment seems to be low, with both entries and exit shares being low relative to the number of total products. Other groups saw a greater overall change in the number of export products. Food and Live Animals have seen an increase in the total share of products exported, while Manufactured Goods by Material have seen a fall in their share of export products.
The high concentration of value in continuing products contrasts sharply with the results of Evenett and Venables, who find that new products are a significant source of export growth in most countries in their sample. To confirm whether this is a function of differences in the methodology employed, or differences in actual export experiences of the sample countries (the Evenett and Venables sample was concentrated on developing countries) we repeated our analysis to match the methodology as closely as possible, through the use of 3 digit SITC classifications, the introduction of a $50,000 threshold for significant export value, and the deflation of trade values to 1995 US dollars.21 One significant difference remains – the observation period in Evenett and Venables's data was from 1970-1997, while our data is available from 1988-2005.
The low levels of new entry do not appear to be driven by differing methodologies. Rather, from the very start of the period under observation, New Zealand has had significant exports in a wider range of products than most of the countries considered by Evenett and Venables. At the start of the period, New Zealand exported 234 discrete products at the 3 digit level. In contrast, only 3 of the countries studied by Evenett and Venables exported more than 200 products at the beginning of the observation period. Those countries (Brazil, China, and Malaysia) also show little or no growth from new products. Countries with extremely high shares of growth from new products (eg Bolivia and Ghana) had very low numbers of product line exports at the beginning of the period (41 and 43 products respectively).
This is related to both the country selection imposed by Evenett and Venables, whose focus was on developing countries, and the time period covered by the data. The NZ data covers the period from 1988 to 2005, while the international data is from 1970 to 1997. Given the longer time period available to Evenett and Venables, there is greater scope for change in the composition of merchandise exports. Similarly, the composition of New Zealand's data in 1970 may have been very different from that in 1988, with fewer products exported. By the end of the observation period (1993-1997) many of the countries covered by the Evenett and Venables study were nearing the level of export product variety found in New Zealand in 1988.
5.2.2 Product by Destination Analysis
While there was little evidence of growth from new products, or from exporting to completely new countries,22 we now turn to an analysis of products by destinations.
Here, we decompose those products which were exported both at the start and the end of the period according to whether they were always exported to the same countries, began being exported to new countries, or exited from any countries they had previously been exported to.
The breakdown of product-destination relationships is expressed in Figure 9. The top two tiers of the diagram refer to the preceding analysis. The top tier shows the total value and growth of all 2976 products exported over the 18 year period. The second tier shows the number of products in each status group, the total value and the amount contributed to export value growth over the observation period. The lowest tier decomposes the changes in export value of products which were exported both at the beginning and end of the period according to the dynamics regarding the destinations they are exported to.
Only a small minority of the 636864 potential product-destination pairings (2976 products by 214 destinations) are ever observed. That is, most countries never received most goods from New Zealand. Although tier two showed that there were very few new products exported, among ongoing exports the number of new relationships exceeds the number of continuing relationships, implying a significant amount of geographic spread of product exports.
Averaging across product groups, the average product was exported to 12.21 destinations in 1988-1991 (Table 12). By 2002-2005, the mean number of export destinations for a given product had risen to 18.2. On average there were 9.65 new markets per product.
5.2.3 HS 10-digit Product Level Summary
Turning to the shorter time period, and more detailed product classifications available for the HS10 aggregate and firm-level data, we focus on this final decomposition of export growth. Changing the number of years of data and the level of disaggregation forces us to reconsider our definitions of new, exiting, ongoing and intermittent exports.
Due to the shorter time period available, we reduce the number of years at the start and end of the period on which we base our definitions. Also, in order to cope with the higher degree of intermittency of exporting at highly disaggregated levels, we explore three alternative definitions for a product to be classified as an entry or exit.
Our first option is to use the same definitions as above, but reduce the time period considered at the start and end to three years rather than four, due to the shorter total length of time available.
A second option is to reduce the time period to three years and tighten the requirements around the consistency of exports for a product/firm to be recognised as an entry or exit, in order to separate out those exports which are intermittent. We therefore define:
Entry: a firm (product/country) is defined as an entry if it does not export (is not exported/is not exported to) in any of the first three years, but is exported in at least two of the last three years.
Exit: exported in at least two of the first three years but not exported at all in the last three years.
Continuing: exported in at least two of the first and last three years
Intermittent: everything else – those products which are not exported consistently at the start or the end of the period, or both.
A final, and even more restrictive option, would be to require that in order to be classed as a consistent export at either the start or end of the period, a product (or firm, or country) must have been present in each of the first or last three years.
While the choice between these definitions, is fairly subjective, this will be the case with any classification of this type. Table 13 summarises the impact of the different definitions by noting the number of products, total export value over the ten year period, and average export value and growth which would fall into each product type depending on the definition applied. In all cases, the merged product category is excluded.
Regardless of the definition used, the bulk of export value and growth is attributed to continuing products. However, changing the definitions makes a significant difference to the numbers of products and the value in each of the other categories. In particular, placing a more stringent restriction at either end raises the number products designated as "intermittent". This increases the overall value associated with these products but lowers their share of growth. In future, we hope to complete further robustness tests of different definitions of entry and exit. For the moment, our preferred option is to take the intermediate option.
Using the intermediate definition above, we briefly cover the distribution of value at the aggregate level over the past ten years. Rather than repeating the full analysis presented above, we simply reproduce Figure 9, showing the relative contributions of different types of product under the second definition above (Figure 10). Growth is determined as the difference in average export values over the first and last three years. Overall, the HS10 data support the conclusion that the vast bulk of export value, and the growth therein comes from increasing value of our existing exports. The increase in value attributable to new products only just outweighs that lost from exiting products.
The 10-digit data also shows a lower share of growth attributable to geographic spread among existing products, with the value gained from new entries being substantially counterbalanced by destruction of export value due to exit of specific products from some markets.
Although the number of products exiting exceeds the number of new products, the share of growth is greater for entries. This is in part explained by the use of current New Zealand dollars rather than constant, leading to over emphasis of the value of later years relative to earlier ones.
5.2.4 Firm Level Analysis
Turning to the firm level, we restrict the sample to those firms which export in at least one year over the ten year period in order to focus on the degree of entry and exit. Anywhere between 37 and 60% of the firms which are ever identified as exporters have zero export value in any given year (Table 14).
The number of firms which export has grown significantly each year until 2005. Figure 11 distinguishes the number of firms exporting per year according to whether they are new to exporting, had exported in the previous year, or had exported previously but not in the most recent year, in the same format as the product level analysis presented in Figure 8. While at the product level many "new" products had in fact been exported in previous years (Figure 8), the firm level data suggests that growth in numbers has come mainly from the entry of a substantial number of new firms each year (Figure 11).
There is a wide dispersion in the number of trading partners that each firm exports to, and the number of products which they export (Table 15 and Table 16). Across the ten year period, the median number of countries exported to and the median number of products exported were two and three respectively, but some firms exported to more than 28 countries and had a product range of up to 100 products in some years.
As would be expected, larger value exporters tend to export both more products in any given year, and to more countries. This is particularly evident in the number of products exported, with the average number of products exported per year for those firms in the top quartile of export value being more than ten times that of firms in the bottom quartile (Table 17).
Among those exporters who export continuously over the ten year period, there is a tendency for them to expand the number of products they export, and the number of destinations they export to (Table 18 and Table 19).
Decomposition of Export Growth
Returning to the decomposition of export growth at the firm level, Table 20 shows the number of firms classified as entries, exits, continuing or intermittent according to the definitions set out in Section 5.2.3 above. Again, our preferred specification is to require a firm to export in at least 2 of the three years in question.
Figure 12 presents a decomposition of export growth at the firm level. The first level mimics that shown in Figure 10 above, but considers the degree of entry and exit of exporting at the firm level. We also note the number of new, exiting and intermittent firms whose change in export status is also accompanied by firm birth or death. That is, for export entrants we note the percentage of that group of firms which was "born" between 1996 and 2005, for exits the percentage that ceased, and for intermittent exporters, the proportion which either started or ceased over this time period. As noted above in Section 2.2, some of these births and deaths may be spurious, as they may reflect either the actual start or ceasing of a firm or changes in enterprise numbers due to changes in legal status.
The consideration of export dynamics within continuing firms is complicated by the large number of potential ways to define relationships at the firm level. We therefore focus on those relationships that can be recognised as being new, ongoing, or ending, and disregard the various types of "intermittent" relationship. Table 21 presents a quick view of the relationship breakdown within continuing firms for a subset of relationship types. This is an area which will be covered in more depth in future analytical work, looking at the determinants of market and product entry and exit at the firm level, including the impact of exchange rates, geographical factors, firm characteristics and the economic environment on export dynamics. In the meantime, it is worth noting that the vast majority of new relationships come from the introduction at firm level of new products into established relationships with existing partners, while the majority of relationships ended due to changes in the firms product range, rather than the end of relationships with partners.
5.3 Duration of Trade
While the discussion above considers the importance of new products and new destinations in generating export growth, a second aspect of export dynamics is associated with the end of export relationships. The evidence in Section 5.1 showed that while a large number of products were exported continuously over the period from 1988 to 2005, there were also a large number of products which were exported for only a short time or were exported intermittently. In this section we consider the duration of trading spells at the level of the product by country relationships, firm exporting spells, and firm-country-product export relationships. A spell is defined as a continuous period of exports over one or more years.
In future work we will focus more closely on the firm and environmental determinants of both entry and exit of export markets. At this stage we simply provide some background descriptive information on the length of exporting relationships at various levels.
To our knowledge, the only published studies of the duration of trade relationships to date are those of Besedes and Prusa (2006a,b). Besedes and Prusa (2006a) show that trade relationships are very unstable when viewed at the disaggregated 7 digit Tariff Schedule level. They discuss this in terms of the duration of US trading relationships, where a trade relationship is defined as a period in which there is consistently non-zero trade in a given product with a given country over one or more years. They also find that there is a significant amount of duration dependence in export relationships. That is, while the median duration of trade relationships with the United States is only around two to four years, those relationships that survive the first few years have a very low probability of failure and tend to survive for a long time.
In other work (Besedes and Prusa 2006b), the authors show that there are significant differences in the average duration of trade relationships depending on the level of product differentiation. Using the classification developed in Rauch (1999), they find that in their data, differentiated products had a median survival time of five years, while homogenous products and reference priced goods had a median duration of only two years.23 They suggest that this is because homogenous goods are more highly affected by changes in tariffs and trade costs than differentiated goods.
Alternatively, higher costs of market entry might deter all but the most successful differentiated product firms from beginning to export, while those that enter are more likely to stay in markets once they have begun to export, knowing that if they leave and wish to return they will again have to incur significant entry costs.
Here we examine patterns of trade duration using New Zealand data at the firm and product level, considering difference in duration according to simple product, market and firm characteristics. The results are purely descriptive, but throw up a number of areas for further investigation.
In their analysis of trade relationships involving exports to the US, Besedes and Prusa assume that multiple spells within the same product can be treated as independent of each other, showing that this assumption has little impact on their results. The main effect of treating multiple spells as independent and equally important is to place greater weight on those relationships which experience multiple spells, and less weight on those which experience only a single spell. For example, a relationship which experiences a single, 18 year spell of trade will be counted towards the average trade duration only once, while relationships experiencing multiple short trading periods will be counted several times. Equally weighting all spells will therefore produce a lower estimated trade duration than weighting according to relationships.
Throughout our analysis we present two alternative specifications side-by-side, one in which each spell receives equal weight (the unweighted results) and one in which each relationship receives equal weight. The latter is achieved by weighting each observed spell by the inverse of the number of spells observed for the relationship. We also provide some indicative comparisons across various treatments of multiple spell relationships.
The different specifications have different impacts according to the level of disaggregation and the length of the observation period. Table 22 sets out the distribution of numbers of spells observed within relationships at different levels of analysis. This suggests that treating multiple spells as independent will have a greater impact on results for SITC 5-digit products than for that of HS 10-Digit products, for which we have a shorter observation period.
Additionally, we note that there are various ways in which we might wish to treat censored spells – that is, spells which are observed in the first and last years of the sample period. For these spells we cannot determine whether they continue prior to, or after, the observed period. In the current paper we consider mean durations of observed spell lengths, and some simple methods of allowing for censoring by the use of survival analysis techniques. A more formal treatment of censored spells will follow in future.
5.3.1 SITC 5-Digit Relationships
As noted above, there are various ways in which we can account for multiple spell relationships. We could continue to treat all spells as independent and equally important. We could consider only relationships which experience only a single spell. We could consider only the first spell of each relationship. We could consider sub-groups of relationships, where the sub-group is defined by the number of discrete export spells experienced by that relationship. Finally, we could weight analysis according to the number of spells experienced, with greater weight placed on those relationships which experience only a single spell.
Considering these options, we first present summary statistics of observable spell durations for the SITC 5-Digit data in Table 23 (that is, not accounting for censoring at the start or end of the sample period). By definition, all spells last longer than 0 years. No spells last longer than 18 years, as this is the period of observation.
At the product level, we see that the mean spell length is reasonably long, at 8 years, but that this is mainly driven by products which are exported continuously throughout the 18 year period. The median spell length is only 4 years. As expected, reducing the sample to a single spell for each relationship shows up as having higher proportions of short spells, and lower proportions of longer spells. Interestingly, the distribution of first spells differs from that of final spells, with final spells being considerably longer overall. This difference is understated in the reported results, as these include single spell relationships. Removing these would suggest an even larger difference in the average spell length between first and last spells among repeated spell exports.
At the level of product-destination relationships we see a very different picture (Table 24). Regardless of the choice of how multiple spells are treated, the median observed spell length is very low. Again, median spell lengths are somewhat longer for the last than first spells.
Allowing For Censored Relationships
The figures above are based on observed spell lengths. As such, a five year spell ending in 2001 is treated equivalently to one ending in 2005. In reality, we would expect that many spells which are observed in the first or last year of the sample period actually extend longer than we have record of them. In particular, it seems likely that the majority of relationships which are consistent over the entire 18 year period will extend beyond those 18 years, in one or both directions.
Survival analysis techniques allow us to control for this type of censoring. Survival analysis is used to calculate expected durations, based on the observable data while allowing for spells that are not observed to finish. Spells which are censored are treated as surviving for "at least x years".
Here we consider some simple descriptions of spell durations, both for the sample as a whole and comparing across a number of simple distinctions. We report two common measures of duration – the survival function and the estimated hazard function. The survival function plots the share of all observed spells which survive at each duration. The hazard function plots the estimated probability of failure at each duration, conditional on a spell having lasted a certain length of time.
Figure 13 below plots the survivor function across a number of comparisons across all spells, showing both unweighted and weighted estimates.24 Perhaps the most striking feature is the drop which occurs in the first year – over half of all observed spells fail after a single year.
The survival rate is noticeably higher using the weighted analysis. In the unweighted analysis, the results place greater emphasis on relationships with multiple, short spells, and less on those with a single, longer spell. Plotting the hazard rates for the weighted and unweighted spells shows that the weighted results does indeed have a noticeable effect in spreading the survival function in the SITC5 data. When equal weights are placed on relationships, rather than spells, the impact of multiple, short spells within relationships is significantly reduced.
The hazard function estimates suggest that the probability of export relationship failure is high in the first few years, but drops rapidly in the following five years (Figure 14), as found by Besedes and Prusa in relation to exports to the United States.
We also consider differences in the duration of export spells according to the characteristics of the products or destinations involved. Figure 15 and Figure 16 compare the impact of some simple distinctions on observed relationship durations.
We first consider descriptive statistics comparing the duration of exporting spells between high-income and lower-income destination countries. To consider this question we plot survival functions separately for OECD and non-OECD destinations.25 The results suggest that export relationships tend to last longer when they are with OECD countries (Figure 15). This may reflect a combination of factors, including the higher income of the destinations and hence their greater demand for imports, and closer political and historical relationships than with many other potential destinations.
For an initial indication of the relationship between trade duration and trade value we compare the survival rates of trade relationships based on the value of exports in the first year. We cannot be sure whether export spells which are observed in 1988 are actually in their first year or whether they are part of a long-standing trade relationship. Therefore, these relationships (including the 30% of relationships which are observed to trade in each of the 18 years of observation) are excluded from this analysis. For this initial, bilateral comparison, we compare the survival rate across quartiles of export value. As anticipated, durations are longer where the value of trade is higher (Figure 16).
5.3.2 HS 10-Digit product analysis
Results at the 10 digit level are quite similar. Here we present only the summary statistics at the simple product and relationship level, with their associated survival functions (Table 25 and Figure 17). The results tell a similar story to those of the SITC 5-Digit products above. Lower mean durations are most likely due to the shorter sample period.
The relationship data again shows somewhat shorter durations that the product data (Table 26 and Figure 18). Somewhat unexpectedly, however, we see a lower proportion of relationships failing in the first year using the HS 10-digit classifications than we did using the SITC 5-digit classifications. This implies that durations tend to be longer among products in which New Zealand exports a range of products within the wider 5-digit groups, hence placing more weight on longer duration spells.
Comparing across product and destination characteristics paints a similar picture to that reported above, with longer durations in relationships with OECD countries and for high value products.26 For HS10 products we also compare survival rates across broad product groups. Figure 19 shows high survival rates in agricultural products, with the lowest survival rates among manufactured goods. This presents a somewhat different picture to that found by Besedes and Prusa (2006b), with longer durations among those product groups which tend to be less easily differentiated.
5.3.3 Firm Level Analysis
Interestingly, comparing the product results to the firm results suggests that the mean spell length for firms is in fact longer than that for product-market relationships (Table 27). This implies that many firms maintain exporting while changing the mix of products and countries with which they trade, as discussed above is Section 5.2.4.
Again, the probability of failure is highest within the first few years, then drops rapidly (Figure 20). However, the proportion of firms failing after a single year is lower than the proportion of product-market relationships.
The failure rate differs somewhat by industry, with manufacturers tending to have significantly higher survival rates than either wholesale trade or agriculture, forestry and fishing enterprises (Figure 21). The duration of trade is also very closely linked to the value of trade, with the largest differences in survival functions being between "large" exporters (those in the top quartile of export value in the first year of the spell) and smaller ones (Figure 22). These differences are more substantial than those shown at the product-market relationship level (Figure 16).
There is also a slight difference according to ownership, with foreign owned companies having somewhat longer-lived export spells (Figure 23). As noted above, however, this may be due to the higher export value of these firms.
Turning to export relationships at the firm level, we see that these are even shorter, with the mean export relationship lasting only 1.6 years (Table 28). This reinforces the conclusion above, that many exporting firms maintain exports while shifting between products and markets frequently.
Comparing across different characteristics of the destination income level and product type reflect the results at the firm-level. Relationships with OECD countries are somewhat longer than with non-OECD countries, although the difference is quite small (Figure 24). Agricultural products have the longest spell durations, while manufactured goods have the shortest (Figure 25).
In contrast, while Manufacturing firms have the longest average export spell duration, with Agriculture, Forestry and Fishing, and Wholesale Trade firms having significantly lower survival rates at the firm level (Figure 21), at the product and destination relationship level, firms in Agriculture, Forestry and Fishing have longer export relationships than in other industries (Figure 26). This is likely to reflect the relative stability of the products they are producing, compared to manufacturers which may have a greater variety of products exported for a shorter duration.
Clearly, many of these factors are interrelated. Similarly, there are a number of additional factors that could be put forward to explain why some trading relationships last longer than others – it could be specific to the product (some products inherently experience more continuous demand – butter, stationery and paper, while others are likely to be required only in response to specific circumstances –helicopters, exposure meters), the country concerned (the depth of our relationship and trading experience with that country eg Australia vs China), or the relationship between product and country (eg. having a trading history in some areas but not in others). In future work we will examine in more detail the interrelationship between various factors which are associated with both entry and exit.
Back to Top