Export performance and longevity of new Canadian exporters between 2005 and 2018

July 2025

Table of contents

• 1. Key messages
• 2. Executive summary
• 3. Introduction
• 4. Literature review
• 5. Data, definitions, and basic statistics
• 6. Survival methodology
• 7. Export survival results
• 8. Discussion of results
• 9. Conclusion
• 10. Bibliography
• 11. Annex 1 – Methodology
• 12. Annex 2: Alternative model specifications

1. Key messages

• This report analyzes the export longevity—or duration of export activity—of Canadian exporters, and whether export longevity differs based on firms’ different approaches to international expansion.
• In Canada, the median duration of export activity for new exporters is 2 years. This is a similar export duration as businesses in other countries. Most exporters start and stop exporting periodically due to a few factors not limited to fluctuating resources, contracts, or demand.
• There are some benefits to longer export durations. In general, the duration of export activity correlates with growth in the value of exports, number of export products, number of export destinations, firms’ revenue, and number of employees. Furthermore, the longer a firm export, the lower the likelihood that it will stop exporting as exporting becomes entrenched in their business models.
• In contrast to traditional theories that advocate for establishing a significant domestic presence before exporting, this report does not find significant differences in the long-term export longevity between those who exported early on versus those who established domestic operations first.
• Rather, what is more important for export longevity is firms’ export commitment and confidence; higher initial values of exports, number of export products, and number of export destinations are positively correlated with export longevity for Canadian firms.

2. Executive summary

While it is the objective of policy makers to have Canadian firms participate and grow in export markets, export longevity—or duration of export activity—has received less attention as an area of study. However, export longevity is just as important as the initial decision to export since it is a necessary condition for continued participation and growth in the international markets. This report focuses on the export longevity of Canadian firms that began operations   between 2005 and 2018, and whether different export strategies result in differences in export longevity.

The main results in this report are that Canadian exporters do not generally export uninterrupted for long durations; about 40% of export activity periods end after the first year that a firm starts to (or resumes) export, with a median export duration of 2 years. Only around 3 in 10 export activity periods last 5 years. These results are similar to past results and studies from other countries. Typically, businesses start and stop exporting periodically, due to a number of factors including contract start and end dates, fluctuating demand or clientele, or limited resources.

There are benefits to becoming entrenched in exporting. Export longevity correlates with improved firm performance, including growth in the value of exports, number of export products, number of export destinations, firms’ revenue, and number of employees. This analysis finds that the median export value was around $26,000 in the first year of exporting with an average of 2.6 export products and 1.5 export destinations. For firms that exported continuously, by their 6th year of exporting, their median export values grew to $432,000, with an average of 6.2 export products and 3.7 export destinations.

There are different strategies that a firm can use to reach international markets. The traditional theory presents a sequential approach, where a firm first establishes itself in its domestic market, and then incrementally increases its commitment and resources to international markets. In this report, exporters that followed this traditional theory are referred to as Gradual Global exporters. In contrast to the traditional theory, Born Global exporters and Born Regional exporters are firms that export within 2 years of beginning operations. Born Global exporters follow the “international new ventures” framework in which young firms benefit from early internationalization by benefiting from the learning advantages that accompany youth. Born Global exporters differ from Born Regional exporters in that they have a global reach from the start. Meanwhile, Born Regional businesses export only to the U.S. in the beginning. Born Regional exporters follow the regionalization hypothesis which states that rapid early internationalization is possible but will be most valuable if revenues are coming from the firm’s home region to diminish the liability of foreignness. The U.S. can be considered Canada’s home region since it accounts for the majority of Canadian trade, with business networks and infrastructure focusing heavily on north-south routes.

Once export confidence and commitment are taken into account (represented by various export performance variables and other variables), there are no differences in the long-term export longevity between Born Regional and Gradual Global exporters. Meanwhile the longevity for Born Global exporters is only marginally worse off. In other words, the age of the firm at the start of exporting does not seem to have an impact on long-term export longevity. What seems to matter is the export confidence and commitment. The duration of exporting periods has a positive correlation with the initial value of exports, initial number of products exported and initial number of export destinations.

3. Introduction

While there is a lot of focus on getting firms to export and to grow exports both in terms of value and destinations, export longevity (or export survival) has received less attention but is just as important since without longevity, a firm cannot participate consistently and grow in the export markets. This report focuses on the export survival of Canadian exporters and whether or not different internationalization strategies can result in differences in export survival. The results in this report are similar to results from past studies, Canadian exporters do not survive well with more than half of exporting periods stopping after the second year.

The different internationalization strategies do not seem to have large impacts on the long-term export survival as once the export confidence and commitment (proxied by the size of the value of exports, the number of products exported and the number of destinations) are accounted for, the differences in survival between the different internationalization strategies became marginal. This means that the age of the firms at the start of exporting does not have an impact on export survival, which run against the traditional internationalization model which indicates that a firm should first establish itself in its domestic market, and then incrementally increase its commitment and resources to the international market. Section 4 of the report reviews the literature on Canadian export survival. Section 5 provides basic information about the data, definitions, and basic statistical summaries. Section 6 reviews survival methodologies and section 7 provides the results of export survival of Canadian exporters. Sections 8 and 9 provide high-level discussions of results and conclusions.

4. Literature review

4.1. The different internationalization strategies

There are different ways in which a Canadian firm can expand internationally. Some firms expand early in age, close to their inception, while others take longer time as they try to establish domestically before expanding internationally. Another dimension is the geographic location of expansion, as many Canadian firms will expand to the U.S. only first as the U.S. is Canada’s number trading partner by a distance, while other Canadian firms will attempt to expand to multiple international markets at once.  

The traditional internationalization model, also known as the stage model or the Uppsala model, is the earliest internationalization model that was developed by a group of Swedish economists (Johanson & Wiedersheim-Paul, 1975; Forsgren & Johanson, 1975). The Uppsala model presents a sequential approach, where a firm should firstly establish itself in its domestic market, and then incrementally increase its commitment and resources to the international market. According to the stage model, the risk and uncertainty associated with foreign market entry is high and therefore businesses should be cautious and gradually enter international markets. In this report, Canadian firms that follow the stage model are called “Gradual Global” firms.

Another framework for internationalization process of firms is called the “international new ventures” (INV) framework. INVs are companies “that from or near foundation, obtain a significant portion of total revenue from sales in international markets” (Knight & Cavusgil, 2005). According to the INV framework, firms may benefit from going international early by capitalizing on “learning advantages that accompany newness” (Autio et al., 2000). That is, a young firm can more easily adapt its processes and structure to the international environment. Early internationalization is also argued to be important in the development of young firms as exposure to foreign markets is expected to trigger the exploration and exploitation of new opportunities and resources (Sapienza et al., 2006). In this report, Canadian firms that follow the INV framework are called “Born Global” firms. 

Another theory called the regionalization hypothesis states that rapid early internationalization is possible but will be most valuable if revenues are coming from the firm’s home region to diminish the liability of foreignness (Rugman & Verbeke, 2004). This can be seen as a way to balance the benefits from learning opportunities in the foreign markets at an early age with the high risks and uncertainty associated with international trade. The regionalization hypothesis is more applicable to a country like Canada, where a majority of international trade activities occur with one market, the United States, and therefore the United States can be considered Canada’s home region. In this report, Canadian firms that follow the regionalization hypothesis are called “Born Regional” firms.

4.2. Export survival of Canadian exporters and international comparison

A key challenge for exporters is to stay exporting and not become a victim of export failure, but research on export market exit has received limited attention (Sandberg et al, 2019; Bernini et al, 2016; Chen et al., 2016; Gima et al, 2003). In Canada, new exporters are likely to stop exporting within a few years, but for those who manage to survive, their export revenues increase significantly. According to Chen & Yu (2010), who explored Canadian exporter dynamics between 2000 and 2006 using Statistics Canada’s Exporter and Business Register databases, about 50% of new exporters who started in 2000 failed by the end of the 2^nd year in the export market, and only a quarter survive and become established continuing exporters by the end of the 6^th year. Furthermore, they found that once the new exporters established themselves in the export market, the value of exports increased significantly. This highlights the importance of sustaining trading relationships in order to remain and expand in the exports market.

Studies in other countries have found similar results, as most studies also find that most exporting spells generally do not last long. Most export flows have been found to cease within 2-3 years (Besedeš and Prusa, 2006; Esteve-Pérez et al., 2013). Lodefalk and al. (2022) concluded that export relations in their study are short-lived. According to the literature review by Bandick (2019), Besedes and Prusa (2011) found that the median survival rate of the manufacturing exporters of 46 developed and developing countries is only 1-2 years. Furthermore, the short‑lived export episode has been found in other studies: Eaton et al (2007) for Colombian exporters, Volpe and Carballo (2009) for Peruvian exporters, Ilmakunnas and Nurmi (2010) for Finnish exporters, and Choquette (2019) for Danish exporters.

In another study, Sabuhoro and Gervais (2004) used Statistic Canada’s Exporter Registry data to study the factors that determine the export exit of Canadian exporters. The study found that most Canadian exporters do not export for long, with the probability of exiting the export market before 12 months being 42.2% and the median survival time for Canadian exporters being 20 months. Factors that increase the likelihood of survival on foreign markets include:

• an increase in relative size of the establishment, proxied by the ratio of the firm’s monthly value of exports to the average monthly value of exports,
• an increase in the number of exported products and destinations,
• exporting to the U.S. Eastern Seaboard,
• being a manufacturer, except when compared against firms in agriculture and related services, and fishing and trapping, logging and forestry,
• having previous exporting experience.

Esteve‐Pérez et al. (2005) studied Spanish firms’ manufacturing data from 1990 to 2001 to find the factors that determine export survival. They found evidence that supports the existence of negative duration dependence–that is, the longer a firm continuously exports, the longer it will continue doing so. Another important finding is the fact that exporting spells to closely related markets are significantly longer. Spanish firms selling to the European Union and to the rest of the OECD countries face a lower risk of ending an exporting spell. Other factors such as an increase in firm size and productivity also enhance export survival.

4.3. Export survival of born global and born regional firms in Canada

There are few studies on the export survival of Canadian Born Global firms. Sui (2009) compared the export survivability of Canadian Born Global and Canadian Gradual Global firms. The author defined “Born Global” firms as a firm that “started to export within two years of its inception and that during its first year of export activity, no less than 25% of its revenue is from exporting.” The authors also filtered firms based on the following conditions and reasons:

• firms in the manufacturing industry only, as this study was intended to investigate firms that manufacture their own products,
• firms that were established between 1997 and 2004 only (the study period), as this took care of the left-censored problem (i.e. an exporting spell that started before 1997 and ended during the study period),
• firms with less than 500 employees as the aim of the study was to investigate the internationalization process of small and medium-sized enterprises (SMEs),
• firms where annual revenue has to be higher than $30,000 for at least one year during the study period, to ensure that firms in the sample are active,
• firms where annual value of exports has to be higher than $2,000 for at least one year during the study period, to ensure that exporting is an important part of a firm’s business.

The result from Sui (2009) shows that with all else held constant, the probability of survival of Born Global Canadian firms in the exports markets is 6% lower than that of Gradual Global firms. After correcting for endogeneity, the difference in the probability of export survival became statistically insignificant.

Sui and Baum (2014) repeated the same process, but this time there was a differentiation between Born Global firms and Born Regional firms. Born Regional Canadian firms were defined as firms that commenced exporting within 2 years of inception, have an export intensity of 25% or higher and only exported to the US market during its first year of export activity. Born Global Canadian firms were defined as firms that exported within 2 years of its inception, had an export intensity of 25% or higher and exported to more than just the U.S. markets during the first year of export activity. Gradual Global firms were all other exporters. Results show that without correcting for the endogeneity in firms’ internationalization choice, Born Global firms have the highest probability of exit from exporting, followed by Born Regional firms. Once the endogeneity in firms’ internationalization choice is corrected for, neither being Born Global nor being Born Regional have a statistically significant effect on the survival of firms in the export market when compared against Gradual Global firms.

Born global as a strategic choice

An important aspect of Born Global firms is that “their organizational form is a conscious value maximizing strategic choice by their management” (Mudambi & Zahra 2007). In other words, the internationalization choice between Born Global and Gradual Global is a strategic choice of a firm “determined by its particular set of resources and competencies and is therefore endogenous” (Mudambi & Zahra 2007). To correct for this endogeneity, the authors used a two-stage model, with the strategy choice determined in the first stage and firm survival probability (between Born Global and Gradual Global) determined in the second stage.

Similarly, Sui (2009), Sui and Baum (2014), and Sandberg et al. (2019) also corrected for this endogeneity in internationalization choice using a two-stage estimation. In the first stage, the probability that a firm will choose either Born Global, Born Regional or Gradual Global was predicted. The second stage used the predicted probability from first stage to estimate the survival of firms in the export market. Lastly, to verify the robustness of their two-stage estimations, a split-sample method was used.

4.4. The definition of born global firms

A literature review by Bader and Mazzarol (2009) found that there wasn’t any consistent definition of what constitutes a “Born Global” firm, but some common criteria do exist, such as a firm having operation (sales) in at least one foreign country early in its life cycle. Bader and Mazzarol (2009) found that the majority of the studies on “Born Global” firms referenced the definitions of either Oviatt and McDougall (1994), Knight and Cavusgil (1996), Rennie (1993) or used a combination.

Oviatt and McDougall (1994) defined Born Global firms as “a business organization that, from inception, seeks to derive significant competitive advantage from the use of resources and the sale of outputs in multiple countries.” They also proposed that “a new venture should control assets, especially unique knowledge that creates value in more than one country.”

Knight and Cavusgil (1996) listed the following criteria for a Born Global firm:

• small, technology-oriented companies,
• operate in international markets from the earliest days of establishment,
• rely on cutting edge technology in the development of ‘relatively’ unique product or process or product innovations,
• managed by “entrepreneurial visionaries” who view the world as a single, borderless marketplace,
• export one or several products within two years of establishment,
• export at least a quarter of total production.

The definition of Born Global firm used by Rennie (1993) can be summarized as “a firm that has an acceptable time lag of two years from formation of the firm to its first international sale to at least one foreign country, with minimum foreign sales of at least AUD$12 million comprising seventy-five per cent of total sales” (Bader and Mazzarol, 2009).

These three definitions and other definitions of Born Global firms have many criteria and Bader and Mazzarol (2009) wanted to find the most common criteria among all the definitions in order to have a reduced and simplified definition of Born Global firms. They settled at the following definition: “a new firm that makes at least one international sale to any new market within two years of formation”. However, Bader and Mazzarol (2009) agreed that the exclusion of an export to sales ratio might not be acceptable to everyone since many empirical research include the criteria of “in the first year of exporting, no less than 25% of a Born Global firm’s revenue is from exporting”. Therefore, the export to sales ratio of 25% will also be used in this study as one of the defining criteria for a Born Global/Born Regional firm, in addition to other criteria described in further details below (see Section 5.2).

5. Data, definitions, and basic statistics

5.1. Data

The main database is Statistics Canada’s Trade in Goods by Exporter Characteristics (TEC), where important export variables such as the value of good exports, numbers of countries the firm exported goods to, and number of products exported (see Annex 1.1 for more details). For this report, TEC data from 2005 to 2018 was used. This was merged with National Accounts Longitudinal Microdata File (NALMF) to obtain firm characteristics such as firm’s birth date, revenue, sales of goods and services, size, and industry. This was further merged with Statistics Canada’s Trade in Goods by Importer Characteristics (TIC) and Activities of Multinational Enterprises in Canada (AMNE) to obtain importer and multinational status. However, TIC and AMNE are only available from 2010 onward.

5.2. Defining born global, born regional, and gradual global firms

In this report, since the data is from 2005 to 2018, firms born before 2005 cannot be classified as either Born Global, Born Regional or Gradual Global and therefore they were classified as “Censored”. For firms born in or after 2005, some firms were classified as “Other” since they 1) had missing data and therefore was unclassifiable, 2) never had revenue over $30,000 per year or 3) never had exports over $2,000 per year. “Censored” and “Other” firms are left out of some of the analyses in this report since the purpose is to make comparison between the different internationalization strategies (Born Global vs. Born Regional vs. Gradual Global). Filtering out firms that never had revenue over $30,000 per year or never had exports over $2,000 per year is similar to Sui (2009). 

For the rest of the firms, a firm is Born Global or Born Regional if it exports within 2 years of inception and exports 25% or more of revenue within 2 years of inception, otherwise, it was classified as Gradual Global. To separate between Born Global and Born Regional firms, a firm is classified as Born Regional if it exports to the U.S. only during the first year of exporting, otherwise, if the firm exports to a country outside of the U.S. or to the U.S. and another country, it would be classified as Born Global (see Figure 1).

Figure 1: Classification of born global, born regional and gradual global firms

5.3. Basic statistics on Canadian export dynamics between 2006 and 2017

Between 2006 to 2017, approximately 40,000 firms exported goods each year, for a total of 485,600 export observations over the entire period (export observations represent the number of exporting firms each calendar year) from Canada (see Table 1). Of these 485,600 export observations, 14.8% were one-timers, which means that the exporting spell^{Footnote 1} lasted only one year, that is the same firm did not export in the previous year and did not export in the year after. The majority (62.1%) of export observations in this period were continuers, that is the same firm exported in the year before and after the current year. Over this period, 11.3% of export observations were entrants, which are the beginning of an export spell, that is the same firm did not export in the previous year but exported in the current year and the year after. A similar share (11.8%) were exiters, which are the end of an export spell, that is the same firm exported in the previous year and current year but did not export in the year after (see Table 1).

Of the 485,600 export observations between 2006 and 2017, approximately 10,400 were made by Born Global firms, 8,500 by Born Regional firms and 58,600 by Gradual Global firms (see Table 1). The rest were either exporting observations from censored firms or other firms where the identification of international expansion strategy was not possible. The results show that exporting spells from Born Global firms are less likely to exit than exporting spells from Born Regional firms, which in turns are less likely to exit than exporting spell from Gradual Global firms. Only 7.0% of exporting observations from Born Global firms were one-timer, slightly less than the 7.2% share for Born Regional firms and a lot less than the 23.1% share for Gradual Global firms (see Table 1). The share of exiters is also smallest for Born Global firms, followed by Born Regional firms and then Gradual Global firms (see Table 1).

Table 1: Number of export observations between 2006 and 2017, by international expansion strategy

5.4. Export survival’s relation with export performance and firm performance

While the literature review (section 4) points to the fact that most exporters do not last long in the international market and data in section 5.3 (table 1) shows the existence of many one-year exporters, those firms that managed to stick with exporting see benefits in terms of improved export performance and firms’ performance. On the left side of figure 2, it can be seen that longer lasting exports correlates with higher value of exports, higher number of products exported and more export destinations. On the right side of figure 2, longer lasting exports also correlates with better firms’ performance such as increased revenue and employment.

Figure 2: Export performance (left graph) and firms’ performance (right graph) through years of exporting

Note: All exporters where the firms could be identified as either Born Global, Born Regional or Gradual Global between 2005 and 2018
Data: Custom data from Statistics Canada
Source: Office of the Chief Economist – Global Affairs Canada

5.5. Export survival and export performance at the start of exporting

In general, there is a correlation between the export performance in the first year of exporting and the length of exporting. For example, in figure 3, firms with longer lasting export tends to export more value, products and to more destinations in the first year of exporting. The difference in performance is most pronounced for export value, where firms that lasted 6 years in the export market began exporting with a median export value almost 7 times firms that lasted only 1 year. For the average number of export products and the average number of export destinations, the difference between firms that lasted 6 years and 1 year is only slightly more than double. In theory, initial export values is a proxy to the level of confidence and commitment between the trading partners and the expectation is that higher initial value of exports lead to higher likelihood of export survival (Nicita et al., 2013). In the same study, Nicita et al. (2013) found that the higher initial value of export lead to longer export duration.   

Figure 3: Export performance in the first year of exporting, by length of exporting

Note: All exporters where the firms could be identified as either Born Global, Born Regional or Gradual Global between 2005 and 2018
Data: Custom data from Statistics Canada
Source: Office of the Chief Economist – Global Affairs Canada

5.6. Export performance by international expansion strategy

In general, Born Global firms have better export performance than Born Regional firms, which in turn have better export performance than Gradual Global firms. From Figure 4 (left graph), one can see that at the start of exporting, Born Global exporters had highest median export value and average number of products exported, followed by Born Regional exporters, with Gradual Global exporters in last place. Born Global exporters also started with highest average number of export destinations, followed by Gradual Global exporters then Born Regional exporters. Furthermore, Born Global exporters retain higher median export value than both Born Regional and Gradual Global exporters throughout the entire length of exporting, as evidence by Figure 4 (right graph).

Figure 4: Export performance in the first year of exporting (left graph) and median value of exports through years of exporting (right graph), by international expansion strategy

Data: Custom data from Statistics Canada
Source: Office of the Chief Economist – Global Affairs Canada

5.7. International expansion strategy and multinational status

Multinational enterprises are corporations with majority-owned operations in more than one country. Foreign Multinationals are firms in Canada controlled by a foreign parent. Canadian Multinationals are Canadian-controlled firms with a foreign affiliate.

Born Global exporters are more likely to be Foreign Multinationals, as at the first year of exporting, about 9.6% of Born Global exporters are Foreign Multinational, compared to 6.1% for Born Regional and 6.5% for Gradual Global (see Figure 5). On the other hand, there doesn’t seem to be much difference in the share of Canadian Multinational between the three International Expansion Strategy.

Figure 5: Share of multinational in the first year of exporting, by international expansion strategy

Data: Custom data from Statistics Canada
Source: Office of the Chief Economist, Global Affairs Canada

Manufacturers and wholesalers account for the largest shares of goods exporters in Canada, with about one thirds of Canadian goods exporters in 2022 being manufacturers and about one fifth being wholesalers. In distant third place is retailers accounting for almost 1 in 10 of goods exporters. In the first year of exporting, Born Global exporters are more likely to be wholesalers than Born Regional and Gradual Global exporters (see Figure 6), but Born Regional exporters are more likely to be manufacturers.

Figure 6: Share of manufacturers and wholesalers in the first year of exporting, by international expansion strategy

Data: Custom data from Statistics Canada
Source: Office of the Chief Economist, Global Affairs Canada

While there is a higher share of Born Global exporters being Foreign Multinationals and also a larger share of Born Global exporters being wholesalers, Born Global exporters are not just indirect exporters being created by large Foreign Multinationals for the purpose of indirect exporting. Among wholesaler and retailer exporters, the share of Foreign Multinationals in the first year of exporting is not larger for Born Global exporters (see Figure 7). In fact, it is among manufacturer exporters, which produces and exports their own products, that the share of Foreign Multinationals in the first year of exporting is much larger for Born Global exporters (see Figure 7). The combination of these two facts reduces the concern that Born Global exporters are just indirect exporters being created by large Foreign Multinationals for the purpose of indirect exporting.

Figure 7: Share of foreign multinationals in the first year of exporting, by international expansion strategy and industry

Data: Custom data from Statistics Canada
Source: Office of the Chief Economist, Global Affairs Canada

6. Survival methodology

6.1. Survival function and hazard function

The survival function gives the probability that an object of interest will survive past a certain time. Even though it is named “survival function”, it is applicable to all kinds of time-to-event analyses. In this report, for example, it is used to analyze the time it takes for an exporting spell to end.

The hazard is the instantaneous risk that an event will end. In this case, the hazard is the risk that a firm will end their export spell for any given year. In survival regression modelling, the hazard is often the dependent variable, and the coefficient estimates can be thought of as hazard ratio. Once the hazard is found, it can be converted back to a survival function (see Annex 1.2 for an example of the relation between hazard function and survival function).

In this report, the two methods used to measure survival and hazard are the Kaplan-Meier method and the Cox-Proportional Hazard model.

6.2. Kaplan Meier survival analysis

The Kaplan-Meier estimator is a statistic used to estimate the survival function and is one of the estimators used in this report to estimate the export survival of Canadian exporting spells. It is a non-parametric statistic which means that we do not make any assumption about the underlying distribution, shapes or parameters of the survival function. An important advantage of the Kaplan-Meier estimator is that it takes into account censored data, in particular right-censoring, which occurs when the object does not die (or the event does not occur) by the end of the study period or if the object dropped out of the study before the end. (See annex 1.3 for more details).

In addition to estimating the survival curve, the Kaplan-Meier statistic is simple to interpret. A disadvantage is that it has no functional form, which means there is no simple mathematical function that describes the shape of the survival curve. Another disadvantage is that it does not allow us to calculate the hazard ratio between groups. Lastly, the Kaplan-Meier curve is an example of univariate analysis (similar to histogram), and it can only compare survival curves between a few categorical variables using the log-rank test.

6.3. Cox proportional hazard model

The Cox proportional-hazards model is a regression model commonly used for investigating the association between the survival time (or time to event) of objects and one or more predictor variables and is the main model used in this report to analyze export survival. The dependent variable in the Cox model is the hazard, denoted by h(t).

• t represents survival time
• h(t) is the hazard determined by a set of p covariates (x₁, x₂, … , x_p) and their respective estimated coefficients (b₁, b₂, … , b_p)
• h₀ is the baseline hazard, which is the value of the hazard if all the covariates are zero. The t in h₀(t) means that the baseline hazard might vary over time.

The Cox model is estimated in its logarithm form:

In the cox model, the coefficients (b₁, b₂, … , b_p) are estimated without specifying the baseline hazard (h₀(t)) using a method called partial likelihood. The exponential of the coefficients (for example: $e^{b_1}$ ) is the hazard ratio. A value of b_i greater than zero, or $e^{b_1}$  greater than 1, indicates that as the value of x_i increases, the hazard increases and thus the length of survival decreases. 

An important assumption of the Cox proportional-hazards model is that the hazard ratio is constant and proportional. For example, a constant hazard ratio of 2 between group A and group B means that throughout the study period, group A are always twice as likely to die (or have event occurred) than group B. If the proportional hazard assumption is violated, one can use a Cox model with time dependent coefficients (also known as extended Cox model), as described by Therneau et al. (2023). In this study, for example, exporting spells from Born Regional firms have lower hazard in the first year of exporting than exporting spells from Born Global firms but have higher hazard in later years.

Furthermore, the data in this paper has time-varying covariates (the independent variables changes through time), the Cox model used will also have time dependent covariates, as once again described by Therneau et al. (2023). For more information, see Annex 1.4.

Two-stage regression method

As previously mentioned in Section 4.3, the firm’s strategic choice between Born Global, Born Regional and Gradual Global is endogenous. To correct for this endogeneity in internationalization choice, Mudambi & Zahra (2007), Sui (2009), Sui and Baum (2014), and Sandberg et al. (2019) all used a two-stage regression method. In the first stage, the author used a (multi)logit or probit model:

• IS_i is the observed international strategy (Born Global, Born Regional and Gradual Global)
• z₁ … z_p are independent variables and α₁ … α_p are their respective coefficients; these are used to predict a firm’s internationalization strategy (${\widehat{\mathrm{IS}}}_i$)

The predicted internationalization strategy (${\widehat{\mathrm{IS}}}_i$) is then used in the second stage with a Cox model:

Furthermore, Sui and Baum (2014) and Sandberg et al. (2019) used a split-sample method to verify the appropriateness of the model and the robustness of results. The split-sample method involves splitting the data into 2 random sub-samples, then:

1. Using the first sub-sample, obtain estimated coefficients (${\hat{z}}_1,\dots ,{\hat{z}}_p$ ) in the multi-logit model,
2. Using the estimated coefficients (${\hat{z}}_1,\dots ,{\hat{z}}_p$ ) on the second sub-sample to get predicted internationalization strategy (${\widehat{\mathrm{IS}}}_i$)
3. Using the second sub-sample and predicted internationalization strategy ( ${\widehat{\mathrm{IS}}}_i$) to perform the second stage Cox model. 

In this study, the split-sample method is performed 1,000 times by randomly splitting the sample 1,000 times to obtain a distribution of the coefficients of interest (b_IS). The split-sample method has the advantage of producing an estimate bias toward zero (Sandberg et al, 2019; Angrist & Krueger, 1995), being reliable and powerful (Sandberg et al, 2019; Dulfour & Jasiak, 2001), and controlling effectively for Type I errors (Sandberg et al, 2019; Bolduc et al., 2008).

7. Export survival results

7.1. Kaplan Meier survival curves

In general, Canadian exporters^{Footnote 2} do not survive for a long time. Among all exporting spells from firms that can be identified as either Born Global, Born Regional, or Gradual Global, about 40% stopped after the first year, with a median survival time of 2 years (see Figure 8). By the 5^th year, only around 30% of exporting spells are still active. These results are similar to the results of Sabuhoro and Gervais (2004). Another noticeable trend is that the longer an exporter survives, the less likely that exporter will stop exporting as seen by the smaller drop off in survival between year 4 and year 5 compared to the large drop off between year 1 and year 2. This is similar to the existence of negative duration dependence found by Esteve‐Pérez et al. (2005), which means that the longer a firm continuously exports, the longer it will continue doing so.

Figure 8: Kaplan-Meier export survival curve for Canadian exporters

Note: Includes all Canadian exporting spells where the firms could be identified as either Born Global, Born Regional or Gradual Global between 2005 and 2018. Error bars represent the 95% confidence interval.
Data: Custom data from Statistics Canada
Source: Office of the Chief Economist, Global Affairs Canada

There are some differences in the export survival rate between the different international expansion strategies, with Born Global exporters surviving better than Born Regional exporters, who in turn survived better in the export market than Gradual Global exporters (Figure 9). For Born Global firms, 78% of exporting spells managed to survive past the first year, with 50% of exporting spells surviving past the 5^th year, resulting in a median survival time of 6 years. For Born Regional firms, exporting survival is a little bit lower, with 76% of exporting spells surviving past the 1^st year and 42% surviving past the 5^th year, resulting in a median survival time of 4 years. Gradual Global firms’ exporting survival is much lower, with only 57% of exporting spells surviving past the first year and 27% surviving past the 5^th year, resulting in a median export survival time of 2 years. However, it should be noted that the Kaplan-Meier analysis does not take into account other factors that could influence export survival.

Figure 9: Kaplan-Meier export survival curve, by international expansion strategy

Note: Error bars represent the 95% confidence interval.
Data: Custom data from Statistics Canada
Source: Office of the Chief Economist, Global Affairs Canada

7.2. Cox proportional hazard model – Regression results

The base model has only the International Expansion Strategy status as covariate, which can be represented equationally by:

Where h(t) is the hazard of stopping exports, and h₀ is the baseline hazard of stopping exports. IS_i is the International Expansion Strategy status of firm i, with Born Global status as the baseline. $e^{b_{\mathrm{IS}}}$ is the hazard ratio, where a value of greater than 1 means a higher hazard than Born Global firms, while a value of less than 1 means a lower hazard than Born Global firms.

In the base model (see table 2), Born Regional exporters have 22% higher hazard (or exporting spells from Born Regional firms are 22% more likely to stop exporting) than Born Global exporters. The 22% came from the exponentiated coefficient (see last column of Table 2), where 1.22 means that compares to Born Global exporters, Born Regional exporters have 1.22 times the hazard, or 22% higher hazard (see Section 6.3 or Annex 1.4 for more explanations).

The hazard is even higher for Gradual Global exporters as these are 96% more likely to stop exporting when compared against Born Global exporters. The results are unsurprising given what was seen in section 5.3, where Born Global firms’ exporting observations were less likely to be a one‑year exporter and also less likely to be the exit of multi-year exporting spells (see Table 1). Similarly in section 7.1, the Kaplan-Meier survival curve shows that Born Global exporters survive better than Born Regional exporters and Gradual Global exporters, hence the lower hazard rate of exiting exports.

Table 2: Cox proportional hazard base model results

However, a statistical test based on the scaled Schoenfeld residuals (Table 3) shows that the proportional hazard was violated for the International Expansion Strategy status variable (the dummy variables with Born Global status as the base category and Born Regional status and Gradual Global status as comparison categories) and therefore an extended Cox model with time dependent coefficients using the step function method (Therneau et al., 2023) was used to solve the violation of proportional hazard (see section 6.3 and annex 1.4 for more details). The conclusion that proportional hazard was violated can be reached since the p-value in Table 3 was below 0.05 and therefore the null hypothesis that proportional hazard exists was rejected. Therefore, for all models discussed from this point onward, the hazard ratio of the International Expansion Strategy status variable was divided into 2 periods, 1) the first year of an exporting spell and 2) the rest of the years.

Table 3: Scaled Schoenfeld residuals test for the Cox proportional hazard base model

The main model is an extended Cox model with time dependent coefficients using the step function method. It can be represented equationally by:

where h(t), h₀, and IS_i is the same as in the base model. b(t)_IS now has a time (t) component to it, which means the hazard ratio is no longer constant through time. Time (t) is divided into 2 periods: 1^st year of exporting spells and the rest of the years. This division of time works since for all the years after the 1^st year, the hazard ratio is constant as seen in table 5 below.

Additionally, the main model has a vector of Z control variables and their corresponding coefficients (a). The control variables include:

• firm size,
• the log of deflated export value,
• the log of number of products exported,
• the log of number of destinations exported to,
• previous exporting experience,
• the stratification of exporting cohorts, and
• industry of firms.

It is important to control for the industry of firms since some industries in Canada are more export-oriented, such as manufacturing or wholesale. The value of exports was included as a control variable since Sabuhoro and Gervais (2004) found that the monthly value of exports was an important determinant in reducing the hazard of an export exit. Furthermore, Nicita et al. (2013) found that the higher initial value of export lead to longer export duration. Sabuhoro and Gervais (2004) also found the number of products exported and the number of export destinations to be statistically significant. The number of export destinations were also controlled for by Sandberg et al. (2019) and was statistically significant. Sui (2009) also controlled for the number of products exported and the number of export destinations and found statistically significant coefficients for these variables. Previous exporting experience was added as a control variable since this was found to be statistically significant by Sabuhoro and Gervais (2004) and Sui (2009). Lastly, exporting year cohorts was added to control for the economic condition at the beginning of an exporting spell.

According to the main model (see Table 4), Born Regional exporters are 29% less likely to stop than Born Global exporters in the first year of exporting, but 18% more likely to stop in later years. For Gradual Global exporters, they are 15% more likely to stop exporting in the first year compared to Born Global exporters, and 29% less likely to stop in later years.

A statistical test based on the scaled Schoenfeld residuals (see Table 5) shows that the International Expansion Strategy status variable no longer violates the proportional hazard assumption, as the p-value are high enough to fail to reject the null hypothesis of proportional hazard.

Table 4: Cox proportional hazard main model results

Table 5: Scaled Schoenfeld residuals test for Cox proportional hazard main model

To summarise the main model, once the extended Cox model with time dependent coefficients was implemented and other control variables were introduced, Born Regional exporters are generally less likely to stop in the first year than Born Global exporters, but the reverse is true for later years. Robustness checks using the two-stage method, the split sample method and other (see Annex 2) provide further assurance that Born Regional exporters having lower hazard in the first year of exporting is consistent, but having a higher hazard in later years of exporting is less robust, losing statistical significance under a few alternative specifications of the main model (see Annex 2).

For Gradual Global exporters, once the extended Cox model with time dependent coefficients was implemented and other control variables were introduced, their export spells generally are more likely to stop in the first year, but less likely to stop in the later years when compared against Born Global exporters. The lower hazard in the later years is consistent through various robustness checks, while the higher hazard in the first year is less robust, as the coefficient lost statistical significance under a few alternative specifications of the model (see Annex 2).

The hazard ratios calculated from the Cox models can be turn into predicted survival curves (see Figure 10). After turning the results from the main model into survival curves, it can be seen that Born Regional exporters had higher survival rates than both Born Global exporters and Gradual Global exporters in the first few years of exporting (see Figure 10). The predicted survival curves were predicted with the following value for other covariates:

• Firm size: small
• Previous exporting experience: False
• Exporting cohort: 2005 to 2007 cohort
• Industry: Manufacturing
• Export Value: log of median export value
• Number of products exported: log of median number of products exported
• Number of export destinations: log of median number of export destinations.

However, by the fourth year of exporting and beyond, there are no statistical differences between the survival curves.

Figure 10: Predicted survival curves using the main model, by international expansion strategy

Note: Error bars represent the 95% confidence interval.
Data: Custom data from Statistics Canada
Source: Office of the Chief Economist, Global Affairs Canada

A reason for the better survival of Born Regional exporters is because of the existence of one-year exporters, i.e. those exporters who exported for only one year in length. As seen in section 5.3, one-year exporters accounted for approximately 15% of total export observations and are more prevalent in Gradual Global exporters. As a result, they cause the hazard rate in the first year of exporting to be different from the rest of the years (see Table 4). Addition to these facts, the difference in short-term survival is of less interest than long-term survival since it was shown in section 5.4 that consistent participation in the exports market led to better export performance. If we were to remove all exporting spells that lasted only one year from the main model (see Table 6), the hazard (likelihood) of export exit became the largest for Born Global exporters. Born Regional exporters and Gradual Global exporters have similar hazard of export exit. Turning these results into predicted survival curves show that Born Regional exporters and Gradual Global exporters survive at a similar rate (Figure 11). Born Global exporters survive slightly worse in the first few years of exporting, but the results are within the 95% confidence interval and by year 4 and beyond, there are no differences in the survival rate.

Table 6: Cox proportional hazard main model results (excluding exporters that lasted one year)

Figure 11: Predicted survival curves using the main model but excluding exporters that lasted only one year, by international expansion strategy

Note: Error bars represent the 95% confidence interval.
Data: Custom data from Statistics Canada
Source: Office of the Chief Economist, Global Affairs Canada

8. Discussion of results

This study found that Canadian exporters do not last long in the export market, with about 40% stopping after the first year and only about 30% surviving past the fifth year. As mentioned in the literature review (Section 4), most studies in other countries also find that most exporting relationships do not last long.

There are some differences with regards to export survivability between different International Expansion Strategies but with the control of other covariates and the consideration of only long-term export survival, the differences in survival became marginal. Without controlling for other covariates, Born Global exporters survive the longest, followed by Born Regional exporters and in last place are Gradual Global exporters. However, Born Global exporters are generally better at exporting in that they export more in terms of value, number of products, and number of destinations. These export performance variables, in particular the value of initial export, can be considered, in theory, as a proxy to the level of confidence and commitment of the trading relationship (Nicita et al., 2013). Once these export performance variables and other variables are controlled for, in addition to the subtraction of exporters that only lasted one year as we are more interested in long-term survival, the differences in survival rates between the three different International Expansion Strategies became marginal. In another word, there are no major differences in the long-term export survival rates between Born Global, Born Regional and Gradual Global exporters once we account for the differences in export performance such as the value of export, a common proxy to the level of confidence and commitment to exports.

There are very few studies on the export survivability of Canadian exporters. This study differs from Sui and Baum (2014) in that it studies a more recent time period, 2005 to 2018 instead of 1997 to 2004. Sui and Baum (2014) also only included manufacturers and small and medium enterprises, this study included all Canadian exporters with firms’ birth date later than 2004 that could be identified as either Born Global, Born Regional, or Gradual Global. Lastly, this study used an extended Cox model with time dependent coefficients, signifying that hazard ratios are not constant, while Sui and Baum (2014) had constant hazard ratios and used a regular Cox proportional hazard model. 

This study faces a few limitations, which offers opportunities for future research. While the study controlled for firm characteristics, the competencies of a firm’s main decision maker can also have an impact on firm’s performance, and therefore, future research can look to control for management characteristics.

Secondly, export is also just one path to engage internationally and future research can look at internationalization through a more holistic view, including such as through foreign investment and imports. Blum et al. (2020) found that in Chile, new exporters that are also new firms^{Footnote 3} do not fit the typical export pattern, they start as relatively large exporters and are significantly less likely to quit exporting in the first year. These new exporters that are also new firms are indeed new economic entities in that they operate plants that did not previously belong to any other firms, but legal documents show that they are often owned by existing firm despite being new economic entities. In most cases, these new exporters that are also new firms were constituted under arrangements that are essentially foreign and/or domestic direct investments, which indicates that a more holistic view of internationalization which includes foreign investment might be necessary.

Lastly, export exits were observed in the data, but it is not possible to determine in this dataset if the exit was due to export market failure or a strategic decision. The consequences of export failure on the firm is also another avenue for future research, and the consequences might differ if export exits were due to strategic decisions rather than export market failures.

9. Conclusion

The objective of policy makers is often to have Canadian firms participate and grow in the export markets; however, an often-overlooked area is the export longevity of Canadian exporters. This study finds that Canadian exporters do not generally export for long durations continuously, with about 40% of export activity periods stopping after the first year, a similar result to other studies.

In contrast to the traditional theory, where it is recommended that a firm first establishes itself domestically before incrementally increasing its commitment to the international market, this report finds that the age of the exporter at the beginning of the exporting journey does not play an important role in export longevity. The difference in the long-term export longevity between firms that exported early in their operation and firms that exported at a later age becomes marginal once export confidence and commitment, represented by export performance variables, are taken into account.

While this research sheds some light on the often-overlooked subject of Canadian export longevity, there were some limitations to the research which provide opportunities for future projects. On area would be to take into account firms’ management expertise, as experience may determine export success. Alternatively, export periods end for different reasons—due to either export market failure or strategic decisions—which could have different consequences for the firm. Furthermore, one could also look at internationalization through a more holistic lens by incorporating imports and investment into the definition.

10. Bibliography

Angrist, J.D. & Krueger, A.B. (1995). Split-sample instrumental variables estimates of the returns to schooling. Journal of Business and Economic Statistics, 13(2), 225-235.

Autio, E., Sapienza, H., & Almeida, J. 2000. Effects of age at entry, knowledge intensity, and imitability on growth. Academy of Management Journal, 43(5): 909–924.

Bernini, M., Du, J., & Love, J.H. (2016). Explaining intermittent exporting: Exit and conditional re-entry in export markets. Journal of International Business Studies, 47(9), 1058-1076

Besedeš, T. and Prusa, T.J. (2006). ‘Ins, outs, and the duration of trade’, Canadian Journal of Economics/Revue canadienne d’économique, vol. 39(1), pp. 266–295.

Besedes, T. and Prusa T., (2011) “The Role of Extensive and Intensive Margins and Export Growth”. Journal of Development Economics, 96:2, 371-379

Bader, Anton Vasilievich and Tim Mazzarol. “'Defining the Born Global Firm' A Review of the Literature.” (2009).

Bandick, Roger (2019) : Global sourcing, firm size and export survival, Economics Discussion Papers, No. 2019-62, Kiel Institute for the World Economy (IfW), Kiel

Blum, B.S. & Claro, S. & Horstmann, I. & Tombe, T. (2020). The DNA of New Exporters: Spin-offs and FDI at the Extensive Margin of Trade. AER: Insights 2020, 2(3): 397-408.

Bolduc, D., Khalaf, L., & Moyneur, E. (2008). Identification-robust simulation-based inference in joint discrete/continuous models for energy markets. Computational Statistics and Data Analysis, 52(6), 3148-3161.

Chen, Shenjie and Yu, Emily (2010). Export Dynamics in Canada: Market Diversification in a Changing International Economic Environment. Trade Policy Research 2010: Exporter Dynamics and Productivity.

Chen, J., Sousa, C.M., & He, X. (2016). The determinants of export performance: A review of the literature 2006-2014. International Marketing Review, 33(5), 626-670.

Choquette, E., (2019), “Import-based market experience and firms´ exit from export markets”. Journal of International Business Studies, 50; 423-449.

Dulfour, J.M., & Jasiak, J. (2001). Finite sample limited information inference methods for structural equations and models with generated regressors. International Economic Review, 42(3), 815-843.

Eaton, J., Eslava, M., Kugler, M. and Tybout, J., (2007) “Export Dynamics in Colombia: FirmLevel Evidence”. NBER Working Paper 13531.

Esteve‐Pérez, S., Máñez-Castillejo, J.A., Rochina, M.E., & Sanchis‐Llopis, J.A. (2005). A survival analysis of manufacturing firms in export markets.

Esteve-Pérez, S. and Requena-Silvente, F. and Pallardó-Lopez, V.J. (2013). ‘The duration of firm-destination export relationships: Evidence from Spain, 1997–2006’, Economic Inquiry, vol. 51(1), pp. 159–180.

Forsgren, M. and Johanson, J. (1975), Internationall företagsekonomi. Stockholm: Norstedts.

Gima, S., Greenaway, D., & Kneller, R. (2003). Export market exit and performance dynamics: A causality analysis of matched firms. Economic Letters, 80, 181-187.

Ilmakunnas, P. and Nurmi, S., (2010), “Dynamics of export and market entry and exit”. The Scandinavian Journal of Economics 112: 101–126

Johanson, J., and Wiedersheim-Paul, F. (1975), “The Internationalization Process of the Firm-Four Swedish Case Studies”, Journal of Management Studies, Vol. 12, pp. 305-322.

Kaplan, E. L., Meier, P. (1958). “Nonparametric estimation from incomplete observations”. Journal of the American Statistical Association 53 (282): 457-481

Knight, G. A. & Cavusgil, S. T. 1996, 'The Born Global Firm: A Challenge to Traditional Internationalization Theory', Advances in International Marketing, vol. 8, pp. 11 - 27.

Knight, G. A., & Cavusgil, S. T. 2005. A taxonomy of born-global firms. Management International Review, 45(3): 15–35.

Lodefalk, M., Tang, A., & Yu, M. (2022). ‘Stayin’ Alive: Export Credit Guarantees and Export Survival.’ Working Paper 10/2022 (Economics). ISSN 1403-0586. Örebro University School of Business, SE-701 82 Örebro, Sweden

Mudambi, R., & Zahra, S. A. 2007. The survival of international new ventures. Journal of International Business Studies, 38(2): 333–352.

Nicita, A., Shirotori, M., and Klok, B.T. (2013) Survival analysis of the exports of least developed countries: the role of comparative advantage. UNCTAD: Policy Issues in International Trade and Commodities Study, Series No. 54.

Oviatt, B. M. & McDougall, P. P. 1994, 'Toward a Theory of International New Ventures', Journal of International Business Studies, vol. 25, no. 1, pp. 45-64.

Rennie, M. W. 1993, 'Born Global', McKinsey Quarterly, no. 4, p. 45.

Rugman, A. M., & Verbeke, A. 2004. A perspective on regional and global strategies of multinational enterprises. Journal of International Business Studies, 35(1): 3–18.

Sapienza, H. J., Autio, E., George, G., & Zahra, S. A. 2006. A capabilities perspective on the effects of early internationalization on firm survival and growth. Academy of Management Review, 31(4): 914–933.

Sabuhoro, J.B. & Gervais, Y. (2004). “Factors Determining the Success or Failure of Canadian Establishment on Foreign Markets: A Survival Analysis Approach.” Analytical Studies Branch research paper series – Statistics Canada. Catalogue no. 11F0019MIE – No. 220.

Sandberg, S., Sui, S. & Baum, M. (2019) What resources hinder SME export failure? Market experience and moderating effects of firm-specific resources on emerging market exit, Journal of Business Research: 98: 489-502.

Sui, S. (2009). “Born Global, Gradual Global, and their Determinants of Exit from Exporting”. Job Market paper. Department of Economics – Carleton University.

Sui, S. & Baum, M. (2014). “Internationalization strategy, firm resources and the survival of SMEs in the export market.” Journal of International Business Studies; Basingstoke Vol. 45, Iss. 7

Therneau, T. & Crowson, C. & Atkinson, E. (2023). “Using Time Dependent Covariates and Time Dependent Coefficients in the Cox model”. R vignette retrieved March 11, 2023.

Volpe, C. and Carballo, J., (2009) “Survival of new Exporters: Does it Matter How They Diversify?”, IDB Working Paper Series IDB-WP-140.

Wagner, Joachim (2005): Exports and Productivity: A survey of the evidence from firm level data, HWWA Discussion Paper, No. 319, Hamburg Institute of International Economics (HWWA), Hamburg

11. Annex 1 – Methodology

11.1. Annex 1.1: Variable descriptions

Table A1: List of variables

11.2. Annex 1.2: Relationship between the survival function and the hazard function

Assuming a simple example where 10% of the original population die every time period (t), this can be graphically represented by the following Probability Density Function [f(t)] and Cumulative Density Function [F(t)]. 

Figure A1: Probability density, f(t), and cumulative density, F(t), function

By the 10th period, 100% of the population have died, leading to 0% survival after the 10th period. The Survival Function [S(t)] is then the inverse of the F(t). In another word, $S\left(t\right)=1-F\left(t\right)$ .

Figure A2: Survival, S(t), and cumulative density, F(t), function

The hazard rate can be calculated using the number of death in a given time period divided by the number at risk in the same time period. In another word, the hazard function [h(t)] is the probability density function divided by the survival function.

Figure A3: Probability density, f(t), survival, S(t), and hazard, h(t), function

The cumulative hazard function [H(t)] is the area underneath the curve of the hazard function [h(t)], or in another word, its integral. It measures how the risk of death has accumulated over time.

Figure A4: Hazard, h(t), and cumulative hazard, H(t), function

Mathematically, these relationships can be represented by:

Or in another word, the cumulative hazard function is just the negative log of the survival function.

11.3. Annex 1.3: Kaplan Meier survival analysis

The Kaplan-Meier curve is a graphical representation of the survival function; it shows the probability that a subject will survive past a time (t). This non-parametric method was invented by Kaplan & Meier (1958). The advantage of the Kaplan-Meier method is that it can account for data that are censored, truncated or have missing values. 

The estimator of the survival function S(t) (the probability of surviving longer than t) is given by the following formula:

t_i is a time when at least one death (or event) happened; d_i is the number of deaths at time t_i; and n_i is the number of objects known to have survived up to time t_i.

∏ is used in math to indicate repeated multiplication, hence $\prod _{i:t_i\le t}\left(1-\frac{d_i}{n_i}\right)$ means to multiply $1-\frac{d_i}{n_i}$ in period i with all $1-\frac{d}{n}$ calculated in all previous periods, which would be:

S(t) is often calculated in its log form, and the variance of ln⁡ (S(t)) can be calculated as follow:

Therefore, the 95% confidence interval can be calculated as follow:

To compare Kaplan-Meier curves between different groups, one can use the log-rank test. The log-rank test resembles a chi-square test of independence, and is constructed by computing the observed and expected number of deaths (or events) in each groups at each observed death (event) time.

Assuming that there are 2 groups, A and B. At each event time (t), we can get the observed number of event for group A, ${O_t}^A$ , and B, ${O_t}^B$ , from the data. The expected number of event, ${E_t}^A$ and ${E_t}^B$ , can be calculated as following:

$E^t=\frac{n^t{}^A{d}_t}{n_t}$ and $E^t{}^B=\frac{n^t{}^B{d}_t}{n_t}$ where $d_t={d_t}^A+{d_t}^B$ and $n_t={n_t}^A+{n_t}^B$ and $E^t{}^A+E^t{}^B={d_t}^A+{d_t}^B$

is the number of events, n is the number of objects at risk

The log-rank test statistics is then:

The log-rank test statistic in this case has a chi-square distribution with one degree of freedom.

11.4. Annex 1.4: Cox proportional hazard model

The Cox proportional-hazards model is a semi-parametric model that investigate the association between the survival time (or time to event) of objects and one or more predictor variables. It is semi-parametric because the parameters are only partially defined. The regression parameters (b₁, b₂, … , b_p) are known, but the distribution of the outcome remains unknown since the form or shape of the baseline hazard (h₀ (t)) is not specified.

The Cox model is estimated in its logarithm form:

t represents survival time. The dependent variable, h(t), is the hazard and is a combination of survival time and whether or not death (event) occurred. h(t) is estimated by a set of p covariates (x₁, x₂, … , x_p) and their respective estimated coefficients (b₁, b₂, … , b_p).

h₀ (t) is the baseline hazard, the value of the hazard if all the covariates are equal to zero. The baseline hazard is not defined in the Cox model and is allowed to vary over time.

In the cox model, the coefficients (b₁, b₂, … , b_p) are estimated without specifying the baseline hazard (h₀ (t)) using a method called partial likelihood. The exponential of the coefficients (for example: $e^{b_1}$ ) is the hazard ratio. A value of b_i greater than zero, or $e^{b_1}$ greater than 1, indicates that as the value of x_i increases, the hazard increases and thus the length of survival decreases.

Below is an example of the data set up to run a Cox proportional hazard model. The Object column is a unique identifier. Death is a binary variable, a value of 1 means the object died, a value of 0 means the object was censored. Time is the time that it took for the object to die or became censored. For example, Object 1 in the table below died at time period 3, while Object 2 was lost to followed up (censored) at time period 2. x₁ and x₂ are covariates and are not time-variant in this case.

Table A2: Data example for Cox proportional hazard model

If the covariates vary through time, one will have to use a time dependent covariates Cox model as demonstrated by Therneau et al. (2023). Assume that x₁ varies through time such that:

The main difference in the table below is that the Time column is split into two columns, Time_start and Time_stop. Object 1 now has 3 lines of data, with Death being 0 in the first 2 lines and equal 1 in the last line which corresponds to dying in the third time period. Object 2 now has 2 lines of data, with Death equal 0 in both periods signifying it was censored in the second period. x₂ is not time varying and remains constant for the same object. x₁ varies through time and is not constant for the same object.

Table A3: Data example for time dependent covariates Cox model

An important assumption of the Cox proportional-hazards model is that the hazard ratio is constant and proportional. For example, a constant hazard ratio of 2 between group A and group B means that throughout the study period, group A are always twice as likely to die (or have event occurred) than group B. If the proportional hazard assumption is violated, like in this paper, one can use a Cox model with time dependent coefficients, as described by Therneau et al (2023).

In the equation above, b₁ does not have constant and proportional hazard and therefore varies through time. The proportional hazards assumption can be checked using statistical tests and graphical diagnostics based on the scaled Schoenfeld residuals.

One of the simplest Cox model with time dependent coefficients is a step function for b₁(t), i.e., different coefficients over different time intervals (Therneau et al. 2023). For example, one would split t into 3 different time groups, time_group1, time_group2 and time_group3, and then would estimate a different b₁ for each time_group.

12. Annex 2: Alternative model specifications

Several alterations were made to the main model to verify the robustness of results:

The variable province, indicating which province the firm was located in, was added to the main model. However, the addition of this province variable did not change the estimated coefficients much (see the changes between Table 7 and Table 8 below), and the standard errors remained similar. As a result, we can conclude that the province variable was not a confounder to the International Expansion Strategy status variable. Therefore, it was left out of the model since it has missing values.

Table 7: Results from the province model (main model plus the addition of province as covariates)

Table 8: Main model but with the same dataset as the province model

Instead of using a 2-year threshold to classify Born Global and Born Regional firms, a 3-years threshold was used, that is a firm is Born Global or Born Regional if export within 0 – 3 years of inceptions instead of within 0 – 2 years of inceptions. Compared to the main model (see Table 4), the coefficients in this model remain similar (see Table 9 below), except for Gradual Global exporters in the first year of exporting, where they are now even more likely to stop exporting, 28% more likely to stop than Born Global exporters instead of just 15% more that was reported in the main model.

Table 9: Main model but with a 3-year threshold for born global and born regional status

Another threshold that could be modified for robustness check is the 25% export to revenue ratio, which was lowered to 10% and 5%. A firm is now Born Global or Born Regional if their exports are more than or equal to 10%/5% of revenue within 0 – 2 years of inception. The coefficients in these models (Table 10 and Table 11 below) remain similar to the coefficients in the main model (Table 4), providing additional robustness to the results in the main model.

Table 10: Main model but with a ratio of 10% export to revenue to define born global and born regional status

Table 11: Main model but with a ratio of 5% export to revenue to define born global and born regional status

An interaction variable between size and International Expansion Strategy status was added but some combinations of size and International Expansion Strategy status did not have enough exporting spells that stopped exporting, resulting in unreliable estimated coefficients. The interaction between size and International Expansion Strategy status was therefore left out of the modelling process.

The log of the deflated export value was also interacted with International Expansion Strategy status but the coefficients for International Expansion Strategy status and log of deflated export remain similar to the main model. The interaction terms also were not statistically significant and did not improve the prediction of the model, hence it was left out of the modeling process.

The main model was also applied for just exporting spells that began between 2005 and 2009. The coefficients for Born Regional exporters remained similar (see Table 12 below and Table 4) but lost statistical significance in the second time period. The coefficients for Gradual Global exporters also remained similar but lost statistical significance in the first year. These losses of statistical significance are likely due to having smaller sample size.  

Table 12: Main model but using only exporting spells that began between 2005 and 2009

The main model was also applied for the rest of the data, exporting spells that began between 2010 and 2018. The coefficients for Born Regional and Gradual Global remain similar, but there was a loss of statistical significance in the second time period for Born Regional exporters, likely due to smaller sample size.

Table 13: Main model but using only exporting spells that began between 2010 and 2018

One factor that often impacts a firm’s export performance is its productivity. Wagner’s (2005) survey of the literature shows that “exporters are more productive than non-exporters, and that the more productive firms self-select into export markets” and therefore, it would be reasonable to include a measure of productivity into the survival model. Here, labour productivity, measured by deflated sales of goods and services divided by the number of workers, was introduced into the survival model as a control variable. Similarly, multinational status (foreign vs. Canadian) of an enterprise was also included as a control variable in the model since in Canada, more than 80% of the value of merchandise exports came from multinational enterprises. Another variable that was added is importer status, as exporters that are also importers accounted for over 85% of total merchandise trade in Canada in 2021, as these two-way traders are strongly integrated in global supply chains and are likely to have higher survivability in the export market. Unfortunately, firms have missing data for labour productivity, multinational status and importer status, and the inclusion of these variables resulted in a smaller data frame of approximately 45,000 rows of observations, down from approximately 86,000 observations when these variables were excluded. In other words, the inclusion of these variables resulted in a database of approximately 20,000 exporting spells, down from approximately 35,000 exporting spells without them. Nonetheless, it is important to find out whether or not labour productivity, multinational status and importer status are confounders to the variable of interest in this research, International Expansion Strategy status.

With the inclusion of labour productivity, multinational status and importer status to the main model and using the data frame with approximately 45,000 rows of observations (see Table 14), Born Regional exporters now have a hazard that is 30% smaller than Born Global exporters in the first year of exporting, but in later years the hazard is 21% higher but is not statistically significant. In the first year of exporting, Gradual Global exporters are 27% more likely than Born Global exporters to stop exporting, but in later years, they are 32% less likely to do so. When these results were compared against the main model using the same reduced dataset (see Table 15), the coefficients and standard errors for Born Regional and Gradual Global exporters did not change much, indicating that labour productivity, multinational status and importer status are not confounders to the International Expansion Strategy status variable. Therefore, labour productivity, multinational status and importer status can be left out when modelling for the effect of International Expansion Strategy on the hazard rate since there are missing values for these variables.

Table 14: Main model with the addition of labour productivity, multinational status, and importer status as covariates

Table 15: Main model but using the same reduced dataset as the model with the addition of labour productivity, multinational status, and importer status as covariates

As previously mentioned in section 4.3 and section 6.3, a two-stage regression model was employed to account for the fact that the firm’s strategic-choice between Born Global, Born Regional and Gradual Global is endogenous. The two-stage regression method can be represented equationally by:

First stage: ${\mathrm{IS}}_i=\gamma 0_{}+\gamma Y$

Second stage: $\ln (h(t\left)\right)={b\left(t\right)}_{\mathrm{IS}}{\mathrm{IS}}^{^}{i}_{}+\alpha Z$

Where in the first stage, IS_i is the International Expansion Strategy status of firm i similar to the base model. Y is a vector of predictors and their coefficient (γ). These are used to predict the International Expansion Strategy $\left({\widehat{\mathrm{IS}}}_i\right)$ using a multinomial logistic regression. The predictors used include: size, exporting year cohorts, firms’ industries, log of number of products exported, log of number of export destinations, log of deflated export value, indicator if firm has activities in more than one industry, indicator if firm has operations in more than one province or territory.

The second stage is the same as the main model, but instead of using IS_i, the International Expansion Strategy status, the second stage uses ${\widehat{\mathrm{IS}}}_i$ , the predicted International Expansion Strategy from the first stage.

The two-stage results (see Table 16) show that in the first year of exporting, the hazard for Born Regional exporters is 50% smaller than Born Global exporters. But in the later years, the hazard is 56% higher than Born Global exporters. For Gradual Global exporters, they have a hazard that is 20% higher than Born Global exporters in the first year of exporting (albeit with less significance level since p-value is slightly larger than 10%), but the hazard is 48% lower in later years.

The split sample method, described in section 6.3, is essentially the same as the two-stage regression method but the data was randomly split into two halves. The first half was used to get estimates for γ and γ₀, which was then used in the second half to get estimated ${\widehat{\mathrm{IS}}}_i$. These estimated ${\widehat{\mathrm{IS}}}_i$ and the second half of data was then used to retrieve b(t)_IS. The split sample method was repeated 1000 times to obtain a 95% confidence interval for the estimated coefficients, b(t)_IS.

The results (see Table 17) from the split sample method described above provide further robustness. The 95% interval from the split sample method confirms that compared to Born Global exporters, Born Regional exporters are less likely to stop in the first year of exporting, since the 95% confidence interval of the estimated coefficients contains only negative values, but more likely to stop in the later years of exporting while Gradual Global exporters are less likely to stop in the later years. For Gradual Global exporters in the first year, the result has less confidence as the 95% interval has both negative and positive results.

Table 16: Two-stage Cox proportional hazard model results

Table 17: Cox proportional hazard split sample results

Predicted survival curves were also created using the two‑stage model (see Figure 12) for further robustness check, and the results are similar to the predicted survival curves created using the main model (see Figure 10). Born Regional exporters had a higher survival rate in the first few years of exporting but by the fourth year, there are no statistical significance differences in the survival rates.

Figure 12: Predicted survival curves using the two-stage model, by international expansion strategy

Note: Error bars represent the 95% confidence interval.
Data: Custom data from Statistics Canada
Source: Office of the Chief Economist, Global Affairs Canada